JB Ies 109 Exercises Answers

Introductory Statistics Explained (1.
09)
Answers to Exercises
©2015 Jeremy Balka

All rights reserved
Balka ISE 1.09 2
Contents
1 Introduction 1
2 Gathering Data 3
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2 Populations and Samples, Parameters and Statistics . . . . . . . . . . . . 3
2.3 Types of Sampling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.3.1 Simple Random Sampling . . . . . . . . . . . . . . . . . . . . . . . 3
2.3.2 Other Types of Random Sampling . . . . . . . . . . . . . . . . . . 4
2.4 Experiments and Observational Studies . . . . . . . . . . . . . . . . . . . 4
2.5 Chapter Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
3 Descriptive Statistics 11
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
3.2 Plots for Qualitative and Quantitative Variables . . . . . . . . . . . . . . . 11
3.2.1 Plots for Qualitative Variables . . . . . . . . . . . . . . . . . . . . 11
3.2.2 Plots for Quantitative Variables . . . . . . . . . . . . . . . . . . . . 12
3.3 Numerical Measures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3.3.1 Summation Notation . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3.3.2 Measures of Central Tendency . . . . . . . . . . . . . . . . . . . . . 12
3.3.3 Measures of Variability . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.3.4 Measures of Relative Standing . . . . . . . . . . . . . . . . . . . . 15
3.4 Boxplots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.5 Linear Transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.6.1 Basic Calculations . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.6.2 Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.6.3 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
4 Probability 25
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
4.2 Basics of Probability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
4.2.1 Sample Spaces and Sample Points . . . . . . . . . . . . . . . . . . 25
i
Balka ISE 1.09 CONTENTS ii
4.3 Rules of Probability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

4.4 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
4.5 Bayes’ Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.6 Counting rules: Permutations and Combinations . . . . . . . . . . . . . . 33
4.7 Probability and the Long Run . . . . . . . . . . . . . . . . . . . . . . . . . 33
4.8.2 Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4.8.3 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
5 Discrete Random Variables and Discrete Probability Distributions 51

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
5.2 Discrete and Continuous Random Variables . . . . . . . . . . . . . . . . . 51
5.3 Discrete Probability Distributions . . . . . . . . . . . . . . . . . . . . . . . 51
5.3.1 The Expectation and Variance of Discrete Random Variables . . . 52
5.4 The Bernoulli Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
5.5 The Binomial Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
5.5.1 Binomial or Not? . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
5.5.2 A Binomial Example with Probability Calculations . . . . . . . . . 56
5.6 The Hypergeometric Distribution . . . . . . . . . . . . . . . . . . . . . . . 57
5.7 The Poisson Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
5.7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
5.7.2 The Relationship Between the Poisson and Binomial Distributions 59
5.7.3 Poisson or Not? More Discussion on When a Random Variable has
a Poisson distribution . . . . . . . . . . . . . . . . . . . . . . . . . 60
5.8 The Geometric Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . 60
5.9 The Negative Binomial Distribution . . . . . . . . . . . . . . . . . . . . . 61
5.10 The Multinomial Distribution . . . . . . . . . . . . . . . . . . . . . . . . . 62
5.11.2 Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
5.11.3 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
5.11.4 Extra Practice Questions . . . . . . . . . . . . . . . . . . . . . . . 70
6 Continuous Random Variables and Continuous Probability Distribu-

tions 77
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
6.2 Properties of Continuous Probability Distributions . . . . . . . . . . . . . 77
6.2.1 An Example Involving Integration . . . . . . . . . . . . . . . . . . 78
6.3 The Continuous Uniform Distribution . . . . . . . . . . . . . . . . . . . . 78
6.4 The Normal Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
6.4.1 Finding Areas Under the Standard Normal Curve . . . . . . . . . . 80
6.4.2 Standardizing Normally Distributed Random Variables . . . . . . . 81
Balka ISE 1.09 CONTENTS iii
6.5 Normal Quantile-Quantile Plots . . . . . . . . . . . . . . . . . . . . . . . . 82

6.5.1 Examples of Normal QQ Plots . . . . . . . . . . . . . . . . . . . . 82
6.6 Other Important Continuous Probability Distributions . . . . . . . . . . . 83
6.6.1 The χ2 Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . 83
6.6.2 The t Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
6.6.3 The F Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
6.7.2 Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
6.7.3 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
7 Sampling Distributions 91
7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
7.2 The Sampling Distribution of the Sample Mean . . . . . . . . . . . . . . . 93
7.3 The Central Limit Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . 94
7.4 Some Terminology Regarding Sampling Distributions . . . . . . . . . . . . 95
7.4.1 Standard Errors . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
7.4.2 Unbiased Estimators . . . . . . . . . . . . . . . . . . . . . . . . . . 95
7.6 Basic Calculations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
7.7 Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
7.8 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
8 Confdence Intervals 105

8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
8.2 Interval Estimation of µ when σ is Known . . . . . . . . . . . . . . . . . . 105
8.2.1 Interpretation of the Interval . . . . . . . . . . . . . . . . . . . . . 107
8.2.2 What Factors A˙ect the Margin of Error? . . . . . . . . . . . . . . 107
8.2.3 An Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
8.3 Confdence Intervals for µ When σ is Unknown . . . . . . . . . . . . . . . 107
8.3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
8.3.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
8.3.3 Assumptions of the One-Sample t Procedures . . . . . . . . . . . . 109
8.4 Determining the Minimum Sample Size n . . . . . . . . . . . . . . . . . . 110
8.5.2 Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
8.5.3 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
9 Hypothesis Testing 123

Balka ISE 1.09 CONTENTS iv
9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123

9.2 The Logic of Hypothesis Testing . . . . . . . . . . . . . . . . . . . . . . . 124
9.3 Hypothesis Tests for µ when σ is Known . . . . . . . . . . . . . . . . . . . 124
9.3.1 Constructing Appropriate Hypotheses . . . . . . . . . . . . . . . . 124
9.3.2 The Test Statistic . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
9.3.3 The Rejection Region Approach to Hypothesis Testing . . . . . . . 124
9.3.4 P-values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
9.4 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
9.5 Interpreting the p-value . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
9.5.1 The Distribution of the p-value When H0 is True . . . . . . . . . . 128
9.5.2 The Distribution of the p-value When H0 is False . . . . . . . . . . 128
9.6 Type I Errors, Type II Errors, and the Power of a Test . . . . . . . . . . . 129
9.6.1 Calculating Power and the Probability of a Type II Error . . . . . 129
9.6.2 What Factors A˙ect the Power of the Test? . . . . . . . . . . . . . 132
9.7 One-Sided Test or Two-Sided Test? . . . . . . . . . . . . . . . . . . . . . . 132
9.8 Statistical Signifcance and Practical Signifcance . . . . . . . . . . . . . . 133
9.9 The Relationship Between Hypothesis Tests and Confdence Intervals . . . 133
9.10 Hypothesis Tests for µ When σ is Unknown . . . . . . . . . . . . . . . . . 133
9.10.1 Examples of Hypothesis Tests Using the t Statistic . . . . . . . . . 134
9.11 More on Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
9.12 Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
9.12.2 Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
9.12.3 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
10 Inference for Two Means 153

10.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
10.2 The Sampling Distribution of the Di˙erence in Sample Means . . . . . . . 153
10.3 Hypothesis Tests and Confdence Intervals for Two Independent Samples
(When σ1 and σ2 are known) . . . . . . . . . . . . . . . . . . . . . . . . . 154
10.4 Hypothesis Tests and Confdence Intervals for µ1 − µ2 (When σ1 and σ2
are unknown) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
10.4.1 Pooled-Variance t Tests and Confdence Intervals . . . . . . . . . . 154
10.4.2 Welch (Unpooled Variance) t Tests and Confdence Intervals . . . . 155
10.4.3 Guidelines for Choosing the Appropriate Two-Sample t Procedure 156
10.5 Paired-Di˙erence Procedures . . . . . . . . . . . . . . . . . . . . . . . . . 156
10.5.1 Paired-Di˙erence t Tests and Confdence Intervals . . . . . . . . . 156
10.6 Investigating the Normality Assumption . . . . . . . . . . . . . . . . . . . 159
10.7.2 Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
10.7.3 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
Balka ISE 1.09 CONTENTS v
11 Inference for Proportions 175

11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
11.2 The Sampling Distribution of p̂ . . . . . . . . . . . . . . . . . . . . . . . . 175
11.3 Confdence Intervals and Hypothesis Tests for the Population Proportion p 175
11.4 Determining the Minimum Sample Size n . . . . . . . . . . . . . . . . . . 177
11.5 Inference Procedures for the Di˙erence Between Two Population Proportions178
11.5.1 The Sampling Distribution of p̂1 − p̂2 . . . . . . . . . . . . . . . . . 178
11.5.2 Confdence Intervals and Hypothesis Tests for p1 − p2 . . . . . . . 178
11.6.2 Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
11.6.3 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
12 Inference for Variances 193
13 χ2 Tests for Count Data 195

13.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
13.2 χ2 Tests for One-Way Tables . . . . . . . . . . . . . . . . . . . . . . . . . 195
13.2.1 The χ2 Test Statistic . . . . . . . . . . . . . . . . . . . . . . . . . . 195
13.2.2 Testing Goodness-of-Fit for Specifc Parametric Distributions . . . 197
13.3 χ2 Tests for Two-Way Tables . . . . . . . . . . . . . . . . . . . . . . . . . 198
13.4.2 Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
13.4.3 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
14 One-Way ANOVA 209

14.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
14.2 One-Way ANOVA . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
14.3 Carrying out the One-Way ANOVA . . . . . . . . . . . . . . . . . . . . . 209
14.3.1 The Formulas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
14.3.2 An Example with Full Calculations . . . . . . . . . . . . . . . . . . 210
14.4 What Should We Do After a One-Way ANOVA? . . . . . . . . . . . . . . 211
14.4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
14.4.2 Fisher’s LSD Method . . . . . . . . . . . . . . . . . . . . . . . . . . 211
14.4.3 The Bonferroni Correction . . . . . . . . . . . . . . . . . . . . . . . 212
14.4.4 The Tukey Procedure . . . . . . . . . . . . . . . . . . . . . . . . . 214
14.5 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
14.6 A Few More Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
Balka ISE 1.09 CONTENTS vi
14.8 Basic Calculations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215

14.9 Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
14.10Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
15 Introduction to Simple Linear Regression 225

15.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
15.2 The Linear Regression Model . . . . . . . . . . . . . . . . . . . . . . . . . 226
15.3 The Least Squares Regression Model . . . . . . . . . . . . . . . . . . . . . 226
15.4 Statistical Inference in Simple Linear Regression . . . . . . . . . . . . . . 227
15.4.1 Model Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . 227
15.4.2 Statistical Inference for the Parameter β1 . . . . . . . . . . . . . . 227
15.5 Checking Model Assumptions with Residual Plots . . . . . . . . . . . . . . 229
15.6 The Correlation Coeÿcient and the Coeÿcient of Determination . . . . . 229
15.7 Estimation and Prediction Using the Fitted Line . . . . . . . . . . . . . . 230
15.8 Transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
15.9 A Complete Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
15.10Outliers and Infuential Points . . . . . . . . . . . . . . . . . . . . . . . . . 231
15.11Some Cautions about Regression and Correlation . . . . . . . . . . . . . . 231
15.11.1 Always Plot Your Data . . . . . . . . . . . . . . . . . . . . . . . . 231
15.11.2 Avoid Extrapolating . . . . . . . . . . . . . . . . . . . . . . . . . . 231
15.11.3 Correlation Does Not Imply Causation . . . . . . . . . . . . . . . . 232
15.12A Brief Multiple Regression Example . . . . . . . . . . . . . . . . . . . . . 232
15.13Chapter Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232
15.13.2 Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
15.13.3 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236
Chapter 1
Introduction
(No exercises.)
1
Balka ISE 1.09 2
Chapter 2
Gathering Data
J.B.’s strongly suggested exercises: 1, 2, 3, 4, 5, 6, 8, 9, 12, 15, 17, 18, 21, 24, 29, 31, 32
2.1 Introduction
2.2 Populations and Samples, Parameters and Statistics
1. A parameter is a numerical characteristic of a population, whereas a statistic is a

numerical characteristic of a sample. The value of a parameter is typically unknown,
but the value of a statistic is known once the sample is taken.
2. (a) All three-year-old walleye in this lake.

(b) The 52 fsh caught in the netting program.
(c) 273 is a statistic, since it is a value obtained from a sample.
(d) The parameter of interest is the true mean weight of all three-year-old walleye
in this lake.
(e) No, not in a practical sense. The true mean weight is not a knowable quantity,
as to fnd it we’d have to catch every three-year-old walleye in this lake.
2.3 Types of Sampling
2.3.1 Simple Random Sampling
3. (a) The 200 pacemakers in the shipment.
3
Balka ISE 1.09 2.4. EXPERIMENTS AND OBSERVATIONAL STUDIES 4
(b) The 8 pacemakers.

(c) 0% (a numerical characteristic of a sample)
(d) 10% (a numerical characteristic of the population).
(e) No. Only the top pacemakers were drawn (not all possible samples were equally
possible).
4. (a) True. Each possible sample of size 10 had the same chance of being selected,
and the sample obtained is thus a simple random sample.
(b) True.
(c) True.
(d) True (since we are sampling 10 from a population of 50).
2.3.2 Other Types of Random Sampling
5. Statements a) and b) are true, and c) is false. Sample I is a simple random sample.
Sample II is a stratifed random sample. Both sampling designs are reasonable.
2.4 Experiments and Observational Studies
6. (a) It is an experiment.
(b) The 20 volunteers.
(c) The procedures (A and B).
(d) Time until death.
(e) Yes, since it is a randomized experiment.
7. (a) It is an experiment.
(b) The types of corn.
(c) The yield of the corn.
(d) The plots of land.
8. No, it would not be reasonable. It’s an observational study, and as such does not
provide strong evidence of a causal link. There may appear to be an association, but
this relationship could easily be caused by other variables. For example, perhaps
uninvolved parents are more likely to have children that watch lots of TV, and more
likely to have children that are arrested.
Balka ISE 1.09 2.5. CHAPTER EXERCISES 5
2.5 Chapter Exercises
9. (a) All students at this university.

(b) No, it is not a simple random sample. Sampling methods like this are some-
times called convenience sampling.
(c) The volunteers may be fundamentally di˙erent from other students at the
university. (Perhaps the small giveaways and a chance of a $100 gift card
lured them in because they had very little money.) These volunteers may tend
to spend more or less on fast food than other students at the university. And
since this was a survey, the respondents may lie, or may not remember how
much they spend on fast food.
(d) No, it would not be possible. We may be able to draw a simple random sample
of students, but we cannot force these students to participate in the study.
10. (a) All lizards of these species in this region of Inner Mongolia.
(b) No, this is not simple random sampling. Only lizards the researchers could
catch were included in the sample.
(c) The researchers would have an easier time catching certain types of lizard, and
thus certain types of lizard would tend to be overrepresented in the sample.
(Perhaps larger lizards are easier to catch, or smaller lizards are easier to
catch.) It is hard to say how much of a problem this might be, but there are
defnitely possible biases in this study. But there is no better alternative—the
researchers simply have to catch the lizards by hand.
11. (a) If patients knew they were receiving vitamin C rather than the placebo, they
may be more confdent that it is helping to reduce the duration and severity
of colds. This may show up in their reporting of the results, even if there is no
real e˙ect. If the researchers knew which treatment the patient was receiving,
they may be tempted (consciously or subconsciously) to tweak the results to
refect what they want to show.
(b) This may or may not be a problem, depending on the reasons people had for
dropping out of the study. The dropouts will not cause any bias if they are
dropping out for reasons unrelated to the study. (For example, if they move out
of town or die in an unrelated accident.) If they drop out for reasons related
to the study then this may very well cause substantial bias. (For example, if
they drop out because the vitamin C is making them too sick from colds to
report the results.) In the original study, the researchers followed up with the
dropouts, and found that most were dropping out because they were getting
bored of the study and did not want to continue.
12. (a) The weight of the tumour is the response variable.

(b) The exercise variable (forced exercise, no forced exercise) is the explanatory
variable.
(c) This will not cause any bias as long as the reason for the rat catching his foot
on the treadmill is unrelated to the study variables. In the absence of further
information, this is likely the case, so the results of the study are still valid.
But if the reason for the rat catching his foot was related to the study variables,
then this may result in substantial bias. (For example, if the rat caught his foot
on the treadmill because of a lack of mobility due to a massive tumour, then
omitting this rat would be problematic and may lead to misleading results.)
13. (a) Since only the numbers 1 through 7 count, the sample is made up of the
numbers 7, 1, and 3. 98797 18334. This corresponds to the countries Panama,
Belize, El Salvador.
(b) In this case, yes. But only because none of these country names share the
same frst letter.
14. The alphabetical list (c) would provide the sample that has the least bias. (But
this is open to debate.)
The sample in a) has many sources of potential bias, including:
• Only students that showed up early to the lecture following a midterm were
included in the sample. Students that show up to the lecture after a midterm
may very well di˙er from the rest of the class in a fundamental way.
• The lecture was on a Friday morning, and students attending Friday morning
classes may di˙er from the rest of the class in a fundamental way.
• What students say their mark is on a test may very well di˙er from their mark
on the test. (Students might have a tendency say they got a higher mark than
they did.)
The sample in b) has many sources of potential bias, including:
• Only students in line for student loans are in the sample.
• The lecture was on a Friday morning, and students attending Friday morning
classes may di˙er from the rest of the class in a fundamental way.
• What students say their parents’ income is may be quite di˙erent from what
their parents’ income actually is.
While there are possible problems with sample c), they are likely not as severe as for
the other two sampling designs. There is possibly a relationship between the frst
letter of the last name and the proportion of students that are male. For example,
certain cultures may send males to university more often than females, and have
last names that tend to start with a letter near the beginning of the alphabet. But
these issues are likely not as severe as for the frst 2 sampling designs.
15. (a) All full-time female graduate students at the university.

(b) The 67 students that returned the survey.
(c) The proportion of all female students at the university that have a job outside
the university.
(d) The proportion in the sample ( 19
67 ).
(e) No, it’s a survey (a type of observational study).
(f) No, it’s a voluntary response sample.
(g) Possibly, but there could be all sorts of bias introduced. Students who work
may very well be more or less likely to send in their survey than those who do
not.
16. (a) All black bear gall bladders in Ontario.

(b) 22 grams is a characteristic of a sample, and is thus a statistic.
(c) 24 grams is a characteristic of the entire population, and is thus a parameter.
(d) Something else. It’s more of a sample of convenience (the next 3 bears en-
countered), which could possibly be extremely biased.
17. No, they do not represent a simple random sample. Values near the middle are more
likely than those on the ends, and thus middle sized sunfowers are more likely to
be selected than small or large sunfowers.
18. (a) False. Clearly the survey gives some information. As it relates to the opinions
of members of this organization, it would yield valuable information.
(b) False. This type of delay would not be a big problem in this situation.
(c) False. A sample of 912 is a fairly large sample, and can provide valuable
information.
(d) True. The population of interest is Americans in general, but only the opinions
of the members of this pro-gun organization are reported.
19. (a) All third graders in the school.

(b) It is a characteristic of a sample, and is therefore a statistic.
(c) No. The extent of the bias that might be present in this type of sample is
diÿcult to assess, but it’s not a simple random sample. (Not all possible
samples are equally likely under this sampling scheme.)
20. (a) False. The sampling design was not stated, but it is highly unlikely that the
sample represents a simple random sample.
(b) False. The wording is pretty reasonable, and would not cause much bias.
(c) False. The value 72% came from a sample, and is thus a statistic.
(d) False. A sample of size 1500 is quite a large sample for most practical purposes.
21. (a) It is an experiment (the researchers imposed a condition (the aspirin) on some
of the patients).
(b) Whether or not the patient had a heart attack.
(c) Whether or not the patient received aspirin.
(d) No. It is sometimes desirable, but not necessary.
(e) False. This is not what confounding means.
22. (a) No, the conclusions would not be as strong. Here we are discussing an ob-
servational study, and thus we could not reach conclusions of a causal nature.
The experiment in Question 21 could give strong evidence of a causal link.
(b) Physicians might not want to respond if they’ve had a heart attack. Or perhaps
they would be more likely to respond if they’ve had a heart attack, as they may
wish to contribute to research of this nature. (We can suspect possible biases
without knowing the direction or magnitude.) And, of course, any physician
who previously died of a heart attack would be unable to respond.
23. (a) False, there are 2 treatments groups (shock/no shock). (What we might call
a treatment group and a control group.)
(b) False, it is an experiment.
(c) False. Experiments can give strong evidence of a causal e˙ect.
(d) False. The response variable is the number of matching pairs.
24. (a) Task completion time.

(b) The harness variable (whether or not a worker is wearing a harness).
(c) It is an experiment (the researchers imposed an intervention, the harness, on
some workers).
(d) Yes. It appears to be a reasonable experimental design, and as such can provide
evidence of a causal link.
(e) Without the randomization, bias can be introduced. Perhaps workers who
felt they might be slowed down a great deal by the harness would choose to
not wear one. Or perhaps faster workers would choose to not wear a harness.
There is a variety of possible biases.
25. (a) 33696 91544 76248. The frst 3 numbers selected are 3, 6, and 9. So Jerry,
Bo, and Eddie go to the control group. The second set of 3 numbers is 1, 5,
4, so Pete, Renaldo, and Alphonse go to the cheese group. The others (2, 7, 8
– Tom, Little Pete, and Huey) go to the electric shock group.
(b) This may introduce bias. You might want to save Little Pete because he is
too weak and frail, or a variety of other reasons that might introduce bias.
26. There appears to be a relationship, but we cannot say it is due to the yoga. Perhaps
women who practice yoga take care of themselves and are simply healthier overall.
Based on this type of survey, we have little evidence that yoga causes improved
health.
27. a: An unmeasured variable that is related to both the response and explanatory
variables.
28. (a) Reduction in blood pressure.

(b) The types of drug.
(c) It is a randomized experiment.
(d) Yes. A well designed experiment can give strong evidence of a causal link.
29. Only (a) is true. This observational study does not give strong evidence of a causal
link. There may be a causal link, but this study certainly doesn’t show it. Note that
those who are more likely to get into an accident may be more likely to purchase
this service.
30. (a) The bonding agents.

(b) The bonding agents.
(c) It is an experiment.
(d) Yes. This type of experiment can show strong evidence of a causal link.
31. (a) An observational study (a study of coroner records, for example).

(b) If the data is available, an observational study would be best. (Perhaps records
from various companies are available.) This may not be possible, and a survey
of workers may have to suÿce.
(c) An experiment.
(d) An experiment would not be possible here. And it would be tough to get
appropriate data for an observational study. We would need to rely on a
survey. But unfortunately, even the survey results may not be very reliable,
as people may be likely to lie when faced with questions about cocaine use or
marital infdelity.
(e) An observational study (a study of coroner records, for example).
32. (a) The experimental surgery. The results from the experiment are more reliable
(less subject to bias).
(b) The medical condition of the dog. Owners of dogs that are in very poor shape
may be more willing to gamble on the experimental surgery.
Chapter 3
Descriptive Statistics
J.B.’s strongly suggested exercises: 1, 4, 7, 8, 9, 10, 11, 16, 17, 20, 21, 22, 24, 25, 36, 46,
49
3.1 Introduction
3.2 Plots for Qualitative and Quantitative Variables
3.2.1 Plots for Qualitative Variables
1. (a) Most victims (67%) were acquaintances of the male perpetrator. Slightly less
than one quarter (24%) were the male murderer’s intimate partner. Small
percentages of the victims were family members (4%) or strangers (4%).
(b) Compared to the victims of female murderers, victims of male murderers are
more likely to be acquaintances and less likely to be intimate partners or family
members.
(c) No, of course not. This is sample data, and for sample data there is always
the chance that what we observe is simply due to chance and not a real e˙ect.
Later on, we will conduct a χ2 test on this data, in order to assess how much
evidence there is of a real e˙ect.
(d) The proportions in each category are:
• Acquaintances: 61 91 ≈ .670
• Partners: 2291 ≈ .242
• Family: 914
≈ .044
• Stranger: 91 ≈ .044
4
11
Balka ISE 1.09 3.3. NUMERICAL MEASURES 12
These proportions make up the areas of the sections of the pie chart. The pie
chart for this data is illustrated in Figure 3.1.
Acquaintance
Stranger
Family
Partner
Figure 3.1: Relationship to the victim for male murderers in Finland.
3.2.2 Plots for Quantitative Variables
3.3 Numerical Measures
3.3.1 Summation Notation
2. (a) 22 (the third number in the list).

X3
(b) xi = x1 + x2 + x3 = 8 + 14 + 22 = 44.
i=1
n
X n
X
(c) xi is shorthand notation for xi .
P
xi = x1 + x2 + x3 + x4 = 8 + 14 +
i=1 i=1
22 + (−5) = 39.
(d)
P 2
xi = 82 + 142 + 222 + (−5)2 = 769.
3.3.2 Measures of Central Tendency
3.3.2.1 Mean, Median, and Mode

P
xi 1 + 5 + 2 + (−3) + 987
3. (a) x̄ = = = 198.4.
n 5
(b) When ordered from smallest to largest, the 5 observations are: −3, 1, 2, 5,
987. The median is 2, the middle value.
(c) No observation occurs more often than any other (they all occur once), and so
there is no mode.
P 1 + 5 + 2 + (−3)
(d) The extreme value is 987. When this value is removed, nxi = =
4
1.25.
(e) The ordered values are: −3, 1, 2, 5. The median is the average of the two
middle values: 1+2
2 = 1.5.
4. (a) x̄ = 2870+2620+3120+3620
4 = 3057.5.
(b) The ordered values are: 2620, 2870, 3120, 3620. The median is the average of
2870 + 3120
the two middle values: = 2995.
2
(c) Grams (the same units as the variable).
(d) Grams (the same units as the variable).
(e) Either one is perfectly acceptable, and it’s fne to report both. (There are no
extreme values that might make the mean a misleading measure of centre.)
3.3.2.2 Other Measures of Central Tendency
5. The trimmed mean is not as infuenced by extreme values as the untrimmed mean.
This can be advantageous in certain situations, as we don’t want a single value to
have undue infuence on our statistics and conclusions. A disadvantage is that the
trimmed mean omits some possibly valuable information.
3.3.3 Measures of Variability
6. (a) 6.5, −3.5, −8.5, 5.5. The deviation for the ith observation is xi − x̄, where
x̄ = 11.5. x1 − x̄ = 18 − 11.5 = 6.5, x2 − x̄ = 8 − 11.5 = −3.5, x3 − x̄ =
3 − 11.5 = −8.5, x4 − x̄ = 17 − 11.5 = 5.5.
(b) 0. The deviations always sum to 0.
7. (a) Range = P Maximum − Minimum = 3620 − 2620 = 1000.

|xi − x̄|
(b) MAD = , where x̄ = 3057.5.
n
|2870 − 3057.5| + |2620 − 3057.5| + |3120 − 3057.5| + |3620 − 3057.5|
MAD =
4
MAD = 187.5+437.5+62.5+562.5
4 = 312.5 grams.
(xi − x̄)2
P
(c) s =
2 , where x̄ = 3057.5.
n−1
(2870 − 3057.5)2 + (2620 − 3057.5)2 + (3120 − 3057.5)2 + (3620 − 3057.5)2
s2 = =
4−1
182291.7.
√
(d) s = 182291.7 = 426.9563.
(e) grams2 (the square of the units of the variable).

(f) grams (the units of the variable).
8. (a) The 4 values must be: 0, 0, 500, 500.

(b) These 4 values have a standard deviation of s = 288.6751.
9. (a) Any 4 values that are equal will work (e.g. 0,0,0,0, or 75, 75, 75, 75).
(b) If the 4 values are equal, s = 0.
10. (a) True. (The standard deviation will be greater than the variance if 0 < s < 1.)
(b) False. (s ≥ 0.)
(c) True. (The mean is a measure of location, but the standard deviation is a
measure of variability. There is no general relationship between them.)
(d) True. (Q3 is a measure of position, but the standard deviation is a measure
of variability. There is no general relationship between them.)
(e) False. (The standard deviation cannot be less than the average distance from
the mean.)
3.3.3.1 Interpreting the standard deviation
11. (a) Yes (the empirical rule would apply), since the distribution is roughly mound-
shaped.
(b) Approximately 68% of the observations would lie within 7.2 of 22.4 (within 1
standard deviation of the mean).
(c) Approximately 95% of the observations would lie within 14.4 of 22.4 (within
2 standard deviations of the mean).
(d) All or almost all of the observations would lie within 21.6 of 22.4 (within
3 standard deviations of the mean). (At most a very small proportion of
observations would fall outside of that range.) For this particular data set, no
observations fall outside of that range of values.
12. (a) Yes, Chebyshev’s theorem would apply. Chebyshev’s theorem applies to every
data set.
(b) Nothing. Chebyshev’s theorem is only informative if k > 1.
(c) At least 75% of the observations would lie within 14.4 of 22.4 (within 2 stan-
dard deviations of the mean).
(d) At least 88.9% of the observations would lie within 21.6 of 22.4 (within 3
standard deviations of the mean).
13. (a) The empirical rule would not oÿcially apply to this data set, since we would
not call this a mound-shaped distribution (it’s skewed to the right). However,
the empirical rule still works pretty well here. For the data given in the plot,
69.6% of the observations fall within 1 standard deviation of the mean, 95.2%
of the observations fall within 2 standard deviations of the mean, and 99.1%
of the observations fall within 3 standard deviations of the mean. So the
empirical rule can still be informative, even if there is some skewness.
(b) Yes, of course Chebyshev’s theorem would apply, since Chebyshev’s theorem
applies to every data set.
3.3.3.2 Why divide by n − 1 in the sample variance formula?
14. Dividing by n − 1 results in an estimator that, on average, equals the population

variance σ 2 . If we divided by n, the resulting estimator would tend to be too small.
15. 86. Since we do not know the population mean µ, we will need to estimate it with
x̄, resulting in a loss of 1 degree of freedom. In this type of scenario, the degrees of
freedom are n − 1.
3.3.4 Measures of Relative Standing
3.3.4.1 Z-scores
16. (a) z1 = x1s−x̄ = 2870−3057.5

426.9563 = −0.439. z2 =
x2 −x̄
s = 2620−3057.5
426.9563 = −1.025.
x3 −x̄ x4 −x̄
z3 = s = 426.9563 = 0.146. z4 = s = 3620−3057.5
3120−3057.5
426.9563 = 1.317.
(b) 0. The mean of all z scores is always 0.
(c) 1. The standard deviation of all z scores is always 1.
(d) The baby’s weight is very large relative to other babies in the data set. (The
baby’s weight is 4.6 standard deviations greater than the mean.)
(e) The baby’s weight is a just a little small relative to other babies in the data
set. (The baby’s weight is 0.4 standard deviations less than the mean.)
17. (a) True.

(b) True.
(c) False. Todd’s score was better than almost all writers of the test.
(d) True.
3.3.4.2 Percentiles
18. (a) 691. The ordered weights of the 8 boxes of cereal are:
684, 684, 684, 686, 686, 691, 691, 691

Balka ISE 1.09 3.4. BOXPLOTS 16
We need to fnd the 80th percentile. np = 8 × 0.80 = 6.4. We round up to the

next largest integer (7), and the 80th percentile is the 7th ranked value (691).
(b) 684. We need to fnd the 25th percentile. np = 8 × 0.25 = 2. The 25th
percentile is the average of the 2nd and 3rd ranked values: 684+684
2 = 684.
19. 171 cm. (171 is the only plausible value. 143 cm and less are far too small, and
200 cm is far too big.) ? estimate the 90th percentile of women 20 years and over
in the United States to be 170.9 cm.
3.4 Boxplots
20. (a) Approximately 35.

(b) Approximately 80.
(c) 4 (Sample A and Sample B each have 2).
(d) A little greater than 20.
21. (a) Approximately 74 mm.

(b) Approximately 66 mm.
(c) The variances of the two distributions appear to be very similar. The distri-
bution for males appears to be shifted higher than that of females. (The males
of this species seem to have a greater tail length on average.)
3.5 Linear Transformations
22. (a) x̄ = 20 × (100 + 5) = 2100.

(b) s2 = 202 ×202 = 160000. (Note that the additive constant (5) does not change
the variance.)
(c) s = 20 × 20 = 400.
(d) 20 × (90 + 5) = 1900.
23. All of the measures of location or position (mean, median, Q3 ) would change. None
of the measures of variability (IQR, variance, standard deviation) would change.
3.6.1 Basic Calculations
24. (a) x̄ = 41.25.

(b) Median = 30.
(c) s2 = 2471.583.
(d) s = 49.71502.
(e) 109.
25. (a) x̄ = 8.214286.

(b) Median = 8.4. The ordered values are −1.7, 3.1, 8.2, 8.4, 8.8, 9.6, 21.1. The
median is the fourth number in this ordered list.
(c) Q3 = 9.6. By the method given in the text: 7 × 0.75 = 5.25, so we round up
to 6 and choose the 6th value in the ordered list.
(d) IQR = Q3 −Q1 = 9.6−3.1 = 6.5. By the method given in the text: 7×0.25 =
1.75, so we round up to 2 and Q1 is the 2nd value in the ordered list. We found
Q3 in 25c.
(e) s2 = 48.7981.
(f) s = 6.985563
(g) Yes, 21.1 is an outlier. (21.1 falls outside of the range Q1 − 1.5 × IQR = −6.65
to Q3 + 1.5 × IQR = 19.35.)
26. (a) x̄ = 30.

(b) Median = 25. The ordered list is: 1, 4, 18, 32, 36, 89. Since there are 6 obser-
vations, the median is the average of the 3rd and 4th values in the ordered
list. (Median = 18+32
2 .)
(c) Q1 = 4. By the method in the text: 6 × 0.25 = 1.5, so we round up to 2 and
Q1 is the second value in the ordered list.
(d) IQR = Q3 − Q1 = 36 − 4 = 32. (Q1 = 4, Q3 = 36.) We found Q1 in 26c. We
can fnd Q3 using the same approach: 6 × 0.75 = 4.5, so we round up to 5 and
Q3 is the ffth value in the ordered list.
(e) s2 = 1036.4.
(f) s = 32.19317.
27. (a) x̄ = 149.2.

(b) Median = 6.
(c) s = 331.5316.
(d) s2 = 109913.2.
(e) If 742 is thrown out: x̄ = 1. Note that the mean has decreased a great deal.
(f) If 742 is thrown out: Median = 1. Note that the median has changed by a
much smaller amount than the mean.

(g) If 742 is thrown out: s = 11.37248. Note that the standard deviation decreased
a great deal when the outlier was discarded.
28. (a) The values are −1.7, −1.1, −1.1, −1.0, −.7, −.6, −.6, −.3, −.2, 0.0, 0.3, 0.3, 0.3, 0.5, 1.0, 1.1, 1.3, 1.8, 1.9, 2
They result in a mean of x̄ = 0.18.
(b) Median = 0.15.
(c) s2 = 1.264842.
(d) s = 1.124652.
(e) Range = 2.4 − (−1.7) = 4.1.
29. (a) Yes, the mean is greater than the median. (The large values in the right tail
a˙ect the mean more than the median.)
(b) Yes, the distribution shows some right skewness (it is stretched out to the
right).
(c) The ffth value in the fve-number summary is the maximum. The frst value
in the fve number summary is the minimum. Max − Min = 5.8 − 0.0 = 5.8.
30. (a) The values are −4.9, −4.7, −4.5, −3.9, −3.9, −3.4, −3.1, −3, −1.6, 1.1, 2.1, 2.3, 3.2, 3.4, 3.9.
These result in a mean of x̄ = −1.133333.
(b) Median = −3.
(c) Since we would in e˙ect be multiplying every value by 10 1
, every one of the
listed statistics would change. All but the variance would change by a factor
of 10
1
. The variance would change by a factor of ( 10
1 2 1
) = 100 .
3.6.2 Concepts
31. (a) Approximately 12.

(b) Approximately 2.
(c) Approximately 25.
(d) No observations are very extreme.
(e) True. The distribution is approximately symmetric, so the values of the mean
and median would be close.
(f) Boxplot A.
32. (a) Since the deviations must sum to 0 ( (xi − x̄) = 0), the ffth deviation must
P
be −1.9.
(b) This is impossible to determine; the mean could have been any value.
(xi −x̄)2 (−2.3)2 +3.72 +1.22 +(−0.7)2 +(−1.9)2
P
(c) s = 2.475884. s2 = n−1 = 5−1 = 6.13. s =
√
6.13 = 2.475884.
33. The data set might be 12, 12, 12, 12, 22 or 134, 134, 134, 134, 144, or anything else of
this form. The variance of any data set of this form is the same: s2 = 20.
34. (a) True, the median is defnitely less than 4. (It would be close to 1.)
(b) False. The distribution is skewed to the right, and the mean would be greater
than the median.
(c) False. (Very few observations in this data set are greater than 5, but 25% of
the values are greater than Q3 .)
(d) True.
(e) False. The distribution is right-skewed.
(f) False. For a right-skewed distribution such as this one, Q3 would lie farther
above the median than Q1 lies below it.
(g) This is debatable. There are some large values that could be considered out-
liers.
35. (a) Boxplot A, as it shows the right-skewness of the histogram.

(b) Boxplot C.
(c) Boxplot B, which has an interquartile range of approximately 1.2.
(d) Boxplot A and Boxplot C contain the same number of outliers.
36. (a) The mean, variance, and standard deviation would change (they would in-
crease). The median would not change, nor would Q1 , Q3 or the IQR.
(b) The mean, median, and IQR would not change. The standard deviation and
variance would increase.
37. (a) True. If the standard deviation is 0, all the values in the data set must be
equal. This implies all the values in the 5 number summary are equal.
(b) True, since all the deviations are 0.
(c) True. (The standard deviation and variance are both equal to 0.)
(d) True. (The mean and median are both equal to 0.)
(e) True. In the new data set, there would be 99 values that are equal and 1 that
is bigger. The IQR, minimum, and median would still be 0.
38. s = 0. If the standard deviation of the 4 number data set is 0, that means all values
must equal 50. Since the added value is also 50, then all numbers in the data set
are equal and s = 0.
39. (a) False. The standard deviation can be greater or less than the median—they
are measuring completely di˙erent things.
(b) False. The variance is usually greater than the standard deviation, but not if
0 ≤ s2 ≤ 1.
(c) False. The mean can be greater or less than the standard deviation—they are
measuring completely di˙erent things.
(d) True. The standard deviation is a little greater than the average distance from
the mean, which has a maximum of 1 in this case. The standard deviation of
the sample (0, 2) for example, is s = 1.414214.
(e) True. The median is usually less than Q3 , but they can be equal if there are re-
peated values in the data set. For example, if the data set is 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 12, 24,
then the median and Q3 will both equal 0.
40. (a) True. It would not be possible in this case for the distribution to the “stretched
out” to the left.
(b) True. The units of the variance are the square of the units of the variable.
(c) False. (By the same logic that tells us that Q1 is not in general exactly half
of the median.)
(d) False. Any of the values in the fve-number summary can be negative. (Recall
that the fve-number summary is: Minimum, Q1 , Median, Q3 , Maximum.)
(e) False. Outliers increase both the standard deviation and variance (and some-
times increase them a great deal).
41. (a) False. If the distribution is right-skewed, the mean will be greater than the
median.
(b) False. These two statistics are measuring completely di˙erent things. Sym-
metry would not make them equal.
(c) False. All or almost all observations will fall within 5 standard deviations of
the mean.
(d) False.
(e) False. Although values less than Q1 would usually have z-scores that are neg-
ative, it is not impossible for a value less than Q1 to be greater than the mean.
(e.g. consider the data set: −32847193487138947, 1, 1, 2, 3, 3, 8, 48, 52, 113.)
(f) True. If the range is 0, then all values in the data set must be equal. This
implies the variance and IQR will also equal 0.
42. (a) The median will equal 15, regardless of what those 2 values are.
(b) It is impossible to determine the value of the mean.
(c) It is impossible to determine the value of the variance.
43. (a) If we let x represent the original variable, and x∗ represent the transformed
variable, then we need to fnd a and b where x∗ = a + bx. The standard
deviation changed from 5 to 1, and thus the multiplicative constant must be
either b = 15 or b = − 15 . We can now solve for a in either case. The two possible
transformations that satisfy the conditions are x∗ = 12 + 15 x and x∗ = 18 − 15 x.
(b) Carrying out the transformation x∗ = 12 + 15 x, x∗ = 12 + 51 10 = 14.
44. The possible values are 52 and 352. Three consecutive digits have a sample standard
deviation of 1 (e.g. s = 1 for the values 5, 6, 7). Three values that are increasing by
100 (e.g. 500, 600, 700) will have a standard deviation of s = 100. So the missing
value could be either 52 or 352. (Verify this on your calculator or computer.)
3.6.3 Applications
45. (a) Just a little under 70.

(b) Somewhere in the high 50s.
(c) Approximately 20.
(d) It is tough to say with any certainty, but there is a touch of left-skewness,
implying the mean would likely be less than the median.
(e) The range is approximately 80. The standard deviation will likely fall some-
where in or close to the interval 80
6 to 4 , or somewhere in the neighbourhood
80
of 13 to 20. (For mound-shaped distributions, the standard deviation often

falls within or close to the interval Range
6 to Range
4 .)
46. (a) The distribution is left skewed, with a clump of outliers at 0.

(b) It would be more meaningful to report the average grade for students that
handed in the assignment, as well as the percentage of students that handed
it in. For example, we might report that 94% of students handed in this
assignment, and those students had an average grade of 20.0 out of 25.
Total number of eggs 142 · 1 + 194 · 2 + 9 · 314 · 4
47. (a) x̄ = = = 1.7075.
Total number of scars 142 + 194 + 9 + 14
(b) 2. There are 359 observations, so the median is the 180th ranked value. There
are 142 ones and 194 twos in the data set, so the median must be 2.
142(1 − 1.7075)2 + 194(2 − 1.7075)2 + 9(3 − 1.7075)2 + 14(4 − 1.7075)2
(c) s2 = =
359 − 1
0.49243.
√
s = 0.49243 = 0.7017.
48. (a) The distribution is slightly skewed to the left.

(b) The 90th percentile would be a little less than 60 kgf.
(c) There is no legitimate way to estimate this quantity from the given data. It we
were to start playing guessing games, we could guess that the 90th percentile
for adult Germans would likely be less than the 90th percentile of this data
set (since the sample was made up of males of university age).
(d) Both the mean and median would be reasonable measures of central tendency
here. It would be perfectly fne to report either or both.
(e) From the boxplot, the median looks to be approximately 52.5. Since the
distribution has slight left skewness, the mean is likely slightly less than this
quantity. (Based on the raw data used to create these plots: Median = 52.25
and x̄ = 50.91.)
(f) While the distribution shows slight left skewness, it is roughly mound-shaped,
so the empirical rule provides a bit of guidance. The standard deviation likely
falls in or close to the interval Range
6 to Range
4 . The range is somewhere between
30 and 35 kgf, so let’s estimate the range to be 33. The standard deviation is
likely in or near the interval 33 6 = 5.5 to 4 = 8.25. (Based on the raw data
33
used to create these plots: s = 7.74.)
49. (a) The distribution looks roughly symmetric, with one very large outlier at ap-
proximately 120%.
(b) The outlier looks to be approximately 85 units above the mean of the remain-
ing data. Removing the outlier would result in a decrease in the mean of
approximately 8523 . The estimated mean without the outlier is approximately
36.6 − 85
23 = 32.90. (If we return to the raw data and remove the outlier,
x̄ = 32.89.)
(c) We should be extremely reluctant to remove a real observation. When possi-
ble, we should check to make sure that an extreme observation was correctly
recorded and was not in error. If the value is correct, then we should never
simply ignore it. We should be more inclined to use measures that are not
as infuenced by extreme values (such as reporting the median instead of the
mean). Reporting the results both with and without the outlier is another
option.
(d) It depends on the reason why the machine did not record the break. If it was
simply a random occurrence that had nothing to do with the elongation of
the cord, then this is not a problem. If the reason for the malfunction was
related to the elongation, then this might be a big problem. For example, if
the machine tends to fail to record the break when the elongation percentage
is very large, then ignoring these data points might incorporate very strong
bias (our reported mean and median will be too small).
50. (a) From the boxplot, the median looks to be approximately 27 grams. The plots
do not show any major outliers or skewness, so the mean and median are
likely close in value (27 grams would also be a reasonable estimate of the
mean). (The mean and median of the raw data are both 27.2 grams.)
(b) It depends entirely on how this sample was drawn, but the sample is very
possibly strongly biased. In practice, it is impossible to get a simple random
sample of crayfsh from a lake. It is very likely that the method of capturing
the crayfsh introduced bias.
51. (a) Since this is just a linear transformation, we can carry out the transformation
on the mean. The new mean is x̄∗ = 47.5975−40
2.54 = 2.991142.
(b) Similarly for the median: 2.54 = 2.803150.
47.12−40
(c) Subtracting 40 will not change the measures of variability, and thus only the
division by 2.54 matters here. sx∗ = 2.541
× 1.212878 = 0.477511.
(d) s2x∗ = 0.4775112 = 0.2280168.
52. A linear transformation is required: x̄∗ = 2.20(x + 2) = 4.4 + 2.2x. The mean in
pounds is x̄∗ = 4.4 + 2.2 × 15.10 = 37.62. The additive constant does not a˙ect the
standard deviation, and the new standard deviation is sx∗ = 2.2 × 1.70 = 3.74.
Chapter 4
Probability
61
4.1 Introduction
4.2 Basics of Probability
4.2.1 Sample Spaces and Sample Points
1. (a) Depending on our needs, we might defne the sample space in di˙erent ways.
One possibility is the set of all 52 × 51 = 2652 orderings of two cards: S =
{(2c, 3c), (2c, 4c), . . . , (As, Ks)}.
Another possibility is the set of all pairs of cards, without regard to order.
(There are 1326 possible pairs of cards, if order does not matter.)
(b) Under the frst defnition of the sample space given in 1a there are 52 × 51 =
2652 sample points: (2c, 3c), (2c, 4c), . . . , (As, Ks).
(c) Yes, any two card ordering has the same chance occurring.
(d) 4 × 3 = 12 of the sample points are a pair of twos. P (Pair of twos) = 2652
12
=
1
221 .
2. (a) The sample space is the set of the 6 × 6 = 36 possible outcomes: S =

{(1, 1), (1, 2), . . . , (6, 6)}.
(b) There are 6 × 6 = 36 sample points: (1, 1), (1, 2), . . . , (6, 6). (Where (1, 2)
represents rolling a 1 on the frst die and a 2 on the second.)
25
Balka ISE 1.09 4.3. RULES OF PROBABILITY 26
(c) Yes, each of the sample points has a 36

1
chance of occurring.
(d) Only one simple event (1,1) results in a sum of 2. P (Sum is 2) = 36 1
.
(e) Six sample points result in a sum of 7: (1, 6), (2, 5), (3, 4), (4, 3), (5, 2), (6, 1).
P (Sum is 7) = 36
6
= 16 .
4.3 Rules of Probability
The Intersection of Events
Mutually Exclusive Events
The Union of Events
Complementary Events
3. See Figure 4.1.
(a) A ∪ B c (b) A ∩ B c (c) Ac ∪ B c
(d) Ac ∩ B c (e) (A ∪ B)c
Figure 4.1: Venn diagrams for Question 3.
4. This scenario is illustrated in Figure 4.2.
(a) P (Ac ) = 1 − P (A) = 1 − 0.6 = 0.4.

(b) P (A ∪ B) = P (A) + P (B) − P (A ∩ B) = 0.6 + 0.3 − 0.2 = 0.7.
(c) P ((A ∪ B)c ) = 1 − P (A ∪ B) = 1 − 0.7 = 0.3.
(d) No. P (A ∩ B) 6= 0, so A and B are not mutually exclusive.
A B
0.40 0.20 0.10
0.30
Figure 4.2: Venn diagram for Question 4.
(e) This one is a little harder to visualize. Ac ∩ B c = (A ∪ B)c (drawing a Venn

diagram helps to visualize this—see Figure 4.11). So P (Ac ∩ B c ) = 0.3, and
Ac and B c are therefore not mutually exclusive.
5. (a) D ∩ E c . P (D ∩ E c ) = P (O negative) = 0.07 (this can be read straight from

the table).
(b) P (O) = 0.39 + 0.07 = 0.46. P (Rh positive) = 0.36 + 0.076 + 0.025 + 0.39 =
0.851.
P (O ∪ Rh positive) = P (O) + P (Rh positive) − P (O ∩ Rh positive)

= 0.46 + 0.851 − 0.39 = 0.921.
(c) The ABO blood groups are mutually exclusive (a person is classifed into one
and only one of A, B, AB, or O), so the following events are mutually exclusive:
(A, B), (A, C), (A, D), (B, C), (B, D), (C, D).
(d) (A∩E c )∪(D∩E c ), or (A∪D)∩E c . P ((A∩E c )∪(D∩E c )) = 0.06+0.07 = 0.13.
Conditional Probability
Independent Events
6. (a) The addition rule tells us that P (A ∪ B) = P (A) + P (B) − P (A ∩ B), so

0.65 = 0.60 + 0.20 − P (A ∩ B), which implies P (A ∩ B) = 0.15.
(b) P (A|B) = P P(A∩B) 0.15
(B) = 0.20 = 0.75.
(c) P (B|A) = P P(A∩B) 0.15
(A) = 0.60 = 0.25.
(d) No, since P (A|B) =6 P (A) (0.75 =
6 0.60).
7. (a) P (B|C) = 1. If we are given that the number on the top face is a 6 (event
C), we know with certainty it is an even number (event B). Alternatively, we
could use the conditional probability formula:
P (B ∩ C) 1/6
P (B|C) = = =1
P (C) 1/6
(P (B ∩ C) = 16 because the intersection of B and C is 6, one of the six possible

equally likely outcomes.)
(b) P (C|B) = 13 . If we are given the number on the top face is an even number
(event B – the numbers 2, 4, 6), then there is a 13 probability it is a 6 (event
C). Alternatively, we could use the conditional probability formula:
P (B ∩ C) 1/6 1
P (C|B) = = =
P (B) 3/6 3
(P (B ∩ C) = 16 because the intersection of B and C is the number 6.)

(c) P (C|B c ) = 0. If we are given the number is odd (event B c , since B is the even
numbers), then the number cannot possibly be a 6. Alternatively, we could
use the conditional probability formula:
P (B c ∩ C) 0
P (C|B c ) = c
= =0
P (B ) 3/6
(P (B c ∩ C) = 0 because B c and C have no events in common.)

(d) P ((B ∩ C)|A) = 13 . B ∩ C = {6} (the only event B and C share is 6). If we
are given A has occurred (one of 4, 5, 6 has occurred), then the probability
a 6 has occurred is 13 . Alternatively, we could use the conditional probability
formula:
P (B ∩ C ∩ A) 1/6 1
P ((B ∩ C)|A) = = =
P (A) 3/6 3
(P (B ∩ C ∩ A) = 16 because the intersection of B and C and A is the number
6.)
(e) None of the pairs are independent. P (A) = 12 , P (B) = 12 , P (C) = 16 .
2
P (A|B) = 6 P (A) =⇒ A and B are not independent
=
3
6 P (A) =⇒ A and C are not independent
P (A|C) = 1 =
6 P (B) =⇒ B and C are not independent
P (B|C) = 1 =
The Multiplication Rule
8. (a)
P (Both cards are hearts) = P (First card is a heart) × P (Second card is a heart | First card is a heart)
13 12
= ×
52 51
156 1
= =
2652 17
Balka ISE 1.09 4.4. EXAMPLES 29
(b)
P (Both cards are fves) = P (First card is a 5) × P (Second card is a 5 | First card is a 5)
4 3
= ×
52 51
12 1
= =
2652 221
(c) 13
52 × 12
51 × 11
50 × 10
49 = 17160
6497400 ≈ 0.00264.
9. (a)
P (Both dice are fours) = P (First die is a four) × P (Second die is a four | First die is a four)
1 1
= × (The rolls are independent.)
6 6
1
=
36
(b) Since the rolls are independent, these two events are independent, and we can
simply multiply the two probabilities together:
3 2 6 1
P (Roll an even number, then a number greater than 4) = × = =
6 6 36 6
4.4 Examples
10. We need to fnd P (A|B) = P P(B∩A)

(A) . P (A) is given in the question, so we just need
to fnd P (B ∩A). We cannot say that P (B ∩A) = P (B)P (A), since we do not know
if A and B are independent (in fact, since P (B|A) 6= P (B), we know that A and B
are not independent). We need to fnd the probability of the intersection in another
way. What is the other bit of information we are given in the question? We are
given that P (B|A) = P P(B∩A)
(A) = 0.25, which implies P (B ∩ A) = 0.25 × 0.80 = 0.20.
Now we can calculate:
P (B ∩ A) 0.20
P (A|B) = = = 0.5
P (B) 0.40
11. (a) See Figure 4.3.

(b) P (A ∪ B) = 0.42 + 0.44 − 0.16 = 0.70.
(c) P (A ∪ B ∪ C) = 0.82.
(d) P (A|B ∩ C) = P P(A∩B∩C) 0.04 1
(B∩C) = 0.12 = 3 .
A B
0.12
0.10 0.20
0.04
0.16 0.08
0.12
0.18 C
Figure 4.3: Venn diagram for Question 11a.
12. (a) Since any one of these 1000 individuals is equally likely to be picked, this
probability is simply the number of Republicans over the total number of
individuals:
60 + 270
P (Republican) = = 0.330
1000
(b) This is the number of individuals that said they approve, over the total number
of individuals:
60 + 361 + 139
P (Approve) = = 0.560
1000
(c) The conditional probability formula can be used here, but we can also an-
swer this question quickly using a little bit of logic. We are given that the
individual is a Republican, which means we are dealing with only the 330
Republicans in the sample. Of the 330 Republicans, 60 approve, and so
P (Approve|Republican) = 330 60
. We could have used the conditional proba-
bility formula had we so desired:
P (Approve ∩ Republican) 60/1000 60

P (Approve|Republican) = = =
P (Republican) 330/1000 330
Note that P(Approve) 6= P(Approve|Republican). This means the events: The

person is a Republican, and The person said they approve of the way Obama is
handling his job as president are not independent. (Which should make sense,
if you know a little about United States politics.)
13. (a) A and B are independent in Plot B. (We can’t determine independence with
certainty by just looking at the plot, without the information that A and B
are independent for one of the plots. It can only be plot B. This plot was
created such that P (A) = 0.1, P (B) = 0.2, and P (A ∩ B) = 0.02.)
(b) Plot A: P (A|B) < P (A).
Plot B: P (A|B) = P (A).
Plot C: P (A|B) > P (A).
Plot D: P (A|B) > P (A).
Balka ISE 1.09 4.5. BAYES’ THEOREM 31
4.5 Bayes’ Theorem
14. The tree diagram for this scenario is illustrated in Figure 4.4.
A P (E ∩ A) = 0.0001 · 0.92 (Alarm sounds)

0.92
Alarm?
Ac
E
0.08 P (E ∩ Ac ) = 0.0001 · 0.08
0.0001
Explosives?
Ec P (E c ∩ A) = 0.9999 · 0.02 (Alarm sounds)
A
0.9999
0.02
Alarm?
Ac
0.98 P (E c ∩ Ac ) = 0.9999 · 0.98
Figure 4.4: The tree diagram for Question 14. E represents the event that the person is
carrying explosives and A represents the event that the alarm sounds.
(a) This is the sum of the probabilities of the two branches of the tree diagram
that lead to the alarm sounding. More formally:
P (A) = P (A ∩ E) + P (A ∩ E c )
= P (E)P (A|E) + P (E c )P (A|E c )
= 0.0001 · 0.92 + 0.9999 · 0.02
= 0.02009
(b) We are given that the alarm has sounded, so either the frst branch or third
branch of the tree diagram has occurred. What is the conditional probability
the frst branch has occurred? This is the probability of the frst branch,
divided by the sum of the probabilities of the frst and third branches. More
formally:
P (E ∩ A)
P (E|A) =
P (A)
P (E)P (A|E)
=
P (E)P (A|E) + P (E c )P (A|E c )
0.0001 · 0.92
=
0.0001 · 0.92 + 0.9999 · 0.02
= 0.00458
Balka ISE 1.09 4.5. BAYES’ THEOREM 32
L P (C ∩ L) = 0.65 · 0.22 (Plane is late)

0.22
Late?
Lc
C
0.78 P (C ∩ Lc ) = 0.65 · 0.78
0.65
Caribou?
Cc P (C c ∩ L) = 0.35 · 0.42 (Plane is late)
L
0.35
0.42
Late?
Lc
0.58 P (C c ∩ Lc ) = 0.35 · 0.58
Figure 4.5: The tree diagram for Question 15. C represents the event that the fight is a
Caribou Air fight, and L represents the event that the fight is late.
15. The tree diagram for this scenario is illustrated in Figure 4.5.
that lead to a late fight. More formally:
P (L) = P (L ∩ C) + P (L ∩ C c )
= P (C)P (L|C) + P (C c )P (L|C c )
= 0.65 · 0.22 + 0.35 · 0.42
= 0.29
(b) We are given that the fight is late, so either the frst branch or third branch
of the tree diagram has occurred. What is the conditional probability that the
frst branch has occurred? This is the probability of the frst branch, divided
by the sum of the probabilities of the frst and third branches. More formally:
P (C ∩ L)
P (C|L) =
P (L)
P (C)P (L|C)
=
P (C)P (L|C) + P (C c )P (L|C c )
0.65 · 0.22
=
0.65 · 0.22 + 0.35 · 0.42
= 0.493
Balka ISE 1.09 4.6. COUNTING RULES: PERMUTATIONS AND COMBINATIONS 33
4.6 Counting rules: Permutations and Combinations
16. (a)
�52
2 = 1326. (The order in which the cards are received is unimportant, so
this is a combinations problem.)
(b) 525 = 2, 598, 960. (The order in which the cards are received is unimportant,
�
so this is a combinations problem.)

(c) As given in problem 16b, the total number of possible hands is 52
�
5 = 2, 598, 960.
For any given suit, there are 5 = 1287 possible fushes. There are 4 suits,
�13
so there are 4 × 135 = 5148 possible fush hands. Since each of the 2598960
�
fve card hands are equally likely, the probability of getting dealt a fush is
2598960 ≈ 0.00198. (About once every 505 deals.)
5148
17. (a)
�200
4 = 64, 684, 950. In a simple random sample, the order of selection is not
important (we care only about which people have been selected).
(b) In a simple random sample every individual has the same chance of being
selected, so the probability any individual students gets chosen is 200
4
= 501
.
(We could also have used an argument based on combinations. There are 4
�200
possible samples, and 11 × 199 of these contain a specifc student. So the

� �
3
(1)(199)
probability any individual student gets selected is 1 2003 = 50
1
.)
(4)
18. (a) 25 = 53, 130. Here the order of selection is not important (we care only
�
5
which 5 people have been picked for the committee).
(b) P730 = 10, 260, 432, 000. Here the order is important (the drivers are being
assigned to 7 di˙erent routes, so among each group of 7 drivers, there are 7!
ways of assigning them to the routes).
4.7 Probability and the Long Run
19. Only A is true.
20. This scenario is illustrated in Figure 4.6.
(a) P (A ∪ B) = 0.3 + 0.8 − 0.125 = 0.975.

A B
0.175 0.125 0.675
0.025
(b) P ((A ∪ B)c ) = 1 − P (A ∪ B) = 1 − 0.975 = 0.025.

(c) P (A|B) = P P(A∩B) 0.125
(B) = 0.80 = 0.15625.
(d) P (B|A) = P P(A∩B) 0.125
(A) = 0.30 = 0.4166667.
(e) No, they are not independent, since P (A) 6= P (A|B).
(f) P (Ac ∪ B) = 0.825. Ac ∪ B is illustrated in Figure 4.7c. From the Venn
diagram in Figure 4.6, we can see that the probability of this region is 0.125 +
0.675 + 0.025 = 0.825.
Alternatively, we could have used the addition rule:
P (Ac ∪ B) = P (Ac ) + P (B) − P (Ac ∩ B)
= 0.7 + 0.80 − 0.675
= 0.825
(Why is P (Ac ∩ B) = 0.675? See Figure 4.7d and Figure 4.6. B = (Ac ∩ B) ∪
(A∩B), and P (B) = P (Ac ∩B)+P (A∩B), so P (Ac ∩B) = 0.8−0.125 = 0.675.)
(a) Ac (b) B
(c) Ac ∪ B (d) Ac ∩ B
Figure 4.7: Ac ∪ B and Ac ∩ B.
21. (a) P (A ∩ B) = 0.02. By the addition rule, 0.68 = 0.5 + 0.2 − P (A ∩ B), and this
implies P (A ∩ B) = 0.02. This scenario is illustrated in Figure 4.8.
A B
0.48 0.02 0.18
0.32
(b) P ((A ∪ B)c ) = 1 − P (A ∪ B) = 1 − 0.68 = 0.32.

(c) P (A|B) = P P(A∩B) 0.02
(B) = 0.20 = 0.10.
(d) P (B|A) = P P(A∩B) 0.02
(A) = 0.50 = 0.04.
(e) No, they are not, since P (A) 6= P (A|B).
(f) P (Ac ∪ B c ) = 0.98. There are di˙erent approaches one could use here. One
method:
P (Ac ∪ B c ) = 1 − P (A ∩ B) = 1 − 0.02 = 0.98
(The only event that is not in Ac ∪ B c is A ∩ B. See Figure 4.12e for an
illustration of this.)
(g) P (Ac ∩ B c ) = 0.32. There are di˙erent approaches one could use here. One
method:
P (Ac ∩ B c ) = 1 − P (A ∪ B) = 1 − 0.68 = 0.32
(Ac ∩ B c is everything that is outside of A ∪ B. See Figure 4.11e for an
illustration of this..)
22. (a) A and B are mutually exclusive. False, since P (A ∩ B) = 0.2 =

6 0.
(b) A and B are independent. False, since P (A∩B) =6 P (A)P (B) (0.2 6= 0.2×0.2).
P (A∩B)
(c) P (A|B) = 0. False. P (A|B) = P (B) = 0.2 = 1.
0.2
(d) P (A|B) = 1. True. P (A|B) = P P(A∩B) 0.2

(B) = 0.2 = 1.
(e) A and B c are not independent. True. P (A|B c ) = 0 = 6 P (A). It might
c c
help to remember that A and B are independent if and only if A and B are
independent.
(f) P (A ∪ B) = P (A ∩ B). True. P (A ∪ B) = P (A ∩ B) = 0.2.
(g) P (A|B) = P (B|A). True. P (A|B) = P (B|A) = 1.
23. (a) 71+19

100 = 0.9.
(b) 71
73 .
(c) No, they are not independent. P (A) = 90
100 , P (B) = 73
100 , P (A ∩ B) = 71
100 6=
P (A) × P (B).
24. (a) P (A|(B ∪ A)c ) = 0, since P (A ∩ (B ∪ A)c ) = 0 (see Figure 4.9).

(b)
P ((A ∩ B) ∩ (A ∪ B))
P (A ∩ B|A ∪ B) =
P (A ∪ B)
P (A ∩ B)
= (See Figure 4.10)
P (A ∪ B)
0.2
=
0.4 + 0.7 − 0.2
2
=
9
(a) A (b) (B ∪ A)c
Figure 4.9: Venn diagrams illustrating that A shares no outcomes with (B ∪ A)c .
(a) A ∩ B (b) A ∪ B (c) (A ∩ B) ∩ (A ∪ B)
Figure 4.10: Venn diagrams illustrating that (A ∩ B) ∩ (A ∪ B) = A ∩ B.
25. (a) 36 = 12 .
(b) 61 .
(c) 0. If we are given the number is even, it cannot possibly be a 3.
(d) 23 . If we are given the number is even, we know it is a 2, 4, or 6. Two of these 3
equally likely possibilities are at least 4 (4 and 6), and so P (The number is at least 4|The number is a 2
3.
2
(e) No. P (Even) = 36 . P (More than 3) = 36 . P (Even ∩ More than 3) = 26 . (The

numbers 4 and 6 are more than 3 and even.) 26 6= 36 × 36 , so these events are
not independent.
(f) 0. If we know the number is at least 4, it cannot possibly be 2 or less.
P (2 or less|At least 4 = P (2 or less∩at least 4)
P (at least 4)
0
) = 3/6 = 0.
4.8.2 Concepts
26. (a) The odds in favour of rolling a 5 or a 6 are 1:2. (Reduced from 2:4 – odds are
usually expressed using the smallest whole numbers that have the appropriate
ratio.)
The odds against rolling a 5 or a 6 are 2:1.
(b) The odds in favour of drawing a jack are 1:12. The odds against drawing a
jack are 12:1.
(c) 5+3
5
= 85 = 0.625.
(d) 19+1
1
= 201
= 0.05.
27. (a) See Figure 4.11.

(b) See Figure 4.12.
(a) A ∪ B (b) (A ∪ B)c
(c) Ac (d) B c (e) Ac ∩ B c
Figure 4.11: Venn diagrams illustrating that (A ∪ B)c = Ac ∩ B c .
28.
P (A∪B∪C) = P (A)+P (B)+P (C)−P (A∩B)−P (A∩C)−P (B∩C)+P (A∩B∩C)
29. C (independent events are never mutually exclusive). If two events A and B are
mutually exclusive, then P (A ∩ B) = 0. If A and B are independent, P (A ∩ B) =
P (A)P (B). P (A)P (B) cannot be 0 unless one or more of the events have probability
0, so for mutually exclusive events P (A ∩ B) 6= P (A)P (B).
30. (a) If P (A) = 0.30 and P (B) = 0.30, then P (A ∩ B) = 0.09. False. P (A ∩ B)
would equal 0.09 if A and B are independent, but this statement is not true in
general.
(a) A ∩ B (b) (A ∩ B)c
(c) Ac (d) B c (e) Ac ∪ B c
Figure 4.12: Venn diagrams illustrating that (A ∩ B)c = Ac ∪ B c .
(b) If P (A) = 0.30 and P (B) = 0.30, and P (A ∪ B) = 0.60, then A and B are
independent. False. The given probabilities imply P (A ∩ B) = 0, which does
not equal P (A) × P (B), and thus A and B are not independent.
(c) If P (A) = 0.50, and the probability that B occurs given A occurs is 0.50, then
P (A ∩ B) = 0.25. True. P (B|A) = P P(A∩B)(A) =
P (A∩B)
0.5 = 0.5. This implies
P (A ∩ B) = 0.5 × 0.5 = 0.25.
(d) If P (A) = 0, then A is independent of any event. True. If P (A) = 0, then
P (A|anything) = 0.
(e) If P (B|A) = 0.80, then P (B) > 0.80. False. The information P (B|A) = 0.80
implies only that 0 < P (A ∩ B) < 1 and 0 < P (B) < 1.
31. (a) If P (A) = 0.6, P (B) = 0.7, then 0.3 ≤ P (A ∩ B) ≤ 0.6. True. P (A ∩ B)
cannot be greater than the lesser of P (A) and P (B) (0.6 in this case). Also,
P (A ∪ B) ≤ 1 =⇒ P (A) + P (B) − P (A ∩ B) ≤ 1 =⇒ P (A ∩ B) ≥
P (A) + P (B) − 1. Here, P (A ∩ B) ≥ 0.3. Putting these inequalities together,
0.3 ≤ P (A ∩ B) ≤ 0.6. (Drawing a Venn diagram can help to visualize this.)
(b) If P (A) = 0.5, then for any event B: 0 ≤ P (A ∩ B) ≤ 0.5. True. With no
information about B we can say only that P (A∩B) ≥ 0 (since it’s a probability)
and P (A ∩ B) ≤ 0.5 (since the probability of the intersection cannot possibly
be greater than P (A).
(c) If P (A|B) = 0, then A and B are mutually exclusive. True. P (A|B) = 0 =⇒
P (A ∩ B) = 0 =⇒ A and B are mutually exclusive.
(d) Suppose that P (A) = 0.99. Event A and its complement are mutually exclu-
sive and independent. False. A and Ac are mutually exclusive, but are not
independent. P (A|Ac ) = 0 6= P (A).
(e) If P (A|B) 6= P (B) then A and B are not independent. False. Knowing
P (A|B) 6= P (B) gives us very little information about A and B or their prob-
abilities of occurring. Given only the information P (A|B) 6= P (B), we cannot
possibly tell if A and B are independent, .
32. No, they are not independent. They are mutually exclusive, so P (A ∩ B) = 0, but
P (A) × P (B) = 0.5 × 0.3 = 0.15 6= 0.
33. (a) A and B are independent. (Knowing the outcome of the frst toss gives us no
information about the outcome of the second toss.)
(b) A and B c are independent. (Knowing the outcome of the frst toss gives us no
information about the outcome of the second toss.)
(c) A and Ac are not independent. P (A|Ac ) = 0 = 6 P (A). If we know the coin
came up tails (A ), we know with certainty it did not come up heads (A) on
c
the same toss.
34. A and B are independent in plots B and D.
35. (a) P (A|B) < P (A). False. Knowing the person weighs more than 100 kg makes
if more likely they are male: P (A|B) > P (A). (The proportion of males that
weigh more than 100 kg is greater than that of females.)
(b) P (B|A) < P (B). False. Knowing the person is a male makes it more likely
they weigh more than 100 kg: P (B|A) > P (B). (The proportion of males that
weigh more than 100 kg is greater than that of females.)
(c) A and B are mutually exclusive. False. It is possible for a Canadian adult to
be both male and weight more than 100 kg.
(d) A and B are not independent. True, these are dependent events. (P (B|A) >
P (B)). For them to be independent events, the proportion of men that weigh
more than 100 kg would have to be exactly the same as the proportion of women
that weigh more than 100 kg.
36. (a) P (B|A) > P (B). A and B are not independent. Weather systems can be
large, and take several days to pass through. If it rains next Tuesday, it is
more likely to rain on the next the day than if there were bright sunshine on
Tuesday.
(b) P (B|A) = P (B). These events are independent. Knowing the weather next
Tuesday will tell us nothing about the weather 10 years from now.
(c) P (B|A) > P (B). These events are not independent. People who do not fnish
high school are more likely to go to prison.
37. (a) No, they are not mutually exclusive. If the frst coin comes up heads and the
second tails, they will both pass the part. P (A ∩ B) = 14 .
(b) Yes, they are independent. P (A) = 12 , P (B) = 12 , and P (A ∩ B) = 14 .
38. (a) The probability that Daniel passes the assignment is a tiny bit greater than
0.40. He has a near zero chance of passing if he is guessing randomly, so his

probability of passing is just a bit larger than 0.5 × 0.8 + 0.5 × 0 = 0.40.
(Daniel’s probability of passing when guessing randomly is 0.0025. We don’t
yet know how to fnd this probability, but it should be clear that his probability
of passing when guessing randomly is very small.)
(b) No, they are not mutually exclusive. (It is possible that they both pass the
test.)
(c) No, they are not independent. If Magnus passes the assignment, Daniel will
also pass if he fipped heads. (And almost surely fail if he fipped tails,
since he is randomly guessing in that event.) If Magnus fails the assignment,
Daniel will almost surely fail. The information that Magnus passes makes it
much more likely that Daniel passes. P (Daniel passes|Magnus passes) ≈ 0.5,
P (Daniel passes|Magnus fails) ≈ 0.
39. (a) Probabilities can be negative. False. Probabilities must lie between 0 and 1.
(b) For any two events A and B, P (A|B) > P (A). False. P (A|B) can be greater
than, less than, or equal to P (A).
(c) If A and B are independent, then P (A|B) = 0. False.
(d) If two cards are drawn from a standard deck, the probability both cards are
red is greater if the cards are drawn with replacement than if they were drawn
without replacement. True. 26 26 26 25
52 × 52 > 52 × 51 .
(e) If two events A and B are independent, then their complements (Ac and B c )
are also independent. True. This isn’t terribly diÿcult to prove, but it requires
a step that might more confusing than it’s worth. It may help to think of it
in these terms: Two events A and B are independent if knowing whether A
has occurred or A has not occurred does not change the probability of B. The
statement “A has occurred or A has not occurred” is equivalent to the statement
“Ac has not occurred or Ac has occurred.” If A and B are independent, so are
their complements.
(f) If A and B are mutually exclusive, then P (A|B) = 0. True. If A and B are
mutually exclusive, P (A ∩ B) = 0, which implies P (A|B) = P P(A∩B) 0
(B) = P (B) =
0.
40. (a) If P (A) = P (B), then A and B are independent. False. We cannot possibly tell
if A and B are independent based solely on the information that P (A) = P (B).
(b) If P (A) > P (B), then P (A|B) ≥ P (B|A). True. P (A|B) = P P(A∩B) (B) and
P (A∩B )
P (B|A) = P P(A∩B) P (A∩B)
(A) . If P (A) > P (B), then P (B ) > P (A) .
(c) If A and B are independent, P (A ∪ B) = P (A) + P (B). False.
(d) If A and B are mutually exclusive, P (A ∪ B) = P (A) + P (B). True. If A and
B are mutually exclusive, P (A ∩ B) = 0 and so P (A ∪ B) = P (A) + P (B) −
P (A ∩ B) = P(A) + P(B).
(e) If P (A|B) = 1, then P (B|A) = 1. False. Consider this counterexample: Roll a
single die once and let A represent rolling an even number, and let B represent
rolling a 6. Here, P (A|B) = 1 but P (B|A) = 13 . (If B is a subset of A, this
does not imply A is a subset of B.)
4.8.3 Applications
41. (a) P (At least 1 is mislabelled) = 1−P (None are mislabelled) = 1−0.754 ≈ 0.68.
(b) No, it’s not a reasonable assumption (for a variety of reasons). For example,
some restaurants would be more likely than others to serve mislabelled sushi,
and thus mislabelled sushi would likely be clumped together. If, say, the frst
3 pieces you get are mislabelled, there is a very good chance the fourth one
will also be mislabelled.
42. Let M represent the event the student is male, and J represent the event the
student has a job. A Venn diagram illustrating the counts in this situation is given
in Figure 4.13.
M J
15 5 20
10
(a) P (J c ) = 25
50 = 0.5.
(b) c
P (M ∩ J ) = 50 15
= 0.3.
(c) 5
P (M |J) = 25 = 0.2.
(d) P (M |M c ∩ J) = 0.
(e) No, in this question job status and gender are not independent. P (J) = 0.5
but P (J|M ) = 20 5
. Since 20
5
6 0.5, job status is not independent of gender (if
=
a student is randomly selected from this group of students).
43. (a) 41+27+36+36+32+16

459 ≈ 0.4096.
(b) 133
459 ≈ 0.2898.
(c) 36+16
41+27+36+36+32+16 ≈ 0.2766.
(d) 27+32+36+16
41+27+36+36+32+16 ≈ 0.5904.
(e) 56
56+37+53 ≈ 0.3836.
(f) 27+32
37+43+27+32 ≈ 0.4245.
(g) 27+32+36+16
37+43+27+32+53+28+36+16 ≈ 0.4081.
(h) No, these events are not independent. P (University) = 133

459 ≈ 0.2898, which
di˙ers from P (University|Smoker) = 41+27+36+36+32+16
36+16
≈ 0.2766.
(i) No, these events are not mutually exclusive. 52 of the people in the sample
have a university education and are smokers.
44. (a) P (Large) = 42+507 549

6272 = 6272 ≈ 0.088.
(b) P (Small, moderate, or large) = 201+2420+209+2769+42+507
6272
6148
= 6272 ≈ 0.980.
(c) P (Cancer) = 14+201+209+42
6272
466
= 6272 ≈ 0.074.
(d) P (Cancer|Never/Seldom) = 14+110 14 14
= 124 ≈ 0.113.
(e) 42 42
P (Cancer|Large) = 42+507 = 549 ≈ 0.077.
(f) 42
P (Large|Cancer) = 14+201+209+42 42
= 466 ≈ 0.090.
(g) P (Cancer|Small or Moderate) = 201+209+2420+2769 = 5599
201+209 410
≈ 0.073.
45. Let G represent the event that the randomly selected student attended the Uni-
versity of Guelph, and F represent the event that the randomly selected student is
female.
16, 783
(a) P (G) = = 0.3214.
52, 217
27, 278
(b) P (F ) = = 0.5224.
52, 217
(c) P (G|F ) = 10,417
27,278 = 0.3819. (This is the number of female U of G students,
over the total number of female students.)
(d) P (F |G) = 10,417
16,783 = 0.6207. (This is the number of female University of Guelph
students, over the total number of U of G students.)
P (F ∩ (G ∪ Lc ))
(e) We require: P (F |G ∪ Lc ) =
P (G ∪ Lc )
P (G ∪ L ) is the probability that a randomly selected student attended the
c
University of Guelph and not Laurier (which means they attended Guelph or
Waterloo):
16, 783 + 24, 377 41, 160
P (G ∪ Lc ) = =
52, 217 52, 217
P (F ∩ (G ∪ Lc )) is the probability that a randomly selected student is a female
student at Guelph or Waterloo (not Laurier):
10, 417 + 10, 607 21, 024
P (F ∩ (G ∪ Lc )) = =
52, 217 52, 217
We can now calculate the fnal answer:
P (F ∩ (G ∪ Lc )) 21, 024/52, 217 21, 024
P (F |(G ∪ Lc )) = = = = 0.5108
P (G ∪ Lc ) 41, 160/52, 217 41, 160
46. (a) P (Good|Man) = 23

30 ≈ 0.767.
(b) P (Man|Woman) = 0.
(c) No, these events are not independent. P (Good) = 40
65 ≈ 0.615, but P (Good|Man) =
30 ≈ 0.767. (Males were more likely than females to call the dining experience
23
good.)
(d) These events are mutually exclusive and not independent.
47. (a) 0.99 × 0.98 = 0.9702.

(b) (1 − 0.99) × (1 − 0.98) = 0.0002.
(c) P (At least 1 goes o˙) = 1 − P (Neither goes o˙) = 1 − 0.0002 = 0.9998.
(d) B or C, as we are interested in the probability of being woken up. (We want
the probability of the event that at least one alarm goes o˙, or the probability
of its complement, the event that neither alarm goes o˙.)
48. (a) 4
52 × 4
52 = 1
169 ≈ 0.006.
(b)
P (Two kings ∩ At least 1 king)

P (Two kings | At least 1 king) =
P (At least 1 king)
P (Two kings)
=
1 − P (No kings)
4
× 4
= 52 4852 2
1 − ( 52 )
= 0.04
(c)
P (Exactly one is red) = P (First card is red and second is black) + P (First card is black and second i
26 26 26 26
= × + ×
52 52 52 52
= 0.5
(d)
P (At least one card is red) = 1 − P (Neither card is red)

26 26
=1− ×
52 52
= 0.75
(e) Drawing a king is independent of drawing a red card (there are 2 red kings and
2 black kings in the deck). The two draws are independent, since the cards
are drawn with replacement. So:
P (At least one king and at least one red card) = P (At least one king) × P (At least one red card)
= (1 − P (no kings)) × (1 − P (no red cards))
48 26
= (1 − ( )2 ) × (1 − ( )2 )
52 52
≈ 0.1109
49. (a) 4
52 × 3
51 = 1
221 ≈ 0.0045.
(b)
P (Two kings ∩ At least 1 king)

P (Two kings | At least 1 king) =
P (At least 1 king)
P (Two kings)
=
1 − P (No kings)
4
× 3
= 52 48 51 47
1 − 52 × 51
≈ 0.0303
(c)
P (Exactly one is red) = P (First card is red and second is black) + P (First card is black and second i
26 26 26 26
= × + ×
52 51 52 51
≈ 0.5098
(d)
P (At least one card is red) = 1 − P (Neither card is red)

26 25
=1− ×
52 51
≈ 0.7549
(e) (One way to approach this is to draw a tree diagram, with 4 branches for each
card: red king, black king, red non-king, black non-king.) The probability of
getting at least one king and at least one red card is: 52
2
× 51 2 26 24
51 + 52 × 51 + 52 ×
4 24 2 298
51 + 52 × 51 = 52×51 ≈ 0.1123680.
50. Let F represent the event that Mateo takes his Ferrari, and A represent the event
that he gets into an accident. P (F ) = P (2 heads) = 12 · 12 = 41 . The tree diagram
for this scenario is illustrated in Figure 4.14.
A P (F ∩ A) = 0.25 · 0.08 (Accident)

0.08
Accident?
Ac
F
0.92 P (F ∩ Ac ) = 0.25 · 0.92
0.25
Ferrari?
Fc P (F c ∩ A) = 0.75 · 0.04 (Accident)
A
0.75
0.04
Accident?
Ac
0.96 P (F c ∩ Ac ) = 0.75 · 0.96
Figure 4.14: The tree diagram for Question 50. F represents the event that Mateo takes
his Ferrari and A represents the event that he gets into an accident.
that lead to an accident. More formally:
P (A) = P (A ∩ F ) + P (A ∩ F c )
= P (F )P (A|F ) + P (F c )P (A|F c )
= 0.25 · 0.08 + 0.75 · 0.04
= 0.05
(b) 0.08. (This is given in the question.)

(c) We are given that Mateo has gotten into an accident, so either the frst branch
or third branch of the tree diagram has occurred. What is the conditional
probability the frst branch has occurred? This is the probability of the frst
branch, divided by the sum of the probabilities of the frst and third branches.
More formally:
P (F ∩ A)
P (F |A) =
P (A)
P (F )P (A|F )
=
P (F )P (A|F ) + P (F c )P (A|F c )
0.25 · 0.08
=
0.25 · 0.08 + 0.75 · 0.04
= 0.40
51. (a) 0.8 · 0.02 + 0.05 · 0.04 + 0.15 · 0.14 = 0.039.

(b) 0.8·0.02+0.05·0.04+0.15·0.14
0.15·0.14
≈ 0.538.
52. (a) 0.55 · 0.40 + 0.35 · 0.90 + 0.10 · 0.55 = 0.59.

(b) 0.55·0.40+0.35·0.90+0.10·0.55
0.55·0.40
= 0.373.
53. (a) It often helps to illustrate this type of problem with a tree diagram, as in
Figure 4.15.
P (Respond “Yes”) = P (Heads) + P (Tails ∩ Stolen)
= 0.5 + 0.5 · 0.2
= 0.6
Y Respond “Yes”
0.2
Stolen?
H N
0.5 0.8 Respond “Yes”
Heads
or Tails?
T Y Respond “Yes”
0.5 0.2
Stolen?
N
0.8 Respond “No”
Figure 4.15: Tree diagram for the random response problem.
(b) There are 3 branches of the tree that lead to a “Yes” response. Combined,
these 3 branches yield us a probability of “Yes” of 0.60. In two of these 3
branches, the person has actually stolen from their employer.
P (Stolen ∩ Yes) 0.5 · 0.2 + 0.5 · 0.2 1
P (Stolen|Yes) = = =
P (Yes) 0.60 3
This is the ratio of the sum of the probabilities of the branches in which the
person has both stolen and responded “Yes”, to the probability that a person
gives a “Yes” response.
(c) This question is a little bit trickier. We know that:
P (Yes) = P (Heads) + P (Tails ∩ Stolen) = P (Heads) + P (Tails)P (Stolen)
(P (Tails ∩ Stolen) = P (Tails) × P (Stolen), since getting tails on the coin
toss is independent of whether the respondent has actually stolen from their
employer.) The probability of getting heads is assumed to be 0.50 (we’re
assuming a fair coin), and thus:
P (Yes) = 0.50 + 0.50 × P (Stolen)
Since the best estimate of P (Yes) is 0.67 (67% of respondents said yes), the
best estimate of P (Stolen) is 0.67−0.5
0.5 = 0.34.
Keep in mind that most people fnd it easier to use the tree diagram approach
for these types of questions.
54. The probability of getting a “yes” response is:

4 2
P (Yes) = + × P (Cheat)
6 6
Since 82% of people said “yes”, the estimated probability of a “yes” response is 0.82,
and so:
4 2
0.82 = + × Estimated proportion that cheated
6 6
0.82− 46
So the estimated proportion of these people who have cheated is: 2 = 0.46.
6
55. (a) P (none have the disorder) = (1 − 1000
1 30
) = 0.9704.
(b) You will have to sample more than 500 mice if the frst 500 mice do not have
the disorder:
P (First 500 do not have the disorder) = 0.999500 = 0.6064
Note that this probability is quite a bit bigger than 50%. Surprising?
(c) If we sample k mice, the probability that at least one has the disorder is:
P (At least one has the disorder) = 1−P (None have the disorder) = 1−0.999k
We need to fnd the minimum value of k such that 1 − 0.999k ≥ 0.50. This
implies we need the value of k such that 0.999k ≤ 0.50. We could use trial-
and-error, or we could fnd k a little more eÿciently using logarithms:
log(0.50)
0.999k ≤ 0.50 =⇒ k·log(0.999) ≤ log(0.50) =⇒ k ≥ =⇒ k ≥ 692.8.
log(0.999)
We would need to sample at least 693 mice to have at least a 50% chance of
fnding a mouse with the defect. Would you have guessed a value this large?
56. (a) P (System failure) = P (All components fail) = 0.324 ≈ 0.01048.

(b) Here we need to fnd k such that 0.32k < 0.0001. We could use trial and
error, but it’s usually a little more eÿcient to use properties of logs: 0.32k <
0.0001 =⇒ kln(0.32) < ln(0.0001) =⇒ k > ln(0.0001) ln(0.32) =⇒ k > 8.083257.
We would need at least 9 components connected in parallel to ensure the
probability of failure is less than 0.0001.
57. (a) 0.775 = 0.2707.

(b) 0.7710 = 0.0732.

(c) Here we need to fnd k such that 1 − 0.77k ≥ 0.99990. We could use trial
and error, but it’s usually a little more eÿcient to use properties of logs:
1 − 0.77k ≥ 0.99990 =⇒ 0.77k ≤ 0.0001 =⇒ kln(0.77) ≤ ln(0.0001) =⇒
k ≥ ln(0.0001)
ln(0.77) =⇒ k > 35.23941. Since k must be an integer, we would need
at least 36 components connected in parallel to have a probability of failure
this low.
58. (a) 0.9999 · 0.0001 = 0.00009999.

(b) 0.99999 · 0.0001 = 0.0000991.
(c) This will occur as long as the frst 2000 do not all not have the disorder:
P (First 2000 have none with the disorder) = 0.99992000 . The probability we
are looking for here is the complement of this: P (First 2000 have at least one with the disorder) =
1 − 0.99992000 = 0.181.
59. (a) 178

200 = 0.89.
(b) 4+20
20
= 0.833.
(c) No. P (A thought it a forgery) = 0.11, P (A thought it a forgery|B thought it a forgery) =
0.8333. Even though they are viewing the piece independently (using the En-
glish language meaning of the term), they are experts, and knowing that one
of them thought the piece a forgery makes it much more likely it is a forgery,
which makes it more likely the other will think so as well.
60. Let J represent the event that John goes to the party, and S represent the event
that Stephanie goes to the party. In the question we are given the conditional
probabilities P (J|S) = 0.94 and P (J|S c ) = 0.03.
(a) This scenario is illustrated in Figure 4.16.

P (J) = P (S)P (J|S) + P (S c )P (J|S c ) = 0.8 · 0.94 + 0.2 · 0.03 = 0.758. (This is
the sum of the probabilities of the frst and third branches in the tree diagram.)
(b) This scenario is illustrated in Figure 4.17.
P (J) = 0.38
=⇒ P (S)P (J|S) + P (S )P (J|S c ) = 0.38
c
=⇒ P (S) · 0.94 + (1 − P (S)) · 0.03 = 0.38

=⇒ P (S) = 0.3846
(c) They are not mutually exclusive since they can both attend the party. They
are not independent; John is much more likely to go to the party if Stephanie
goes.
61. Most people fnd it helpful to draw a tree diagram for this type of problem (see
Figure 4.18).
J P (S ∩ J) = 0.80 · 0.94 (John goes)

John 0.94
goes? Jc
S
0.06 P (S ∩ J c ) = 0.80 · 0.06
0.80
Stephanie
goes?
Sc P (S c ∩ J) = 0.20 · 0.03 (John goes)
J
0.20
John 0.03
goes? Jc
0.97 P (S c ∩ J c ) = 0.20 · 0.97
Figure 4.16: The tree diagram for Question 60a.
J P (S ∩ J) = P (S) · 0.94 (John goes)

John 0.94
goes? Jc
S
0.06 P (S ∩ J c ) = P (S) · 0.06
P (S)
Stephanie
goes?
Sc P (S c ∩ J) = P (S c ) · 0.03 (John goes)
J
P (S c )
John 0.03
goes? Jc
0.97 P (S c ∩ J c ) = P (S c ) · 0.97
Figure 4.17: The tree diagram for Question 60b.
P (+ve test) = P (Defect ∩ +ve test) + P (No defect ∩ +ve test)

= 0.00001 · 0.99 + (1 − .00001) · 0.005
= 0.00500985
+ve P (D ∩ +ve) = 0.00001 · 0.99
Positive 0.99
test? -ve
D 0.01 P (D ∩ −ve) = 0.00001 · 0.01
0.00001
Defect?
Dc
+ve P (Dc ∩ +ve) = 0.99999 · 0.005
0.99999
Positive 0.005
test? -ve
0.995 P (Dc ∩ −ve) = 0.99999 · 0.995
(a)
Figure 4.18: Tree diagram for the genetic defect problem.
(b)
P (Defect ∩ +ve test)

P (Defect | +ve test) =
P (+ve test)
0.00001 · 0.99
=
0.00001 · 0.99 + (1 − 0.00001) · 0.005
= 0.001976
Note that there is very little chance the person has the defect, even though
they tested positive.
62. (a) 0.99726, as given in the table.

(b) 1 − 0.99726 = 0.00274.
(c) This is impossible to determine from the given information (but could be
fgured out from similar information given at the StatsCan site).
(d) 0.99775 · 0.99751 · 0.99726 · (1 − 0.99702) = 0.002957765.
(e) 0.99775 · 0.99751 · 0.99726 = 0.9925386.
Chapter 5
Discrete Random Variables and

Discrete Probability Distributions
30, 48, 51, 53, 54, 57, 58, 62, 64, 66 (a-d).
5.1 Introduction
5.2 Discrete and Continuous Random Variables
1. (a) Discrete.
(b) Continuous.
(c) Continuous.
(d) Discrete.
(e) Continuous.
(f) Discrete.
(g) Discrete.
5.3 Discrete Probability Distributions
2. X can take on the values 0, 1, and 2.

P (X = 0) = 39 52 × 51 = 2652 = 34 . (Don’t spend much time trying to reduce the
38 1482 19
fractions in this question—focus on the topic at hand.)
51
Balka ISE 1.09 5.3. DISCRETE PROBABILITY DISTRIBUTIONS 52
13 39 39 13 1014 13
P (X = 1) = 52 × 51 + 52 × 51 = 2652 = 34
2652 = 17 .
13 12 156 1
P (X = 2) = × =
52 51
In summary, the probability distribution of X is:
x 0 1 2
19 13 1
p(x) 34 34 17
3. (a) Discrete.
(b) 0.1 (since the probabilities must sum to 1).
(c) The most likely value is 30.
(d) P (X < 32) = P (X = 10) + P (X = 20) + P (X = 30) = 0.2 + 0.2 + 0.5 = 0.9.
(e) P (X < 25|X < 35) = 0.40.9 ≈ 0.44.
P (X < 25 ∩ X < 35)

P (X < 25|X < 35) =
P (X < 35)
P (X = 10 or X = 20)
=
P (X = 10 or X = 20 or X = 30)
0.2 + 0.2
=
0.2 + 0.2 + 0.5
0.4
=
0.9
4. (a) and (c) are valid discrete probability distributions, since for these distributions
the probabilities all lie between 0 and 1 and they sum to 1. (Note that some discrete
random variables can take on negative values and non-integer values.) (b) is not a
valid discrete probability distribution, since the probabilities do not sum to 1. (d)
is not a valid discrete probability distribution, since one of the probabilities listed
in the table is negative.
5.3.1 The Expectation and Variance of Discrete Random Variables
5.3.1.1 Calculating the expected value and variance of a discrete ran-

dom variable
5. (a) X is discrete (it takes on only 5 possible values).

(b) 0.30 (since the probabilities must sum to 1).
(c) 5.0.
(d) P (−1 < X < 4) = P (X = 0) + P (X = 2.5) = 0.1 + 0.1 = 0.2.
(e)
P
E(X) = xp(x) = −2.5·0.10+0·0.10+2.5·0.10+5.0·0.40+7.5·0.30 = 4.25.
Balka ISE 1.09 5.3. DISCRETE PROBABILITY DISTRIBUTIONS 53
(f) σ = 3.172.
σ 2 = E[(X − µ)2 ] = (x − µ)2 p(x) = (−2.5 − 4.25)2 · 0.10 + (0 − 4.25)2 · 0.10 +
P
(2.5 − 2 2 2
√ 4.25) · 0.10 + (5.0 − 4.25) · 0.40 + (7.5 − 4.25) · 0.30 = 10.0625.
σ = 10.0625 = 3.172.
6. Let X represent the number of babies.
(a) µ =
P
xp(x) = 1 · 0.988 + 2 · 0.01187 + 3 · 0.000128 + 4 · 0.000002 = 1.012132.
(b) σ = 0.11069.
E(X 2 ) = x2 p(x) = 12 · 0.988 + 22 · 0.01187 + 32 · 0.000128 + 42 · 0.000002 =
P
1.036664.
√ σ 2 = E(X 2 ) − (E(X))2 = 1.036664 − 1.0121322 = 0.01225281.
σ = 0.01225281 = 0.11069.
(c) P (X ≥ 2) = 0.01187 + 0.000128 + 0.000002 = 0.012. Equivalently, P (X ≥
2) = 1 − P (X = 1) = 1 − 0.988 = 0.012.
0.01187
(d) P (X = 2|X ≥ 2) = = 0.989.
0.01187 + 0.000128 + 0.000002
0.000128
(e) P (X = 3|X ≥ 3) = = 0.985.
0.000128 + 0.000002
7. (a) Probabilities of impossible events can be negative. False. (Probabilities must
lie between 0 and 1.)
(b) If a random variable can take on an infnite number of possible values, then it
cannot be a discrete random variable. False. (Some discrete random variables
take on a countably infnite number of possible values.)
(c) The mean of a discrete random variable cannot be negative. False. (Some
discrete random variables take on negative values and have negative means.)
(d) The standard deviation of a discrete random variable cannot be negative. True.
(Standard deviations cannot be negative.)
(e) The mean of a discrete random variable cannot be greater than its standard
deviation. False. (The mean may be greater than, less than, or equal to the
standard deviation.)
5.3.1.2 Properties of Expectation and Variance
8. (a) µX+Y = µX + µY = 8.7 + 14.9 = 23.6.

q √
(b) σX+Y = σX 2 + σ2 =
Y 18.52 + 212 = 27.9866.
q √
(c) σX−Y = σX 2 + σ2 =
Y 18.52 + 212 = 27.9866.
9. Yes, they are independent. Since the Professor of Course B is simply assigning
grades at random, knowing a student’s grade in that course yields no information
about their grade in course A.
Balka ISE 1.09 5.4. THE BERNOULLI DISTRIBUTION 54
10. Let X represent the time to completion of Step 1 (µX = 6, σX = 1.2).
Let Y represent the time to completion of Step 2 (µY = 125, σY = 15).
Let Z represent the time to completion of Step 3 (µZ = 17, σZ = 3).
Let T represent the time to completion of the entire process: T = X + Y + Z. We

need to fnd the mean and standard deviation of T .
(a) E(T ) = E(X + Y + Z) = E(X) + E(Y ) + E(Z) = 6 + 125 + 17 = 148 minutes.

(b) Since the 3 steps are assumed to be independent, V ar(T ) = V ar(X + Y +
Z) = √V ar(X) + V ar(Y ) + V ar(Z) = 1.22 + 152 + 32 = 235.44 minutes2 .
σT = 235.44 = 15.344 minutes.
(c) Dividing T by 60 changes the scale of measurement to hours. E( 60 T
) = E(T )
60 =
V ar(T )
60 = 2.467 hours. V ar( 60 ) = 602 = 0.0654 hours2 . σT /60 = σ60T = 15.344
148 T
60 =
0.2557 hours.
5.4 The Bernoulli Distribution
11. (a) Let X represent the number of eights that are drawn. Then X has the Bernoulli
distribution with p = 13
1
. That is:
1 x 12 1−x
P (X = x) = ( ) ( ) for x = 0, 1.
13 13
(b) µ = p = 13
1
.
q q
(c) σ = p(1 − p) = 13
1 12 12
p
· 13 = 169 ≈ 0.266.
5.5 The Binomial Distribution
12. (a) P (X = 3) = 15
� 3 12 = 0.250.
3 0.2 (1 − 0.2)
(b) Using the binomial probability mass function with n = 15 and p = 0.2:
P (X ≤ 3) = P (X = 0) + P (X = 1) + P (X = 2) + P (X = 3)
= 0.0352 + 0.1319 + 0.2309 + 0.2501
= 0.648
(c) P (X > 3) = 1 − P (X ≤ 3) = 1 − 0.648 = 0.352.

(d) P (2 ≤ X < 5) = P (X = 2) + P (X = 3) + P (X = 4) = 0.231 + 0.250 + 0.188 =
0.669.
Balka ISE 1.09 5.5. THE BINOMIAL DISTRIBUTION 55
P (X = 3 ∩ X ≤ 4) P (X = 3)
(e) P (X = 3|X ≤ 4) = =
P (X ≤ 4) P (X ≤ 4)
P (X = 3) 0.2501
= = =
P (X = 0) + P (X = 1) + P (X = 2) + P (X = 3) + P (X = 4) 0.8358
0.2993.
(f) µX =pE(X) = np = 15 · 0.2 = 3.
√
(g) σ = np(1 − p) = 15 · 0.2 · 0.8 = 1.549.
13. Let X represent the number of newborns with major structural or genetic birth
defects. Then X has a binomial distribution with parameters n = 50 and p = 0.03.
(a) P (X = 2) = 50 2 48 = 0.2555.
�
2 0.03 (1 − 0.03)
(b)
P (X ≤ 2) = P (X = 0) + P (X = 1) + P (X = 2)

50 0 50 50 1 49 50
= 0.03 (1 − 0.03) + 0.03 (1 − 0.03) + 0.032 (1 − 0.03)48
0 1 2
= 0.2181 + 0.3372 + 0.2555
= 0.8108
(c) P (X = 10) = 50 10 40 = 1.8 × 10−6 .

�
10 0.03 (1 − 0.03)
(d) E(X)p= np = 50 · 0.03 = 1.5.
√
(e) σ = np(1 − p) = 50 · 0.03 · 0.97 = 1.206.
14. Which of the following statements are true?
(a) The variance of a binomial distribution is always greater than the mean. False.
(The variance (np(1 − p)) must be less than or equal to the mean (np).)
(b) A binomial random variable can be negative. False. (A binomial random
variable represents a count, and its value must be at least 0.)
(c) The mean of a binomial random variable can be negative. False.
(d) The variance of a binomial random variable can be negative. False.
(e) A binomial random variable is a discrete random variable. True.
(f) Every discrete random variable is a binomial random variable. False. (There
are many other discrete probability distributions.)
(g) A binomial random variable represents a count. True.
(h) The mean of a binomial random variable is np. True.
(i) The standard deviation of a binomial random variable is np(1 − p). True.
p
(j) For any given n, the variance of a binomial random variable is greatest when
p = 0.5. True, since p(1 − p) is greatest when p = 0.5.
Balka ISE 1.09 5.5. THE BINOMIAL DISTRIBUTION 56
5.5.1 Binomial or Not?
15. (a) Not binomial. The number of hearts would not have a binomial distribution,
since the trials are not independent (the probability of getting a heart will
change as cards are removed from the deck).
(b) Binomial. The number of times heads comes up will have a binomial distri-
bution with n = 10 and p equal to the probability of heads on any individual
toss.
(c) Not binomial. The number of putts that are holed will not have a binomial
distribution. The number of trials is fxed, each trial is either a success or a
failure, and we are counting the number of successes, but the probability of
success is changing and the trials are not independent.
(d) Not binomial. The number of newborn babies would not have a binomial
distribution. The conditions of the binomial distribution are defnitely not
satisfed here (there is not even a fxed number of trials).
(e) Not binomial. The amount of money withdrawn would defnitely not have a
binomial distribution. (This random variable is not even a count of a number
of successes.)
16. (a) Binomial. All of the conditions of a binomial distribution are satisfed here.
(b) Not binomial. While this random variable is a count (that almost always
takes on the value 10), it does not represent the number of successes in a fxed
number of trials.
(c) Binomial. The probability of rolling a sum of 7 on a pair of dice is 16 , so here
the number of sevens would have a binomial distribution with n = 14 and
p = 16 .
(d) Not binomial. While the number of children the woman has is a count, it does
not represent the number of successes in a fxed number of trials.
(e) Not binomial. While the number of times the person uses the washroom is
a count, it does not represent the number of successes in a fxed number of
trials.
(f) Not binomial. The weight is not a count.
5.5.2 A Binomial Example with Probability Calculations
17. (a) 0.2499. If they are randomly pulling cards, the number of times the king
of spades is drawn has a binomial distribution with n = 50 and p = 52 1
.
P (X ≥ 2) = 1 − P (X = 0) − P (X = 1) = 0.2499.
(b) No. This experiment yields no evidence that they have ESP, but we cannot
say there is strong evidence they do not. There is, however, strong evidence
against their claimed probability of drawing the king of spades. If their claimed
Balka ISE 1.09 5.6. THE HYPERGEOMETRIC DISTRIBUTION 57
value of p = 0.2 is true, the probability they draw the king of spades 2 or fewer
times in 50 attempts is only 0.00088.
5.6 The Hypergeometric Distribution
18. Let X represent the number of white balls drawn. Then X has a hypergeometric
distribution with a = 22, N = 40, and n = 6.
�22�18
(a) P (X = 4) = 4
�402 = 0.2916.
6 �22�18 �22�18
(b) P (X ≤ 1) = P (X = 0) + P (X = 1) = 0
�40 6
+ 1
�405 = 0.00484 +
6 6
0.04911 = 0.05395.
22
(c) The mean number of white balls is 6 · = 3.3. The mean number of red balls
40
18
is 6 · = 2.7.
40
19. Here we are sampling without replacement (each person can be hired only once).
Let X represent the number of hires that have a criminal record. Then X has a
hypergeometric distribution with a = 7, N = 20, and n = 5.
(a) Here we must draw 3 criminals (from 7 total) and 2 non-criminals (from 13
total): �7�13
3
P (X = 3) = �202 = 0.1761.
5
(b) Here we must draw 2 criminals (from 7 total) and 3 non-criminals (from 13
total): �7�13
2
P (X = 2) = �203 = 0.3874.
5 �7�13
(c) P (X ≥ 1) = 1 − P (X = 0) = 1 − 0�205 = 1 − 0.0830 = 0.9170.
5
(d) µ = n · a
N =5· 7
20 = 1.75.
20. (a) The number of males in the sample has a hypergeometric distribution. The
binomial distribution can be used to approximate the hypergeometric, but
only if the sample size is a small proportion of the population size (less than
5% or so). Here the sample size is too large (we’re drawing 8 from 20), so the
binomial distribution would not provide a reasonable approximation. (Using
the hypergeometric distribution: P (X = 4) = 0.3501. Using the binomial
approximation: P (X = 4) ≈ 0.2734.)
Balka ISE 1.09 5.7. THE POISSON DISTRIBUTION 58
(b) The number of white balls follows a hypergeometric distribution, but since
we are drawing only a small proportion of the population (12 from 300), the
binomial distribution would provide a reasonable approximation. (Using the
hypergeometric distribution: P (X = 3) = 0.2141. Using the binomial approx-
imation: P (X = 4) ≈ 0.2120.)
(c) Here the number of kings has a binomial distribution with n = 60 and p = 13
1
,
so no approximation is required. P (X = 7) = 0.088.
5.7 The Poisson Distribution
5.7.1 Introduction
52 e−5
21. (a) P (X = 2) = 2! = 0.0842.
(b) 0.1247.
P (X ≤ 2) = P (X = 0) + P (X = 1) + P (X = 2)
50 e−5 51 e−5 52 e−5
= + +
0! 1! 2!
= 0.0067 + 0.0337 + 0.0842
= 0.1247
(c) P (X > 2) = 1 − P (X ≤ 2) = 1 = 0.1247 = 0.8753.

(d) P (2 < X ≤ 5) = P (X = 3) + P (X = 4) + P (X = 5) = 0.140 + 0.175 + 0.175 =
0.491.
P (X = 2 ∩ X ≤ 3) P (X = 2)
(e) P (X = 2|X ≤ 3) = = = 0.08422
0.2650 = 0.3178.
P (X ≤ 3) P (X ≤ 3)
(f) µ = 5, since µ = λ √
for a Poisson random variable.
√
(g) σ = 5, since σ = λ for a Poisson random variable.
41 e−4
22. (a) P (X = 1) = = 0.07326. (The number of shark attacks has a Poisson
1!
distribution with λ = 4.)
(b) P (X ≤ 2) = P (X = 0) + P (X = 1) + P (X = 2) = 0.2381. (Calculated with
the Poisson probability mass function with λ = 4.)
84 e−8
(c) P (X = 4) = = 0.05725. (The number of shark attacks in a two-year
4!
period has a Poisson distribution with λ = 2 × 4 = 8.)
23. Which of the following statements are true?
(a) A Poisson random variable can take on a countably infnite number of possible
values. True. (A Poisson random variable takes on the values 0, 1, 2, ...)
Balka ISE 1.09 5.7. THE POISSON DISTRIBUTION 59
(b) A Poisson random variable can take on negative values. False.

(c) A Poisson random variable represents a count of the number of occurrences of
an event.True.
(d) The mean and variance of a Poisson random variable are always equal. True.
(e) If X has a Poisson distribution, then P (X = 0) < P (X = 1). False. (Whether
or not P(X=0) is less than P(X=1) depends on the value of λ.)
(f) Every random variable that represents a count is a Poisson random variable.
False.
5.7.2 The Relationship Between the Poisson and Binomial Distri-

butions
24. Situation (d), where n = 500 and p = 0.01. The Poisson approximation to the
binomial works best when n is large and p is small.
25. (a) Using the binomial pmf: P (X = 2) = 100 2 98 = 0.18486.

�
2 0.01 · 0.99
For the Poisson approximation, λ = np = 100 · 0.01 = 1 and: P (X = 2) =
12 e−1
= 0.18394.
2!
(b) Binomial: P (X ≤ 2) = 0.92063. Poisson approximation: P (X ≤ 2) ≈
0.91970.
Using the binomial pmf:
P (X ≤ 2) = P (X = 0) + P (X = 1) + P (X = 2)

100 0 100 100 1 99 100
= 0.01 · 0.99 + 0.01 · 0.99 + 0.012 · 0.9998
0 1 2
= 0.36603 + 0.36973 + 0.18486
= 0.92063
Using the Poisson approximation:
P (X ≤ 2) = P (X = 0) + P (X = 1) + P (X = 2)
10 e−1 11 e−1 12 e−1
= + +
0! 1! 2!
= 0.36788 + 0.36788 + 0.18394
= 0.91970
(c) Binomial: σ 2 = np(1 − p) = 100 · 0.01 · 0.99 = 0.99.

Poisson approximation: σ 2 = λ = np = 100 · 0.01 = 1.
Balka ISE 1.09 5.8. THE GEOMETRIC DISTRIBUTION 60
5.7.3 Poisson or Not? More Discussion on When a Random Vari-

able has a Poisson distribution
26. The number of students would not follow a Poisson distribution, since students do
not arrive randomly and independently. Students are likely to travel in groups, and
thus there would be "clumping" of students. It is likely that there would also be
time e˙ects (for example, the theoretical rate at which students are arriving might
be very di˙erent at 12:01 and 12:29).
27. The number of cows in a randomly selected 100 m2 area of the pasture would not
follow a Poisson distribution, since the cows would not be distributed randomly and
independently (cows tend to stick together).
5.8 The Geometric Distribution
28. Let X be the number of donors required to get the frst one with type AB blood.
(a) P (X = 5) = (1 − 0.03)5−1 · 0.03 = 0.0266.

(b) 0.0873. This answer can be found in 2 ways. One way is brute force method:
P (X ≤ 3) = P (X = 1) + P (X = 2) + P (X = 3)
= 0.970 · 0.03 + 0.971 · 0.03 + 0.972 · 0.03
= 0.0873
Another way is to use the cumulative distribution function:
P (X ≤ x) = 1 − (1 − p)x
=⇒ P (X ≤ 3) = 1 − (1 − 0.03)3
= 0.0873
(c) 0.401. This could be answered using the brute force method:
P (X > 30) = 1 − P (X ≤ 30) = 1 − [P (X = 1) + P (X = 2) + . . . + P (X = 30)].
However, this would require calculating 30 geometric probabilities which would
be very cumbersome without using software. It is easier to answer this question
if we recognize that the frst success will come after the 30th person if the frst
30 people do not have blood type AB. So P (X > 30) = 0.9730 = 0.401.
(d) µ = p1 = 0.031
= 33.3̇.
q q
(e) σ = 1−p p 2 = 1−0.03
0.032
= 32.83.
29. Let X represent the required number of cups to get the frst winner.
Balka ISE 1.09 5.9. THE NEGATIVE BINOMIAL DISTRIBUTION 61
(a) P (X = 4) = ( 56 )4−1 ( 16 ) = 0.0965.

(b) 0.5177. This answer can be found in 2 ways. One way is brute force method:
P (X ≤ 4) = P (X = 1) + P (X = 2) + P (X = 3) + P (X = 4)
5 1 5 1 5 1 5 1
= ( )0 ( ) + ( )1 ( ) + ( )2 ( ) + ( )3 ( )
6 6 6 6 6 6 6 6
= 0.5177
Another way is to use the cumulative distribution function:
P (X ≤ x) = 1 − (1 − p)x
5
=⇒ P (X ≤ 4) = 1 − ( )4
6
= 0.5177.
(c) 0.0260. This could be answered using the brute force method:
P (X > 20) = 1 − P (X ≤ 20) = 1 − [P (X = 1) + P (X = 2) + . . . + P (X = 20)].
However, this would require calculating 20 geometric probabilities which would
be very cumbersome without using software. It is easier to answer this question
if we recognize that the frst winner will come after the 20th cup if the frst 20
cups have no winners. So P (X > 20) = ( 56 )20 = 0.0260.
(d) µ = p1 = 1/6
1
= 6.
30. (a) The mean of the geometric distribution is p1 . True.

(b) The most likely value of a geometric random variable is always 1. True. The
most likely value is always 1, the next most likely value is always 2, and so on.
(c) If p > 0.5, then P (X = 2) < P (X = 3). False. Geometric probabilities are
always decreasing in x for x = 1, 2, . . ., so P (X = 2) > P (X = 3).
(d) The variance of the geometric distribution is always 1. False. σ 2 = 1− p2
p
.
(e) A geometric random variable can take on a countably infnite number of pos-
sible values. True. X takes on the values 1, 2, . . .
(f) If X has a geometric distribution, then Y = X1 also has a geometric distribu-
tion. False. The possible values of Y would be 1, 12 , 13 , . . . . These are not the
values that a geometric random variable takes on.
5.9 The Negative Binomial Distribution
31. Let X represent the number of attempts required to chip-in 4 times. X has the
negative binomial distribution with parameters r = 4 and p = 0.05. The�probability
mass function of the negative binomial distribution is: P (X = x) = x−1
r
r−1 p (1 −
p)x−r .
Balka ISE 1.09 5.10. THE MULTINOMIAL DISTRIBUTION 62
(a) P (X = 20) =
�20−1
4−1 0.054 (1 − 0.05)16 = 0.00267.
(b) P (X = 20) + P (X = 21) = 20−1
� 4 16
�21−1 4 17 =
4−1 0.05 (1 − 0.05) + 4−1 0.05 (1 − 0.05)
0.00267 + 0.00298 = 0.00564.
(c) µ = pr = 0.05
4
= 80.
q q
(d) σ = r(1−p)
p2
= 4·0.95
0.052
= 38.99.
(e) It will take more than 100 attempts to chip-in 4 times if the frst 100 attempts
have fewer than 4 chip-ins. Let Y represent the number of chip-ins in the frst
100 attempts. Then we need to fnd P (Y < 4), where Y has the binomial
distribution with n = 100 and p = 0.05.
P (Y < 4) = P (Y = 0) + P (Y = 1) + P (Y = 2) + P (Y = 3)
= 0.0059 + 0.0312 + 0.0812 + 0.1396
= 0.2578
This implies that P (X > 100) = 0.2578.
32. Let X represent the number of doors required to make 3 sales. Then X has
the negative binomial distribution with parameters r = 3 and p = 0.015. The
probability
�x−1 r mass function of the negative binomial distribution is: P (X = x) =
x−r .
r−1 p (1 − p)
(a) P (X = 50) =
�50−1
0.0153 (1 − 0.015)47 = 0.00195.
�3−1
(b) P (X = 200) = 200−1 3 197 = 0.00339.

3−1 0.015 (1 − 0.015)
(c) µ = pr = 0.015
3
= 200.
q q
(d) σ = r(1−p)
p2
= 3·0.985
0.0152
= 114.60.
(e) It will take more than 500 attempts to make 3 sales if the frst 500 attempts
have fewer than 3 sales. Let Y represent the number of sales in the frst
500 attempts. Then we need to fnd P (Y < 3), where Y has the binomial
P (Y < 3) = P (Y = 0) + P (Y = 1) + P (Y = 2)
= 0.00052 + 0.00398 + 0.01512
= 0.0196
5.10 The Multinomial Distribution
33. Let X1 , X2 , and X3 be the number of people with arches, loops, and whorls, respec-
tively. Then X1 , X2 , and X3 have the multinomial distribution with parameters
n = 10 and p1 = 0.12, p2 = 0.49, and p3 = 0.39.
10!
(a) P (X1 = 2, X2 = 3, X3 = 5) = 0.122 0.493 0.395 = 0.0385.
2!3!5!
10!
(b) P (X1 = 4, X2 = 4, X3 = 2) = 0.124 0.494 0.392 = 0.0057.
4!4!2!
(c) Since we are counting the number with arches, we will label arches a success
and non-arches (loops or whorls) a failure. The number with arches (X1 ) will
have a binomial
�10distribution with n = 10 and p = 0.12.
P (X1 = 1) = 1 0.12 (1 − 0.12)9 = 0.3798.
1

(d) The number with arches has a binomial distribution with parameters n and p,
so E(X1 ) = np = 10 × 0.12 = 1.2.
34. Let X1 , X2 , and X3 be the number of times the ball lands in red, black, and green
slots, respectively. Then X1 , X2 , and X3 have the multinomial distribution with
parameters n = 20 and p1 = 18 37 , p2 = 37 , and p3 = 37 .
18 1
20! 18 8 18 10 1 2
(a) P (X1 = 8, X2 = 10, X3 = 2) = ( ) ( ) ( 37 ) = 0.0141.
8!10!2! 37 37
20! 18 14 18 3 1 3
(b) P (X1 = 14, X2 = 3, X3 = 3) = ( ) ( 37 ) ( 37 ) = 0.000073.
14!3!3! 37
(c) Since we are counting the number of times the ball lands in a red slot, we
will label the red slots a success and non-red slots (green or black) a failure.
The number of times the ball lands in a red slot (X1 ) will have a binomial
distribution with n = 20 and p = 18 37 .
�20 18 1 18 9
P (X1 = 11) = 11 ( 37 ) (1 − 37 ) = 0.1506.
(d) The number of times the ball lands in a red slot has binomial distribution with
parameters n = 20 and p = 18 37 , so E(X1 ) = np = 20 × 37 = 37 ≈ 9.73.
18 360
35. All of the statements are true.
36. (a) c = 0.05, since the probabilities must sum to 1. So the distribution is:
x −5 10 50 100
p(x) 0.05 0.05 0.40 0.50
(b) P (X > 21) = P (X = 50) + P (X = 100) = 0.40 + 0.50 = 0.90.
P (X > 21 ∩ X < 60) P (X = 50) 0.40
(c) P (X > 21|X < 60) = = = =
P (X < 60) P (X < 60) 0.05 + 0.05 + 0.40
0.80.
37. (a) 115. The missing probability must be 0.4. E(X) = 10 · 0.1 + 20 · 0.2 + 30 ·
0.3 + x · 0.4 = 60 =⇒ x = 115.
38. (a) 0.1 (since the probabilities must sum to 1).

(b) P (X > 11) = P (X = 12) + P (X = 13) = 0.2 + 0.1 = 0.3.
(c)
P
E(X) = xp(x) = 10 · 0.4 + 11 · 0.3 + 12 · 0.2 + 13 · 0.1 = 11.
(d) σ 2 = 1. E(X 2 ) =
P 2
x p(x) = 102 · 0.4 + 112 · 0.3 + 122 · 0.2 + 132 · 0.1 = 122.
σ = E(X ) − [E(X)]2 = 122 − 112 = 1.
2 2
39. (a) 0.2304. Let X represent the number of 10s selected. �5 Then X has a binomial
distribution with n = 5 and p = 0.4. P (X = 3) = 3 0.4 0.62 = 0.2304.
3
(b) P (X ≥ 1) = 1 − P (X = 0) = 1 − 0.65 = 0.92224.

(c) This will occur if there are no 11s in the frst six values, which has a probability
of (1 − 0.3)6 = 0.1176.
40. Let X be the number of 10s selected. Then X has a binomial distribution with
n = 20 and p = 0.4.
(a) P (X < 8) = 0.416. (This requires summing 8 binomial probabilities, so it is

best carried out using software.)
(b) P (2 ≤ X ≤ 6) = 0.249.
(c) E(X) = np = 20 · 0.40 = 8.
41. Both X and Y are binomial random variables. Binomial random variables have a
mean of np and a variance of np(1 − p).
(a) E(X + Y ) = E(X) + E(Y ) = 20 × 0.50 + 18 × 1

6 = 13.
q
(b) σX
2 = 20 × 0.5 × (1 − 0.5) = 5, σ 2 = 18 ×
Y
1
6 × 5
6 = 2.5. σX−Y = 2 + σ2 =
σX Y
√
5 + 2.5 ≈ 2.739.
42. (a) P (X > 1) = P (X = 10) + P (X = 100) = 0.01 + 0.001 = 0.011.

(b) The only way to get a total proft of $100 is to have one $100 ticket and one
$0 ticket. The probability of this is 0.689 · 0.001 + 0.001 · 0.689 = 0.001378.
(c) E(X) = xp(x) = 0 · 0.689 + 1 · 0.30 + 10 · 0.010 + 100 · 0.001 = 0.5.
P
(d) σ 2 = (x − µ)2 p(x) = (0 − 0.5)2 ·√0.689 + (1 − 0.5)2 · 0.30 + (10 − 0.5)2 · 0.010 +
P
2
(100 − 0.5) · 0.001 = 11.05. σ = 11.05 = 3.324.
(e) P (X = 100|X > 0) = P (X=1)+PP(X=10)+P
(X=100) 0.001
(X=100) = 0.3+0.01+0.001 = 0.003215.
(f) 1000 · 0.5 = $500.
5.11.2 Concepts
43.
E[(X − µ)2 ] = E(X 2 − 2Xµ + µ2 )

= E(X 2 ) − E(2Xµ) + E(µ2 )
= E(X 2 ) − E(2Xµ) + E(µ2 )
= E(X 2 ) − 2µE(X) − µ2 (Since µ is a constant)
2 2 2
= E(X ) − 2µ + µ
= E(X 2 ) − µ2
= E(X 2 ) − [E(X)]2
44. No, that statement is false. A variable is not independent of itself.
45. All of the statements are false.
46. (a) E(X) = 15 · 0.2 = 3, E(Y ) = 6, and E(Z) = 5.

E(X + Y + Z) = E(X) + E(Y ) + E(Z) = 3 + 6 + 5 = 14.
(b) σX
2 = np(1 − p) = 15 · 0.2 · 0.8 = 2.4, σ 2 = 6, σ 2 = 182 = 324.
Y Z
q
2 + σ2 + σ2 =
√ √
σX+Y −Z = σX Y Z 2.4 + 6 + 324 = 332.4 = 18.232.
(c) Yes. X is a binomial random variable and cannot take on negative values, and
Y is a Poisson random variable and cannot take on negative values. But Z has
an unknown distribution, so it could possibly take on negative values. Since
X + Y could be 0, X + Y − Z could possibly take on negative values.
47. (e) (None of the above). The number of children in the selected area will be 0 or
1 (since the children spread themselves out). The number of children in the area
would have a Bernoulli distribution, but the precise value of p is a little tricky to
determine.
48. (c). If the trials were independent, then X would have a binomial distribution.
However, in this scenario the trials are not independent, since if Tom makes a
mistake then he is more likely to make more mistakes. X is a count variable with
a minimum of 0 and a maximum of 20, and it is thus a discrete random variable,
but its exact distribution is impossible to determine without further information.
49. (a) A discrete random variable can take on a countable number of possible values.
True.
(b) A discrete random variable always represents a count. False.
(c) The expected value of a discrete random variable must equal one of the vari-
able’s possible values. False.
(d) The binomial distribution has two parameters: n and p. True.

(e) A binomial random variable can take on one of n + 1 possible values. True.
(A binomial random variable takes on the values 0, 1, 2, . . . , n.)
50. (a) The variance and standard deviation of a discrete random variable cannot be
equal. False. (They may both equal 1.)
(b) The mean of a random variable cannot be negative. False.
(c) The standard deviation of a random variable cannot be negative. True.
(d) If X is a discrete random variable, then Y = 2X + 3 is a discrete random
variable. True.
51. (a) In a binomial setting, the probability of success usually changes from trial to
trial. False. (In a binomial setting, the probability of success stays constant
from trial to trial.)
(b) The mean of a Poisson random variable can be negative. False. (µ = λ must
be positive.)
(c) If X represents the number of students in a randomly selected 100 m2 area of
a university campus at 3:00 pm on a Tuesday afternoon, then X has a Poisson
distribution. False. (Students are not distributed randomly—they tend to group
together.)
(d) If p > 0, then the variance of a binomial random variable is less than its mean.
True.
(e) If X is a binomial random variable with n = 10 and p = 0.1, and Y is a
Poisson random variable with λ = 2, then X − Y has a binomial distribution.
False. (X −Y can take on negative values so it cannot possibly have a binomial
distribution.)
5.11.3 Applications
52. (a) E(X) = 1 · 0.38 + 2 · 0.52 + 3 · 0.05 + 4 · 0.05 = 1.77.

(b) σ 2 = (1 − 1.77)2 2 2 2
√ · 0.38 + (2 − 1.77) · 0.52 + (3 − 1.77) · 0.05 + (4 − 1.77) · 0.05 =
0.5771. σ = 0.5771 = 0.7597.
(c) E(ln(X)) = ln(1) · 0.38 + ln(2) · 0.52 + ln(3) · 0.05 + ln(4) · 0.05 = 0.48468.
(d) E(X 2 ) = 12 · 0.38 + 22 · 0.52 + 32 · 0.05 + 42 · 0.05 = 3.71.
53. (a) Let Y represent the number of randomly selected numbers that match the
ticket. Then Y has a hypergeometric distribution with a = 6, N = 70, and
n = 20.
6 64 6 64
� � � �
5
P (Y = 5) = �7015
= 0.0059123. P (Y = 6) = 6
�7014
= 0.0002956.
20 20
P (Y ≤ 4) = 1 − [P (Y = 5) + P (Y = 6)] = 1 − (0.0059123 + 0.0002956) =
0.9937921.
(b) Let X represent

P the payout on the ticket.
E(X) = xp(x) = 0 · 0.9937921 + 25 · 0.0059123 + 1000 · 0.0002956 = 0.4434.
(c) E(X 2 ) =
P 2
x p(x) = 02 · 0.9937921 + 252 · 0.0059123 + 10002 · 0.0002956 =
299.2952.
√ σ = E(X 2 ) − [E(X)]2 = 299.2952 − 0.44342 = 299.0986. σ =
2
299.0986 = 17.29.
(d) A player is paying $1 for the ticket and getting back $0.4434 on average, so
the government is keeping about 56 cents on average.
54. (a) Let X represent the number of tickets that win $25. Then X has a binomial
distribution with
�50 parameters n = 50 and p = 0.0059123.
P (X = 1) = 1 0.0059123 (1 − 0.0059123)49 = 0.2211.
1
(b) P (X ≥ 1) = 1 − P (X = 0) = 1 − 50 0 50 = 1 −
�
0 0.0059123 (1 − 0.0059123)
0.7434 = 0.2566.
(c) Let W represent the number of tickets that win a prize (either $25 or $1000).
Then W has a binomial distribution with parameters n = 50 and� p= 0.0059123+
0.0002956 = 0.0062079. P (W ≥ 1) = 1 − P (W = 0) = 1 − 50 0
0 0.0062079 ·
0.993792150 = 1 − 0.993792150 = 1 − 0.7324 = 0.2676.
(d) E(W ) p= np = 50 · 0.0062079 = 0.3104.
(e) σW = np(1 − p) = 50 · 0.0062079(1 − 0.0062079) = 0.5554.
p
55. (a) 5000 × 0.98 + 100000 × 0.02 = 6900.

(b) Taking the gamble saves you $100 on average.
(c) We are often willing to accept a reduced expectation in order to reduce the
variability. (For example, when we purchase insurance.) Here, perhaps losing
the wager would put you in a bad fnancial position, which is likely something
you want to avoid. (And there is always the possibility that the salesman is
running some sort of con game on you. It is, of course, probably best to avoid
propositions like this.)
56. (a) 1000 − 25000 · 0.01 = $750.

(b) Let X represent the insurance company’s payout. The distribution of X is:
x 0 25000
p(x) 0.99 0.01
E(X) = 0 · 0.99 + 25000 · 0.01 = 250

E(X 2 ) = 02 · 0.99 + 250002 · 0.01 = 6250000
σ 2 =√E(X 2 ) − [E(X)]2 ) = 6250000 − 2502 = 6187500.
σ = 6187500 = 2487.469.
(c) 1000 − 25000 · p = 0 =⇒ p = 25000 1000
= 0.04.
57. (a) The number of siblings that are an HLA-identical match has a binomial dis-
tribution with n = 6 and p = 14 .
(b) Let X represent the number of siblings that are an HLA-identical match.
P (X ≥ 1) = 1 − P (X = 0) = 1 − ( 34 )6 = 0.822.
(c) P (X ≥ 2) = 1 − [P (X = 0) + P (X = 1)] = 1 − (0.1780 + 0.3560) = 0.466.
(d) µ = np = 6 × 14 = 1.5.
58. Let X represent the number of completed surveys. Then X has a binomial distri-
bution with parameters n = 100 and p = 0.09.
(a) P (X = 8) = 100 8 92 = 0.1366.

�
8 0.09 0.91
(b) 0.399. P (X = 8)+P (X = 9)+P (X = 10) = 100 0.098 0.9192 + 100 9 91
� �
�100 8 9 0.09 0.91 +
10 90 = 0.1366 + 0.1381 + 0.1243 = 0.399.
10 0.09 0.91
(c) P (X ≥ 2) = 1 − [P (X = 0) + P (X = 1)] = 1 − (0.00008 + 0.00079) = 0.99913.
(d) µ = np
p = 100 · 0.09√= 9.
(e) σ = np(1 − p) = 100 · 0.09 · 0.91 = 2.862.
59. Let X represent the number of calls needed to get 4 completed surveys. Then X
has the negative binomial distribution with parameters r = 4 and p = 0.09. The
probability
�x−1 r mass function of the negative binomial distribution is: P (X = x) =
r−1 p (1 − p)x−r .
(a) P (X = 50) =
�50−1
4−1 0.094 (1 − 0.09)46 = 0.0158.
(b) P (X = 50) + P (X = 51) = 50−1
� 4 46
�51−1 4 47 =
4−1 0.09 (1 − 0.09) + 4−1 0.09 (1 − 0.09)
0.0158 + 0.0153 = 0.0311.
(c) µ = pr = 0.09
4
= 44.4.
q q
(d) σ = r(1−p)
p 2 = 4·0.91
0.092
= 21.20.
(e) It will take more than 100 calls to get 4 completed surveys if the frst 100
calls result in fewer than 4 completed surveys. Let Y represent the number
of completed surveys in the frst 100 calls. Then we need to fnd P (Y < 4),
where Y has the binomial distribution with n = 100 and p = 0.09.
P (Y < 4) = P (Y = 0) + P (Y = 1) + P (Y = 2) + P (Y = 3)
= 0.00008 + 0.00079 + 0.00388 + 0.01254
= 0.0173
60. Let X represent the number of chickens that have detectable levels of salmonella.
Then, assuming we have a random and independent sample of chickens, X has a
binomial distribution with parameters n = 12 and p = 0.14.
(a) P (X = 2) = 12 2 10 = 0.286.
�
2 0.14 0.86
(b) 0.7697.
P (X ≤ 2) = P (X = 0) + P (X = 1) + P (X = 2)

12 0 12 12 1 11 12
= 0.14 0.86 + 0.14 0.86 + 0.142 0.8610
0 1 2
= 0.1637 + 0.3197 + 0.2863
= 0.7697
(c) Let Y represent the number of chickens with detectable levels of salmonella.
Assuming this retailer’s claim is true, Y has a binomial distribution with
n = 20 and p = 0.02. P (Y ≥ 3) = 1 − P (X ≤ 2) = 1 − [P (Y = 0) + P (Y =
1) + P (Y = 2)] = 0.007. This yields strong evidence against the company’s
claim.
61. Let X represent the number of people in the sample that have given blood at some
point.
(a) X has a binomial distribution with n = 5 and p = 0.10.

(b) P (X = 1) = 0.328.
(c) P (X ≥ 2) = P (X = 2) + P (X = 3) + P (X = 4) + P (X = 5) = 0.0815.
(d) P (X ≤ 2) = P (X = 0) + P (X = 1) + P (X = 2) = 0.9914.
(e) P (X = 1|X ≥ 1) = P (X=1∩X≥1)
P (X≥1) = PP (X=1) 0.32805
(X≥1) = 0.40951 = 0.8011.
62. (a) The number of tickets that win a cash prize follows a binomial distribution
with n = 156 and p = 0.03.
(b) E(X) = np = 156 · 0.03 = 4.68.
(c) P (X = 0) = 0.97156 = 0.0086.
(d) This is a complicated question, without an easy answer, and there are many
factors to take into account (for example, how trustworthy you believe your
friend is). But if your friend was indeed purchasing the tickets and (theoreti-
cally) sharing the cash prizes, there is a less than 1% chance of not winning a
prize. You may not want to end a friendship over this, but it is a little fshy.
63. Let X represent the number of crashes in the next year. Since the fatal crashes can
be thought of as occurring randomly and independently with a mean of 1.1, X has
(approximately) a Poisson distribution with λ = 1.1.
1.10 e−1.1
(a) P (X = 0) = 0! = 0.3329.
2 −1.1 3 −1.1
(b) P (X = 2) + P (X = 3) = 1.1 2!
e
+ 1.1 3!
e
= 0.2014 + 0.0738 = 0.2752.
(c) P (X√> 4) = 1 − P (X ≤ 4) = 0.0054.
√
(d) σ = λ = 1.1 = 1.049.
64. Let X represent the number of crashes in the next 2 year period. Since the fatal
crashes can be thought of as occurring randomly and independently with a mean of

1.1 per year, X has (approximately) a Poisson distribution with λ = 2 × 1.1 = 2.2.
0 −2.2
(a) P (X ≥ 1) = 1 − P (X = 0) = 1 − 2.2 0!
e
= 1 − 0.1108 = 0.8892.
(b) P (X > 1) = 1 − P (X ≤ 1) = 1 − [P (X = 0) + P (X = 1)] = 1 − (0.1108 +
0.2438) = 0.6454.
(c) µ = λ
√= 2.2√.
(d) σ = λ = 2.2 = 1.48.
65. (a) If the mean number of impacts per million years is 2, then the mean number
of impacts in a 100,000 year period is µ = λ = 10 2
= 0.20.
0
0.2 e −0.2
P (X ≥ 1) = 1 − P (X = 0) = 1 − 0! = 1 − 0.8187 = 0.1813.
(b) If the mean number of impacts per million years is 2, then the mean number
of impacts in a 10,000 year period is µ = λ = 100 2
= 0.02.
0
0.02 e −0.02
P (X ≥ 1) = 1 − P (X = 0) = 1 − 0! = 1 − 0.9802 = 0.0198.
66. (a) The number of round/yellow plants has a binomial distribution with n = 50
and p = 16
9
.
(b) µ = np = 50 · 16
9
= 28.125.
q
(c) σ = np(1 − p) = 50( 16 9 9
p
)(1 − 16 ) = 3.5078.
(d) The number of wrinkled/green � 1plants has a binomial distribution with n = 50
and p = 16 1
. P (X = 4) = 50 4 ( 16 ) 4 ( 15 )46 = 0.1805.
16
(e) 28!12!6!4!
50!
) ( 16 ) ( 16 ) ( 16 ) = 0.00152. (The distribution of the number of
9 28 3 12 3 6 1 4
( 16
plants with the 4 phenotypes is the multinomial distribution with n = 50,
p1 = 169
, p2 = 16 3
, p3 = 163
, and p4 = 16 1
.)
5.11.4 Extra Practice Questions
67. (a) µ = np =⇒ 2.4 = 10 · p =⇒ p = 0.24.

(b) P (Xp= 2) = 0.2885.p
√
(c) σ = np(1 − p) = 10 · 0.24 · (1 − 0.24) = 1.824 = 1.350.
68. X is a binomial random variable with n = 2, p = 23 .
(a) µ = np = 2 · 23 = 43 .
q
(b) σ = np(1 − p) = 2 · 2 1
= 23 .
p
3 · 3
69. (a) The number of people in the sample that are over 60 years old will follow
a binomial distribution with n = 20 and p = 0.178. Let X represent this
number.
(b) P (X ≥ 2) = 0.89426.
(c) P (X = 2) = 0.1767.
(d) P (X = 0) = (1 − 0.178)20 = 0.01984.
(e) E(X) = np = 20 · 0.178 = 3.56.
70. (a) The number of people in the sample that voted in the last federal election
has a binomial distribution with n = 5 and p = 0.60. Let X represent this
number.
(b) P (X = 4) = 0.2592.
(c) P (X ≥ 4) = P (X = 4) + P (X = 5) = 0.3370.
(d) P (X = 5|X ≥ 4) = P (X=5∩X≥4)
P (X≥4) = PP (X=5) 0.07776
(X≥4) = 0.33696 = 0.2308.
(e) E(X) = np = 5 · 0.60 = 3.0.
71. Let X be the number of people that are out of work. Then X has a binomial
(a) P (X = 3) = 0.0670.
(b) P (X > 2) = 1 − P (X ≤ 2) = 1 − [P (X = 0) + P (X = 1) + P (X = 2)] =
0.08496.
(c) 1. (If there are at least 2 out of work, there must be at least 1 out of work.)
(d) 0. (If there are at least 3 out of work, then there is more than 1 out of work.)
72. The number of cars that fail the test will have a binomial distribution with n = 10
and p = 0.35.
(a) P (X = 2) = 0.17565.
(b) P (X ≤ 2) = P (X = 0) + P (X = 1) + P (X = 2) = 0.2616.
(c) P (X > 2) = 1 − P (X ≤ 2) = 1 − 0.2616 = 0.7384.
(d) E(X) p= np = 10 · 0.35 = 3.5.
√
(e) σ = np(1 − p) = 10 · 0.35 · 0.65 = 1.508.
(f) Not in and of itself, but something is going on. If this was a random selection
of 10 cars that are at least 13 years old, then the probability of failing all of
them should be approximately 0.3510 = 0.000028. This is quite unlikely, so
there is a concern that the garage might be inappropriately failing cars for
some reason. But there are other possible explanations. For example, perhaps
this garage has an excellent reputation for fxing cars that are almost ready
for the junk heap, so they might tend to test cars that are in worse shape.
73. (a) The number of decays in a one-second period is assumed to follow a Poisson
0.82 e−0.8
distribution with λ = 0.8, so: P (X = 2) = = 0.1438.
2!
(b) The number of events in a three-second period is assumed to follow a Poisson
2.42 e−2.4
distribution with λ = 3 × 0.8 = 2.4, so: P (X = 2) = = 0.2613.
2!
2.40 e−2.4 2.41 e−2.4

(c) P (X ≤ 2) = P (X = 0) + P (X = 1) + P (X = 2) = + +
0! 1!
2.42 e−2.4
= 0.5697.
2!
74. Since the events are occurring randomly and independently of one another, it is
reasonable to assume that the number of events in an 8 hour period follows a
Poisson distribution with λ = 8 × 0.2 = 1.6.
(a) Let X represent the number of log jams in an 8 hour period. P (X = 3) =

0.1378.
(b) P (X ≥ 1) = 1 − P (X = 0) = 1 − 0.2019 = 0.7981.
(c) 0.1582. Let X represent the number of log jams in an hour. The probability
there is at least one log jam is P (X ≥ 1) = 1−P (X = 0) = 1−0.8187 = 0.1813
(calculated using the Poisson pmf with λ = 0.2).
Let Y represent the number of hours in the 12 randomly selected hours that
have at least one log jam. Then Y has a binomial distribution with n = 12 and
p = 0.1813. P (Y ≥ 4) = 1 − P (Y ≤ 3) = 1 − [P (Y = 0) + P (Y = 1) + P (Y =
2) + P (Y = 3)] = 1 − 0.8418 = 0.1582.
75. (a) 0.999955. Let X represent the number of planes that arrive in the two hour
period. P (X ≥ 1) = 1 − P (X = 0) = 1 − 0.000045 = 0.999955 (calculated
using the Poisson pmf with λ = 2 × 5 = 10).
(b) 0.00017. Let X represent the number of planes that arrive in a 3 hour period.
P (X = 3) = 0.00017 (calculated using the Poisson pmf with λ = 3 × 5 = 15).
(c) 0.0821. Let X represent the number of planes that arrive in a half-hour period.
P (X = 0) = 0.0821 (calculated using the Poisson pmf with√ λ = 0.5 × 5 = 2.5).
√
(d) 8 × 5 = 6.325 (since for a Poisson random variable, σ = λ).
76. (a) Under the given assumptions, the number that crash has a binomial distribu-
tion with n = 10 and p = 0.01. For the remaining questions, let X represent
the number of planes that crash.
(b) P (X ≥ 1) = 1 − P (X = 0) = 1 − 0.90438 = 0.0956 (calculated using the
binomial pmf with n = 10 and p = 0.01).
(c) P (X = 1) = 0.09135 (calculated using the binomial pmf with n = 10 and
p = 0.01).
(d) P (Xp= 10) = 10−20p= 0.00000000000000000001.
(e) σ = np(1 − p) = 10 · 0.01 · (1 − 0.01) = 0.3146.
77. (a) The number that need to be selected in order to get the frst doctor has a
geometric distribution with p = 0.002. Let X represent this number. P (X =
4) = (1 − 0.002)3 × 0.002 = 0.001988.
(b) If the frst 1000 people are not doctors, then the frst doctor will occur after
the frst 1000 people sampled. P (X > 1000) = (1 − 0.002)1000 = 0.1351.
(c) The number of doctors in a random sample of 80 Canadians has (approxi-

mately) a binomial distribution with n = 80 and p = 0.002. Let Y represent
this number. P (Y = 1) = 0.13659.
(d) P (Y > 2) = 1 − P (Y ≤ 2) = 1 − [P (Y = 0) + P (Y = 1) + P (Y = 2)] =
1 − 0.9994 = 0.0005858 (calculated using the binomial pmf with n = 80 and
p = 0.002).
78. (a) The number of welds needed to get the frst unsuccessful weld has a geometric
distribution with p = 0.10. For the following questions, let X represent this
random variable.
(b) P (X = 3) = 0.92 · 0.1 = 0.081.
(c)
P (X ≤ 3) = P (X = 1) + P (X = 2) + P (X = 3)
= 0.90 · 0.1 + 0.91 · 0.1 + 0.92 · 0.1
= 0.271
(d) The frst unsuccessful weld will occur after the 13th trial if the frst 13 trials
are successful. So P (X > 13) = 0.913 = 0.254.
79. (a) The number of welds needed to get the third unsuccessful weld has a negative
binomial distribution with r = 3 and p = 0.10. Let X represent this random
variable.
P (X = 22) = 21
� 3 19
2 0.1 0.9 = 0.02837
(b) Let Y represent the number of unsuccessful welds. Then Y has a binomial
distribution with n = 20 and p = 0.10. P (X = 3) = 0.1901.
(c) P (Y ≥ 3) = 1 − P (Y ≤ 2) = 1 − [P (Y = 0) + P (Y = 1) + P (Y = 2)] =
1 − 0.6769 = 0.3231.
(d) µ = np = 20 × 0.1 = 2.
80. (a) Let X represent the number of shipments needed to get the frst rejected
shipment. Then X has a geometric distribution with p = 0.1. P (X = 10) =
0.969 · 0.04 = 0.0277.
(b) The frst rejected shipment will occur after the 10th shipment if the frst 10
are not rejected. P (X > 10) = 0.9610 = 0.6648.
(c) Let Y represent the number of rejected shipments. Then Y has a binomial
distribution with n = 60 and p = 0.04. P (Y ≤ 1) = P (Y = 0) + P (Y = 1) =
0.0864 + 0.2159 = 0.3022.
81. (a) The number of men in the sample follows a hypergeometric distribution. To
get exactly 4�men,
� we must choose 4 men from 55, and 1 woman from 30:
55 30
4 1
P (X = 4) = �85 = 0.3119.
5
(b) To approximate the hypergeometric distribution in this example, we can use

the binomial distribution with n = 5 and p = 55+30 55 11
= 17 . Using the binomial
pmf: P (X = 4) ≈ 0.3093. Note that this is not too far from the correct
probability of 0.3119 (found in 81a), since we are sampling only a small fraction
of the population.
82. Let X represent the number of dandelions in the randomly selected area. Then X
has a Poisson distribution with λ = 4.
(a) P (X = 2) = 0.1465.
(b) P (X > 2) = 1 − P (X ≤ 2) = 1 − [P (X = 0) + P (X = 1) + P (X = 2)] =
1 − 0.2381 = 0.7619.
(c) P (X = 3|X ≥ 1) = P (X=3∩X≥1)
P (X≥1) = PP (X=3) 0.1953668
(X≥1) = 0.9816844 = 0.1990.
√ √
(d) σ = λ = 4 = 2.
83. (a) Let X represent the number of implants needed to get the frst unsuccessful
one. Then X has a geometric distribution with p = 0.05. P (X = 4) =
0.953 · 0.05 = 0.0429.
(b) In order for the frst unsuccessful implant to occur after the 11th trial, we need
the frst 11 trials to be successful. P (X > 11) = 0.9511 = 0.5688.
(c) Let Y represent the number of trials needed to get the third unsuccessful
implant. Then Y has �the negative binomial distribution with r = 3 and
p = 0.05. P (Y = 28) = 27 3
2 0.05 0.95
25 = 0.0122.
84. (a) Here we need to pick 2 footballs and 4 balls that are either soccer balls or
volleyballs: �15�28
P (2 footballs) = 2
�434 = 0.3526.
6
(b) µ = n · a
N =6· 15
43 = 2.093.
�20�23 �20�23
(c) P (0 soccer balls)+P (1 soccer ball) = 0
�436 + 1�435 = 0.01656+0.11039 =
6 6
0.1269.
�15�20�8
(d) 3
�432 1
= 0.1134.
6
85. (a) Since the sampling is done without replacement, the number of computer en-
gineers in the group of 6 will follow a hypergeometric distribution. We must
select 3 computer engineers (from 7 total)
� � and 3 non-computer engineers (from
7 13
5 + 5 + 3 = 13 total). P (X = 3) = 3
�203 = 0.2583.
6
(b) Here we must select 2 biomedical engineers (from 5), 2 computer engineers
(from 7), 1 mechanical engineer (from 5) and 1 water resource engineer (from
�5�7�5�3
3). The probability of this is: 2 2
�201 1
= 0.0812.
6
86. Here we need to pick 1 Honda vehicle (from 4), 1 Ford (from 6), 1 Toyota (from 3),
and 2 Chrysler (from 7). We can calculate the�probability of this event using the
4 6 3 7
� � �
multivariate hypergeometric distribution: 1 1
�20 1 2
= 0.09752.
5
87. 5! 3 1 1
3!1!1!0! 0.4 0.3 0.2 0.1
0 = 0.0768.
Chapter 6
Continuous Random Variables and

Continuous Probability
Distributions
J.B.’s strongly suggested exercises: 1, 5, 6, 8, 9, 10, 11, 13, 14, 15, 21, 23, 24, 27
6.1 Introduction
6.2 Properties of Continuous Probability Distributions
1. (a) For continuous random variables, probabilities correspond to areas under the
density curve. True.
(b) The area under any continuous probability distribution is 1. True.
(c) A probability density function can take on negative values. False. f (x) ≥ 0
everywhere.
(d) A probability density function can take on values greater than 1. True. The
area under the entire curve is 1, but the height of the curve can be greater than
1.
(e) f (x) = P (X = x) for all x. False. The value of the pdf (the height of the curve
at point x) is not a probability.
(f) f (x) = µ whenever the distribution is symmetric. False.
77
Balka ISE 1.09 6.3. THE CONTINUOUS UNIFORM DISTRIBUTION 78
6.2.1 An Example Involving Integration

R∞
2. (a) c = 19 . For any continuous probability distribution, −∞ f (x)dx = 1. For this
R3 3
distribution, 0 cx2 dx = 1 =⇒ cx3 |03 = 1 =⇒ 9c = 1 =⇒ c = 91 .
R3 2 3 33 23
(b) P (X > 2) = 2 x9 dx = x27 |32 = 27 − 27 19
= 27 .
R 1 x2 x3 1 13 03
(c) P (X < 1) = 0 9 dx = 27 |0 = 27 − 27 = 27 . 1
Ra 2
(d) 2.052. We need to fnd the value a such that 0 x9 dx = 0.32. This implies
a3 03
√3
27 − 27 = 0.32. a = 0.32 · 27 = 2.052.
Ra 2
(e) 2.381. We need to fnd the value a such that 0 x9 dx = 0.50. This implies
a3 03
√3
27 − 27 = 0.50. a = 0.50 · 27 = 2.381.
R∞
(f) For any continuous probability distribution, E(X) = −∞ xf (x)dx. For this
R3 2 4 34
distribution, E(X) = 0 x · x9 dx = x36 |03 = 36 = 2.25.
R∞
3. (a) c = 19 . For any continuous probability distribution, −∞ f (x)dx = 1. For this
R 10
distribution, 4 xc2 dx = 1 =⇒ − xc |10 4 = 1 =⇒ −c( 10 − 4 ) = 1 =⇒ c = 3 .
1 1 20
R 9 20 1
(b) P (8 < X < 9) = 8 3 · x2 dx = − 20 1 9
3 · x |8 = − 20 ( 1 − 1 ) = 216
20
≈ 0.093.
R a 203 91 8
(c) We need to fnd the value a such that 4 3 · x2 dx = 0.44. This implies
− 20 1 1
3 ( a − 4 ) = 0.44 =⇒ a = 5.4348.
R∞
4. (a) P (X > 0.22) = 0.22 10e−10x dx = −e−10x |0.22 ∞ = 0 − (−e−10×0.22 ) = e−2.2 ≈
0.111.
R∞
(b) P (X > 0.02) = 0.02 10e−10x dx = −e−10x |0.02 ∞ = 0 − (−e−10×0.02 ) = e−0.2 ≈
0.819.
R 0.12
(c) P (0.08 < X < 0.12) = 0.08 10e−10x dx = −e−10x |0.08 0.12 = (−e−10×0.12 ) −
(−e −10×0.08 ) = −e −1.2 − (−e −0.8 ) ≈ 0.449 − 0.301 = 0.148.

Rx
(d) We need to fnd the value x such that P (X < x) = 0 10e−10u du = 0.25. This
implies −e−10u |x0 = 0.25 =⇒ 1 − e−10x = 0.25. Solving for x, x = ln(1−0.25) −10 =
0.02877.
(e) The median is the value x suchRthat P (X < x) = 0.5. So we need to fnd the
x
value x such that P (X < x) = 0 10e−10u du = 0.50. This implies −e−10u |0x =
0.5 =⇒ 1 − e−10x = 0.5. Solving for x, x = ln(1−0.5) −10 = 0.0693.
6.3 The Continuous Uniform Distribution
5. (a) Since the base is (25 − 5 = 20), the height of the rectangle must be 1
25−5 = 20 .
1
So,
20 for 5 ≤ x ≤ 25
( 1
f (x) =
0 elsewhere
Balka ISE 1.09 6.4. THE NORMAL DISTRIBUTION 79
(b) 0 (for any continuous random variable X, P (X = a) = 0 for all a).

(c) (25 − 8) × 20
1
= 0.85. (The area under the curve to the right of 8 is 0.85.)
(d) P (X > 8.0|X < 7.3) = P (X>8.0∩X<7.3)
P (X<7.3)
0
= P (X<7.3) = 0 (since an observation
cannot be both greater than 8 and less than 7.3.
1
(25−8.0)×
(e) P (X > 8.0|X > 7.3) = P (X>8.0∩X>7.3)
P (X>7.3) = P (8.0<X<25) 0.85
P (X>7.3) = (25−7.3)× 1 = 0.885 =
20
20
0.96045.
(f) 15 (since half the area lies to the left of 15, half to the right).
(g) 15. (For symmetric distributions, the mean equals the median. We could also
fnd the mean by integration.)
(h) Let a represent the 47th percentile. The area to the left of a is 0.47 (by the
defnition of a percentile). So (a − 5) × 201
= 0.47 =⇒ a = 14.4.
(i) IQR = Q3 − Q1 . (Q1 − 5) × 20 = 0.25 =⇒ Q1 = 10. (Q3 − 5) × 20
1 1
= 0.75 =⇒
Q3 = 20. IQR = 20 − 10 = 10.
6. (a) The uniform distribution is symmetric. True.

(b) For any uniform distribution, the mean and median are equal. True. The
mean and median are equal in symmetric distributions.
(c) For any uniform distribution, Q1 = −Q3 . False. Q1 = −Q3 only if the uniform
distribution has a mean of 0.
(d) For any uniform distribution, Q3 − Q2 = Q2 − Q1 . True. For symmetric
distributions, the distance between the median and third quartile will be the
same as the distance between the median and frst quartile.
(e) If a random variable has a uniform distribution, then it cannot take on negative
values. False.
(f) If a random variable has a uniform distribution, then its standard deviation is
greater than its mean. False. Depending on the value of the parameters, the
standard deviation can be greater than, less than, or equal to the mean.
6.4 The Normal Distribution
7. µ and σ 2 are the parameters of the distribution (or µ and σ). µ can be any value
(−∞ < µ < ∞), but σ 2 must be positive (σ 2 > 0).
8. (a) The normal distribution is symmetric about µ. True.

(b) If X is a normally distributed random variable, then P (X < 0) > 0. True. A
normally distributed random variable can take on any fnite value.
(c) If X is a normally distributed random variable, then P (X = 0) = 0. True. The
probability a continuous random variable takes on any single value is always
0.
(d) The standard deviation of a normal distribution cannot be greater than the
mean. False.
(e) The mean and median of a normal distribution are always equal. True. The
normal distribution is symmetric, and the mean and median are equal in sym-
metric distributions.
6.4.1 Finding Areas Under the Standard Normal Curve
9. (a) 0.6443.
(b) 1 − 0.6443 = 0.3557.
(c) 0.1443.
(d) 0.1003.
(e) 1 − 0.1003 = 0.8997.
(f) 0.
(g) 0.3997.
P (Z>2.41∩Z>0)
(h) P (Z > 2.41|Z > 0) = P (Z>0) = PP(Z>2.41) 0.008
(Z>0) = 0.5 = 0.016.
(i) P (−2.41 < Z < 2.41|Z > 0) = P (−2.41<Z<2.41∩Z>0)
P (Z>0) = P (0<Z<2.41)
P (Z>0) = 0.4920
0.5 =
0.984.
(j) ≈ 0.95 (by the table), 0.9541653 (using software).
(k) By the table: Q1 ≈ −.67, Q3 ≈ .67, IQR ≈ 1.34. Using software: Q1 ≈
−0.6744898, Q3 ≈ 0.6744898, IQR ≈ 1.348980
(l) P (Z < 2.13) = 0.9834. If two values are randomly and independently selected,
the probability they are both less than 2.13 is 0.98342 = 0.9671.
10. (a) 0.9968 (the area to the right of −2.73 under the standard normal curve).
(b) 0.0032 (the area to the right of 2.73 under the standard normal curve).
(c) 1.17.
(d) −1.17.
(e) 1.96.
(f) −0.71.
(g) 0.00052. The probability a single value is greater than 2.0 is 0.0228. The
probability two values, independently sampled, are both greater than 2.0 is
0.02282 = 0.00052.
11. 0 (the area to the right of 31788905573847358927509487 under the standard normal
curve is minuscule).
6.4.2 Standardizing Normally Distributed Random Variables
12. (a) P (X < 105) = P (Z < 105−12020 ) − P (Z < −0.75) = 0.2266.

(b) P (X > 87) = P (Z > 87−12020 ) − P (Z > −1.65) = 0.9505.
(c) P (92 < X < 108) = P ( 20 < Z < 108−120
92−120
20 ) = P (−1.4 < Z < −0.6) =
0.2743 − 0.0808 = 0.1935.
(d) P (82 < X < 87) = P ( 82−120 20 < Z < 87−120
20 ) = P (−1.9 < Z < −1.65) =
0.0495 − 0.0287 = 0.0208.
(e) 0.
(f) 103.2. The 20th percentile of the standard normal distribution is approxi-
mately −0.84. The 20th percentile of the distribution of X is approximately
µ + σz = 120 + 20(−0.84) = 103.2. (Using software, a more precise value is
103.1676.)
(g) 136.8. The 80th percentile of the standard normal distribution is approxi-
mately 0.84. The 80th percentile of the distribution of X is approximately
µ + σz = 120 + 20 · 0.84 = 136.8. (Using software, a more precise value is
136.8324.)
(h) IQR = Q3 − Q1 . The frst quartile of X is approximately 120 + 20(−0.67) =
106.6, and the third quartile is approximately 120 + 20 · 0.67 = 133.4. The
IQR is approximately 133.4 − 106.6 = 26.8. (This was calculated using the
table, but there is quite a bit of rounding error. Using software, a more precise
value is 26.97959.)
13. Let X represent the weight (in kg) of a randomly selected newborn female African
elephant that is born in captivity. Then X ∼ N (95.1, 13.72 ).
(a) P (X > 120.0) = P (Z > 120.0−95.1

13.7 ) = P (Z > 1.818) = 0.035.
(b) P (X < 100.0) = P (Z < 100.0−95.1
13.7 ) = P (Z < 0.358) = 0.640.
(c) P (90.0 < X < 110.0) = P ( 13.7 90.0−95.1
< Z < 110−95.1
13.7 ) = P (−0.372 < Z <
1.088) = 0.862 − 0.355 = 0.507.
(d) The 30th percentile of the standard normal distribution is approximately
−0.52. The 30th percentile of newborn weights is approximately 95.1+13.7(−0.52) =
88.0 kg. (There is a little rounding error here. Using software, a more precise
value is 87.92.)
(e) The 80th percentile of the standard normal distribution is approximately 0.84.
The 80th percentile of newborn weights is approximately 95.1 + 13.7 · 0.84 =
106.6 kg.
14. (a) If X has a normal distribution with mean µ and standard deviation σ, then
X −µ
σ has the standard normal distribution. True.
(b) If Z has a standard normal distribution, then the mean and standard deviation
of Z are equal. False. µ = 0, σ = 1.
Balka ISE 1.09 6.5. NORMAL QUANTILE-QUANTILE PLOTS 82
(c) The frst quartile of the standard normal distribution is greater than 0. False.
The frst quartile is approximately −0.67.
(d) The frst quartile of any normal distribution is greater than 0. False. Any of
the percentiles can be negative, depending on the parameters of the distribution.
(e) If X has a normal distribution, then P (X > µ) = P (X < µ). True. Since the
normal distribution is symmetric about µ, P (X > µ) = P (X < µ) = 0.5.
6.5 Normal Quantile-Quantile Plots: Is the Data Ap-

proximately Normality Distributed?
6.5.1 Examples of Normal QQ Plots for Di˙erent

Distributions
15. (a) Since the points on the plot fall close to a straight line, the population from
which the sample is drawn is likely approximately normal.
(b) Since the points on the plot fall close to a straight line, the population from
which the sample is drawn is likely approximately normal.
(c) This type of plot is typical of a distribution that is symmetric, but has shorter
tails than the normal distribution (perhaps the uniform distribution, or some-
thing similar).
(d) This type of plot is typical of a distribution that is skewed to the right.
16. The normal quantile-quantile plot would look like the one in Figure 6.1.
Sample Quantiles
-2 -1 0 1 2
Theoretical Quantiles
Figure 6.1: A normal quantile-quantile plot showing left skewness and one large outlier.
Balka ISE 1.09 6.6. OTHER IMPORTANT CONTINUOUS PROBABILITY DISTRIBUTIONS 83
6.6 Other Important Continuous Probability Distribu-

tions
6.6.1 The χ2 Distribution
6.6.2 The t Distribution
6.6.3 The F Distribution
17. (a) 0.2. The height of the rectangle is 50−0

1
= 501
. f (x) = 50
1
for 0 ≤ x ≤ 50, and
0 otherwise. P (X < 10) = 0.2 (the area to the left of 10).
(b) 0.8 (the area to the right of 10).
(c) P (12 < X < 23) = 23−12 11
50 = 50 = 0.22.
(d) P (−5 < X < 5) = P (0 < X < 5) (since there is no area to the left of 0).
1
P (0 < X < 5) = (5 − 0) × 50 = 0.1.
(e) a = 10 (since the area to the right of 10 is 0.8).
(f) The 20th percentile is equal to 10 (since the area to the left of 10 is 0.2).
(g) 25.
18. (a) 2. The area of a triangle is 12 × base × height. Since the area under the entire
curve must equal 1, 12 × base × height = 1 =⇒ 12 × (1 − 0) × height = 1 =⇒
height = 2. So the highest point of the triangle is 2, which occurs at x = 1.
(b) f (0) = 0 and f (1) = 2, which implies that f (0.5) = 1. (Since f (x) is a straight
line in this question.)
(c) f (x) = 2x for 0 < x < 1 and 0 otherwise. (The curve is a line with slope =
2.)
(d) 0.
(e) 1. P (X < 1.5) is the area to the left of 1.5. Since all of the area lies between
0 and 1, P (X < 1.5) = 1.
(f) 0.25. P (X < 0.5) is the area to the left of 0.5. At x = 0.5, f (0.5) = 1. The
area to the left of 0.5 is a triangle with a base of (0.5 − 0) = 0.5 and a height
of 1, so P (X < 0.5) = 12 × 0.5 × 1 = 0.25.
(g) P (X > 0.5) = 1 − P (X < 0.5) = 1 − 0.25 = 0.75.
√
(h) 0.5. We need to fnd the value a such that the area to the left of a is 0.5.
The area to the left of a point a is 12 × (a − 0) × 2a. Setting this equal to 0.5,
√
a2 = 0.5 =⇒ a = 0.5.
6.7.2 Concepts
19. 1, since the area to the left of −458958643961346 under the standard normal curve
is minuscule.
20. (a) The 95th percentile of a continuous random variable cannot be less than 0.
False.
(b) The mean of a continuous random variable cannot be negative. False. Con-
tinuous random variables often have negative means.
(c) The mean of a continuous random variable can be greater than the third
quartile. True, but it would be unusual.
(d) The normal distribution is symmetric. True.
(e) If X has a normal distribution, then the mean and median of X are equal.
True, since for symmetric distributions the mean and median are equal.
6.7.3 Applications
21. Let X represent the total serum cholesterol (mmol/L) of a randomly selected Cana-
dian male between 60 and 79 years of age. Then X ∼ N (4.7, 1.22 ).
(a) P (X ≥ 6.2) = P (Z ≥ 6.2−4.7

1.2 ) = P (Z > 1.25) = 0.1056.
(b) 0.1056. (The proportion of men in this age group with a total serum cholesterol
of at least 6.2 mmol/L is the same as the probability a randomly selected man
in this age group has a level at least this high (which we found in 21a).)
(c) 0.467. P (X < 4.6) = P (X < 4.6−4.7
1.2 ) = P (Z < −0.083) = 0.467. (There will
be a little rounding error if you use a table: P (Z < −0.08) = 0.4681.)
(d) 0.428. P (4.6 < X < 6.2) = P ( 4.6−4.7
1.2 < Z < 6.2−4.7
1.2 ) = P (−0.083 < Z <
1.25) = 0.894 − 0.467 = 0.428. (There will be a little rounding error if you use
a table: P (Z < −0.08) = 0.4681.)
(e) 3.69 mmol/L. The 20th percentile of the standard normal distribution is ap-
proximately −0.84 and thus the 20th percentile of total serum cholesterol levels
is approximately µ + σz = 4.7 + 1.2(−0.84) = 3.69 mmol/L.
(f) 5.71 mmol/L. The 80th percentile of the standard normal distribution is ap-
proximately 0.84 and thus the 80th percentile of cholesterol levels is approxi-
mately µ + σz = 4.7 + 1.2 · 0.84 = 5.71 mmol/L.
22. Let X represent the height (in cm) of a randomly selected Canadian female between
20 and 39 years of age. Then X ∼ N (163.3, 6.42 ).
(a) P (X ≥ 182.88) = P (Z ≥ 182.88−163.3

6.4 ) = P (Z ≥ 3.059) = 0.0011.
(b) P (X > 170.0) = P (Z > 170.0−163.3
6.4 ) = P (Z > 1.047) = 0.148. (There will be
a little rounding error if you are using a table: P (Z > 1.05) = 0.1469.)
(c) P (X < 170.0) = P (Z < 170.0−163.3
6.4 ) = P (Z < 1.047) = 0.852. (There will be
a little rounding error if you are using a table: P (Z < 1.05) = 0.8525.)
(d) P (X > 150.0) = P (Z > 150.0−163.3
6.4 ) = P (Z > −2.078) = 0.981.
(e) P (−1 < Z < 1) = 0.8413 − 0.1587 = 0.683.
(f) 157.9 cm. The 20th percentile of the standard normal distribution is approx-
imately −0.84. Thus the 20th percentile of female height is approximately
µ + σz = 163.3 + 6.4(−0.84) = 157.9 cm. (Using software, a more precise
value is 157.91.)
23. Let X represent the height (in cm) of a randomly selected Canadian male between
(a) P (X ≥ 182.88) = P (Z ≥ 182.88−177.76.5 ) = P (Z ≥ 0.797) = 0.213. (There will

be a little rounding error if you are using a table: P (Z ≥ 0.80) = 0.2119.)
(b) P (X > 170.0) = P (Z > 170.0−177.76.5 ) = P (Z > −1.184) = 0.882. (There will
be a little rounding error if you are using a table: P (Z > 1.18) = 0.881.)
(c) P (X < 170.0) = P (Z < 170.0−177.7 6.5 ) = P (Z < −1.184) = 0.118. This is 1
minus the probability found in part 23b.
(d) The 99th percentile of the standard normal distribution is approximately 2.33.
The 99th percentile of the heights of Canadian males in this age group is
approximately µ + σz = 177.7 + 6.5 × 2.33 = 192.8 cm. (Using software, a
more precise value is 192.82.)
(e) In 23b, we found the probability a randomly selected Canadian male in this
age group is taller than 170.0 cm is 0.882. The probability that three randomly
and independently selected males are all taller than 170.0 cm is 0.8823 = 0.686.
(f) In 23a, we found the probability a randomly selected Canadian male in this
age group is at least 6 feet fall is 0.213. For two randomly selected males in this
age group, the probability they are both at least 6 feet tall is 0.2132 = 0.045.
(g) The probability that a randomly selected Canadian male in this age group is at
least 6 feet tall is 0.213 (found in 23a). If 5 males are randomly selected from
this age group, the number that are at least 6 feet ball will have a binomial
distribution with n = 5 and p = 0.213. �5 If we let Y be the number that are at
least 6 feet tall, then P (Y = 2) = 2 0.213 (1 − 0.213)3 = 0.221.
2
24. Let X represent the test diameter (in mm) of a randomly selected mature male
green sea urchin in this area. Then X ∼ N (39.7, 3.82 ).
(a) 0.290. P (X > 40) = P (Z > 40−37.9

3.8 ) = P (Z > 0.553) = 0.290. (There will be
a little rounding error if you are using a table: P (Z > 0.55) = 0.2912.)
(b) 0.019. P (X < 30) = P (Z < 30−37.9
3.8 ) = P (Z < −2.079) = 0.019.
(c) 0.691. P (30 < X < 40) = P ( 30−37.9

3.8 < Z < 40−37.9
3.8 ) = P (−2.079 < Z <
0.553) = 0.710 − 0.019 = 0.691. (There will be a little rounding error if you
are using a table: P (Z < 0.55) = 0.7088 and P (Z < −2.08) = 0.0188, so
P (−2.08 < Z < 0.55) = 0.690.)
(d) 37.9 mm. (The normal distribution is symmetric, so the mean and median are
equal.)
(e) 35.34 mm. The 25th percentile of the standard normal distribution is approx-
imately −0.67. Thus the 25th percentile of test diameters is approximately
µ + σz = 37.9 + 3.8(−0.67) = 35.35 mm. (This answer contains a little round-
ing error. Using software, we can fnd that the 25th percentile of the standard
normal distribution is −0.67449, and the 25th percentile of test diameters is
35.34 mm.)
25. Let X represent the weight (in grams) of the cereal in a randomly selected bag of
cereal of this type. Then X ∼ N (374.5, 2.82 ).
(a) 0.010. P (X ≥ 380) = P (Z ≥ 380−374.5

2.8 ) = P (Z ≥ 1.964) = 0.025.
(b) 0.010. P (X < 368) = P (Z < 368−374.5
2.8 ) = P (Z < −2.321) = 0.010.
(c) 0.936. P (370 < X < 380) = P ( 2.8
370−374.5
< Z < 380−374.5
2.8 ) = P (−1.607 < Z <
1.964) = 0.975 − 0.054 = 0.921.
(d) 374.5 grams. (The normal distribution is a symmetric distribution, and the
mean and median are equal in symmetric distributions.)
(e) 371.6 grams. The 15th percentile of the standard normal distribution is ap-
proximately −1.04. Thus the 15th percentile of cereal weight is approximately
µ + σz = 174.5 + 2.8(−1.04) = 371.6 grams. (There is a little rounding error
when fnding a percentile of the standard normal distribution using the table.
Using software, we can fnd that the 15th percentile of the standard normal
distribution is −1.036433, and the 15th percentile of cereal weight is is 371.598
grams.)
26. Let X represent the amount of protein (in grams) of a randomly selected cup of
this type of 2% milk. Then X ∼ N (8.0, 0.302 ).
(a) 0.631. P (X < 8.10) = P (Z < 8.1−8.0

0.30 ) = P (Z < 0.333) = 0.631. (If you use a
table, there will be a little rounding error: P (Z < 0.33) = 0.6293.)
(b) P (X > 14.0) = P (Z > 14.0−8.0
0.30 ) = P (Z > 20) ≈ 0.
(c) 0.125. P (8.05 < X < 8.20) = P ( 8.05−8.0 0.30 < Z < 8.20−8.0
0.30 ) = P (0.167 <
Z < 0.50) = 0.691 − 0.566 = 0.125. (If you use a table, there will be a little
rounding error: P (Z < 0.17) = 0.5674.)
(d) 8.70 grams. The 99th percentile of the standard normal distribution is ap-
proximately 2.33. Thus the 99th percentile of protein in a cup of 2% milk
is approximately µ + σz = 8.0 + 0.3 · 2.33 = 8.70 grams. (There is a little
rounding error when fnding a percentile of the standard normal distribution
using the table. Using software, we can fnd that the 99th percentile of the
standard normal distribution is 2.326348.)
proximately 0.67. Thus the 75th percentile of protein in a cup of 2% milk
is approximately µ + σz = 8.0 + 0.3 · 0.67 = 8.20 grams. (There is a little
standard normal distribution is 0.6744898.)
27. Let X represent the total amount of fat (in grams) in a randomly selected serving
(85 grams) of this type of cooked ground beef. Then X ∼ N (10.2, 1.82 ).
(a) 0.996. P (X < 15.0) = P (Z < 15.0−10.2

1.8 ) = P (Z < 2.667) = 0.996.
(b) 0.730. P (8.0 < X < 12.0) = P ( 1.8 < Z < 12.0−10.2
8.0−10.2
1.8 ) = P (−1.222 < Z <
1.000) = 0.841 − 0.111 = 0.730.
(c) 0.0003. For a single serving: P (X > 14.0) = P (Z > 14.0−10.21.8 ) = P (Z >
2.111) = 0.0174. The probability that two randomly selected servings both
have more than 14.0 grams of fat is 0.01742 = 0.0003.
(d) 8.33 grams. The 15th percentile of the standard normal distribution is ap-
proximately −1.04. Thus the 15th percentile of fat in a serving of this type of
beef is approximately µ + σz = 10.2 + 1.8(−1.04) = 8.33 grams. (There is a
little rounding error when fnding a percentile of the standard normal distri-
bution using the table. Using software, we can fnd that the 15th percentile of
the standard normal distribution is −1.036433, and the 15th percentile of fat
content in a serving is 8.334 grams.)
proximately 1.645. Thus the 95th percentile of fat in a serving of this type of
beef is approximately µ + σz = 10.2 + 1.8 · 1.645 = 13.16 grams.
28. Let X represent the length (in mm) of the head of a randomly selected lizard of
the species P. versicolor. Then X ∼ N (11.90, 0.412 ).
(a) 0.582. P (11.0 < X < 12.0) = P ( 11.0−11.90

0.41 < Z < 12.0−11.90
0.41 ) = P (−2.195 <
Z < 0.244) = 0.596 − 0.014 = 0.582.
(b) 0.835. P (X ≥ 11.5) = P (Z ≥ 11.5−11.90
0.41 ) = P (Z ≥ −0.976) = 0.835. (If you
use a table, there will be a little rounding error: P (Z ≥ −0.98) = 0.8365.)
(c) The probability that a randomly selected adult female lizard of this species has
a head length that is at least 22.5 mm is: P (X ≥ 22.5) = P (Z ≥ 22.5−11.90
0.41 )=
P (Z ≥ 25.9) ≈ 0. Either the person caught an extremely unusual lizard, or
this lizard is not of this species.
(d) 11.37 mm. The 10th percentile of the standard normal distribution is approxi-
mately −1.28. Thus the 10th percentile of head lengths in this type of lizard is
approximately µ + σz = 11.90 + 0.41(−1.28) = 11.38 grams. (There is a little

standard normal distribution is −1.281552, and the 10th percentile of head
lengths is 11.3745 mm.)
imately 1.28. Thus the 90th percentile of head lengths in this type of lizard
is approximately µ + σz = 11.90 + 0.41 · 1.28 = 12.42 grams. (There is a
little rounding error when fnding a percentile of the standard normal distri-
bution using the table. Using software, we can fnd that the 90th percentile of
the standard normal distribution is 1.281552, and the 90th percentile of head
lengths is 12.4254 mm.)
29. Let X represent the weight (in grams) of the heart of a randomly selected adult
Caucasian male. Then X ∼ N (365, 712 ).
(a) P (X > 400) = P (Z > 400−36571 ) = P (Z > 0.493) = 0.311. (There is a little
rounding error if you use the table: P (Z > 0.49) = 0.3121.)
(b) P (X > 453.6) = P (Z > 453.6−365
71 ) = P (Z > 1.248) = 0.106.
(c) P (X < 300) = P (Z < 300−365
71 ) = P (Z < −0.915) = 0.180. (There is a little
rounding error if you use the table: P (Z < −0.92) = 0.1788.)
(d) P (300 < X < 400) = P ( 300−365
71 < Z < 400−365
71 ) = P (−0.915 < Z < 0.493) =
0.689 − 0.180 = 0.509.
(e) The 90th percentile of the standard normal distribution is approximately 1.28.
The 90th percentile of heart weight is approximately 365 + 71 × 1.28 = 455.9
grams. (This answer has a little rounding error. If we use software, we can
fnd that the correct value to 2 decimal places is 455.99.)
30. Let X represent the weight (in kg) of a randomly selected newborn baby in the
United States. Then X ∼ N (3.3, 0.52 ).
(a) P (X < 2.0) = P (Z < 2.0−3.3

0.5 ) = P (Z < −2.60) = 0.005.
(b) 4.0−3.3
P (X < 4.0) = P (Z < 0.50 ) = P (Z < 1.40) = 0.919.
(c) P (2.0 < X < 4.0) = P (−2.60 < Z < 1.40) = 0.919 − 0.005 = 0.914.
(d) 0.50 ) = P (Z > 5.4), which is very close to 0. (Using
P (X > 6.0) = P (Z > 6.0−3.3
software, a more precise value is 3.3 × 10−8 . But it’s not wise to count on the
normal approximation to provide a great deal of precision this far out in the
tail. In any event, the probability is near 0.)
(e) The 90th percentile of the standard normal distribution is approximately 1.28,
and thus the 90th percentile of full-term newborn baby weight is approximately
µ + σz = 3.3 + 0.50 × 1.28 = 3.94 kg.
(f) The 25th percentile of the standard normal distribution is approximately
−0.67, and thus the 90th percentile of full-term newborn baby weight is ap-
proximately µ + σz = 3.3 + 0.50(−0.67) = 2.96 kg.
31. Let X represent the lifetime (in days) of a randomly selected insect of this type.
Then X ∼ N (60.2, 18.02 ).
(a) P (Z > 100−60.2

18.0 ) = P (Z > 2.211) = 0.014.
(b) P (Z > 18.0 ) = P (Z > 7.767) ≈ 0. (Using software, a more precise value
200−60.2
is 4.0 × 10−15 .)
(c) P (Z < 50−60.2
18.0 ) = P (Z < −0.567) = 0.285. (If you use the table, there will be
a little rounding error: P (Z < −0.57) = 0.2843.)
(d) P ( 50−60.2
18.0 < Z < 70−60.2
18.0 ) = P (−0.567 < Z < −0.544) = 0.707 − 0.285 =
0.421.
(e) The 90th percentile of the standard normal distribution is approximately 1.28,
and thus the 90th percentile of life lengths is approximately µ + σz = 60.2 +
18 × 1.28 = 83.2 days. (This answer has a little rounding error. If we use
software, we can fnd that the correct value is 83.27.)
32. Let X represent the purity level in a randomly selected oxygen container of this
type. Then X ∼ N (99.50, 0.0482 ).
(a) P (X < 99.43) = P (Z < 99.43−99.50

0.048 ) = P (Z < −1.458) = 0.072.
(b) P (Z > 99.60−99.50
0.048 ) = P (Z > 2.083) = 0.019.
(c) The 10th percentile of the standard normal distribution is approximately
−1.28, and thus the 10th percentile of purity levels is approximately µ + σz =
99.50 + 0.048(−1.28) = 99.44.
(d) The 25th percentile (Q1 ) of the standard normal distribution is approximately
−0.67, and thus the 25th percentile of purity levels is approximately µ + σz =
99.50 + 0.048(−0.67) = 99.468. Q3 is approximately µ + σz = 99.50 + 0.048 ·
0.67 = 99.532.
(e) The probability that a single container has a purity level greater than 99.40%
is P (Z > 99.40−99.50
0.048 ) = P (Z > −2.083) = 0.981. The probability that 4
independently sampled containers all have purity levels greater than 99.40 is
0.9814 = 0.93.
33. Let X represent the weight (in grams) of a randomly selected adult pigeon in this
city. Then X ∼ N (370, 282 ).
(a) P (X < 350) = P (Z < 350−370 28 ) = P (Z < −0.714) = 0.238. (If you use a
table, there will be a little rounding error: P (Z < −0.71) = 0.2389.)
(b) P (X > 453.6) = P (Z > 453.6−370
28 ) = P (Z > 2.986) = 0.0014.
(c) P (350 < X < 400) = P ( 28 < Z < 400−370
350−370
28 ) = P (−0.714 < Z < 1.071) =
0.858 − 0.238 = 0.620.
(d) P (X > 6000) = P (Z > 6000−37028 ) = P (Z > 201.07) ≈ 0. (Under the assumed
model, this probability is non-zero, but very very close to 0. For all intents
and purposes there is no chance of observing this event.)
(e) The 5th percentile of the standard normal distribution is approximately −1.645.
The 5th percentile of pigeon weights is approximately µ+σz = 370+28(−1.645) =
323.94 grams.
34. Let X represent the cheese demand (in kg) on a randomly selected day. Then
X ∼ N (50, 36). Note that √ the variance is given (σ = 36), and the standard
2
deviation is therefore σ = 36 = 6 kg.
(a) P (X > 60) = P (Z > √−50 ) = P (Z > 1.667) = 0.048.

60
36
(b) P (X < 45) = P (Z < 45√−50
36
) = P (Z < −0.833) = 0.202. (If you use a table,
there will be a little rounding error: P (Z < −0.83) = 0.2032.)
(c) The 80th percentile of the standard normal distribution is approximately 0.84
and thus
√ the 80th percentile of cheese demand is approximately µ + σz =
50 + 36 × 0.84 = 55.04 kg. (There is a little rounding error here. If we use
software we can fnd the correct value of 55.05 kg.)
(d) This is equivalent to asking for the 99th percentile of cheese demand. The 99th
percentile of the standard normal distribution is approximately 2.33 and √ thus
the 99th percentile of cheese demand is approximately µ + σz = 50 + 36 ×
2.33 = 63.98 kg. (There is a little rounding error here. If we use software we
can fnd the correct value of 63.96 kg.)
35. Let X represent the diameter (in cm) of a randomly selected shaft. Then X ∼
N (4.10, 0.052 ).
(a) P (X > 4.23) = P (Z > 4.23−4.10

0.05 ) = P (Z > 2.6) = 0.0047.
(b) P (X < 4.00) = P (Z < 4.00−4.10
0.05 ) = P (Z < −2.0) = 0.0228.
(c) The 10th percentile of the standard normal distribution is approximately
−1.28, and thus the 10th percentile of diameters is approximately µ + σz =
4.10 + 0.05(−1.28) = 4.036 cm.
(d) The 90th percentile of the standard normal distribution is approximately 1.28,
and thus the 90th percentile of diameters is approximately µ + σz = 4.10 +
0.05 · 1.28 = 4.164 cm.
(e) P (Shaft must be scrapped) = P (X < 4.00)+P (X > 4.20) = P (Z < 4.00−4.10
0.05 )+
P (Z > 4.20−4.10
0.05 ) = P (Z < −2.0) + P (Z > 2.0) = 0.0228 + 0.0228 = 0.0456.
Chapter 7
Sampling Distributions
J.B.’s strongly suggested exercises: 1, 6, 7, 8, 9, 10, 20, 21, 22, 23, 24, 25
7.1 Introduction
1. (a) We view statistics as random variables that have probability distributions.

True.
(b) We view parameters as random variables that have probability distributions.
False. A parameter’s value is not usually known, but it is a fxed value that
does not change from sample to sample.
(c) The sampling distribution of a statistic is the probability distribution of the
statistic. True. “Sampling distribution” is another name for the probability
distribution of a statistic.
(d) In repeated sampling, the value of a statistic will vary about the parameter it
estimates. True.
(e) In repeated sampling, the value of a parameter will vary about the statistic
that estimates it. False. A parameter’s value is not usually known, but it is a
fxed value that does not change from sample to sample.
2. (a) µ = 14.
(b) 42 = 6.
�
91
Balka ISE 1.09 7.1. INTRODUCTION 92
Sample values Sample mean Probability of occurring

12, 13 12.5 1/6
12, 15 13.5 1/6
(c) 12, 16 14.0 1/6
13, 15 14.0 1/6
13, 16 14.5 1/6
15, 16 15.5 1/6
x̄ 12.5 13.5 14.0 14.5 15.5

(d)
p(x̄) 1/6 1/6 2/6 1/6 1/6
(e)
¯) =
X 1 1 2 1 1
µX̄ = E(X ¯ x) = 12.5 · + 13.5 · + 14.0 · + 14.5 · + 15.5 · = 14
xp(¯
6 6 6 6 6
Note that E(X
¯ ) = µ (on average, the sample mean equals the population
mean).
Sample values Sample variance Probability of occurring

12, 13 0.5 1/6
12, 15 4.5 1/6
3. (a) 12, 16 8.0 1/6
13, 15 2.0 1/6
13, 16 4.5 1/6
15, 16 0.5 1/6
s2 0.5 2.0 4.5 8.0

(b)
p(s2 ) 2/6 1/6 2/6 1/6
(c) 2
6 = 13 .
4. (a) Recall that the probability distribution of a discrete random variable is a

listing of all possible values of that random variable and their probabilities of
occurring. So the probability distribution of X is:
x 0 1
p(x) 0.5 0.5
This is a Bernoulli distribution with p = 0.5. (We’ll get heads half the time.)
(b) µX = E(X) = xp(x) = 0 · 0.50 + 1 · 0.50 = 0.50.
P
(c) σX2 = (x − µ)2 p(x) = (0 − 0.5)2 · 0.5 + (1 − 0.5)2 · 0.5 = 0.25.

P
(d) The following table gives the 4 possible outcomes, the corresponding values of
X¯ , and their probabilities of occurring.
The probability distribution of the sample mean X ¯ is thus:
x̄ 0 0.5 1
p(x̄) 0.25 0.5 0.25
Balka ISE 1.09 7.2. THE SAMPLING DISTRIBUTION OF THE SAMPLE MEAN 93
Outcome TT TH HT HH
Value of X¯ 0 0.5 0.5 1
Probability 0.25 0.25 0.25 0.25
This is the sampling distribution of the average number of heads per toss (X ¯)
when a fair coin is tossed twice. It was not necessary to actually toss the coin
to derive the sampling distribution.
(e) The mean of the sampling distribution of X ¯ is represented by µ ¯ . We can use
X
the formula for the mean of a discrete random variable to fnd the mean of X: ¯
X
¯) =
µX̄ = E(X ¯ x) = 0 · 0.25 + 0.5 · 0.5 + 1 · 0.25 = 0.50
xp(¯
Note that the mean of the probability distribution of the sample mean is equal
to mean of the probability distribution for a single toss: µX¯ = µX = 21 .
(f) The variance of the sampling distribution of X ¯ is represented by σ 2 . We
X̄
can use the formula for the variance of a discrete random variable to fnd the
variance of X
¯:
X
2
σX̄ = (x̄−µX̄ )x p(x̄) = (0 = 0.5)2 ·0.25+(0.5−0.5)2 ·0.5+(1−0.5)2 ·0.25 = 0.125
Note that the variance of X

¯ is the variance of X divided by the number of
σ 2
tosses: (σX̄
2 = X = 0.25 = 0.125).
2 2
5. (a) Let X represent the number of winning cups. Then X has a binomial distri-
bution with n = 2 and p = 61 . The distribution of X in table form is:
x 0 1 2
25 10 1
p(x) 36 36 36
(b) The proportion of winning cups is the number of winning cups divided by the
number of cups purchased (p̂ = X n ). So the sample proportion takes on 3
possible values: p̂ = 02 = 0, or p̂ = 12 = 0.5, or p̂ = 22 = 1. The sampling
distribution of the sample proportion is:
p̂ 0 0.5 1
Probability 25 36
10
36
1
36
7.2 The Sampling Distribution of the Sample Mean
6. (a) µX¯ = µ = 10.

σ √
(b) σX¯ = √ = √625 16
= 6.25.
n
(c) X¯ ∼ N (10, 625 ).
16
Balka ISE 1.09 7.3. THE CENTRAL LIMIT THEOREM 94
30−µ
(d) P (X¯ > 30) = P (Z > √ )
σ/ n
= P (Z > 30−10
√ ) = P (Z > 3.2) = 0.0007.
25/ 16
(e) P (0 < X¯ < 20) = P ( 0−10
√
25/ 16
<Z< 20−10
√ ) = P (−1.6 < Z < 1.6) = 0.8904.
25/ 16
7. Plot B is the sampling distribution of the sample mean. In this situation the
sampling distribution of X ¯ will be approximately normal with a mean of 20 and a
standard deviation of σX̄ = √σn = √20 400
= 1. Although A, B, and D all have means
of 20, plot B is the only one with a standard deviation that appears to be close to
1.
7.3 The Central Limit Theorem
8. The central limit theorem tells us that sums and means of random variables have
distributions that tend toward the normal distribution as the number of observa-
tions increases, regardless of the distribution from which we are sampling. This
implies that for large sample sizes, the distribution of the sample mean will be
approximately normal, even when we are sampling from a non-normal population.
The central limit theorem is very important in the world of statistics. Because of
the CLT, we can often make probability calculations and use statistical inference
procedures based on the normal distribution, even when the underlying population
is not normally distributed.
9. False. µ is a parameter, and as such it does not have a sampling distribution. (We
view µ as a fxed, unchanging value.)
10. (a) The standard deviation of the sampling distribution of the sample mean de-
pends on the value of µ. False. σX¯ = √σn depends only on σ and n.
(b) We cannot possibly determine any characteristics of a statistic’s sampling dis-
tribution without repeatedly sampling from the population. False. We can
often mathematically determine characteristics of the sampling distribution.
For example, we have shown that E(X ¯ ) = µ.
(c) The sampling distribution of X is always at least approximately normal for
¯
large sample sizes, and is sometimes approximately normal for small sample
sizes. True. The distribution of X ¯ is normal for any sample size if we are
sampling from a normally distributed population. If we are sampling from a
distribution that is approximately normal, then the distribution of X¯ will be
approximately normal for any sample size.
(d) If the sample size is quadrupled, then standard deviation of the sampling
distribution of the sample mean would decrease by a factor of 2. True. √σn
√
decreases by a factor of 4 = 2 if n = 4.
Balka ISE 1.09 7.4. SOME TERMINOLOGY REGARDING SAMPLING DISTRIBUTIONS 95
7.4 Some Terminology Regarding Sampling Distributions
7.4.1 Standard Errors
11. The standard error of a statistic is the estimate of the standard deviation of that
statistic’s sampling distribution. (It is a measure of the variability of the statistic
in repeated sampling.)
(Note: Some sources defne standard error as the true standard deviation of the
statistic’s sampling distribution, and the estimated standard error as the estimate.
But here we reserve the term standard error for the estimated standard deviation
of a statistic’s sampling distribution.)
We view parameters as fxed, usually unknown quantities. As such, parameters do

not have sampling distributions or standard errors.
12. SE(X
¯) = √s
n
= 13
√.7
6
= 5.593.
13. (a) Increasing the sample size decreases SE(X ¯ ).

(b) An increase in the sample variance increases SE(X ¯ ).
(c) SE(X ¯ ) is a˙ected by the variability and the sample size, abut not by the value
of X. The sample mean has no e˙ect on SE(X
¯ ¯ ).
7.4.2 Unbiased Estimators
14. A statistic is an unbiased estimator of a parameter if its expected value is equal to

that parameter. (In other words, a statistic is an unbiased estimator if, on average,
the statistic equals the parameter it estimates.)
15. Yes, since E(X

¯ ) = µ.
7.6 Basic Calculations
7.7 Concepts
16. (a) The letter n represents the number of samples that are drawn in repeated sam-
pling, and not the number of observations in each sample. False. n represents
the number of observations in the sample.
(b) In practice, we usually take a large number of samples so that we can accurately
estimate the sampling distribution of X. ¯ False. In practice, we draw only a
single sample. (We may draw a sample with a large number of observations,
but we draw only a single sample.) We use mathematical techniques to derive
¯
characteristics of the sampling distribution of X.
(c) The standard deviation of the population from which we are sampling de-
creases as the sample size increases. False. The standard deviation of the
population is what it is, and drawing more observations won’t change that.
(d) The standard deviation of the sampling distribution of X ¯ decreases as the
sample size increases. True. √n decreases as n increases.
σ
17. (a) The probability distribution of X is:

x 1 2 3 4 5 6
p(x) 1/6 1/6 1/6 1/6 1/6 1/6
(b) E(X) = xp(x) = 1 · 16 + 2 · 16 + 3 · 16 + 4 · 16 + 5 · 16 + 6 · 16 = 3.5.
P
(c) σ 2 = (x − µ)2 p(x) = (1 − 3.5)2 · 16 + (2 − 3.5)2 · 16 + (3 − 3.5)2 · 16 + (4 − 3.5)2 ·

P
6 + (5 − 3.5) · 6 + (6 − 3.5) · 6 = 12 .
1 2 1 2 1 35
(d) To fnd the sampling distribution of X,¯ we can list all 36 possible outcomes
and calculate X
¯ for each possibility. The results are given in the following
table.
Roll 1 / Roll 2 1 2 3 4 5 6
1 1.0 1.5 2.0 2.5 3.0 3.5
2 1.5 2.0 2.5 3.0 3.5 4.0
3 2.0 2.5 3.0 3.5 4.0 4.5
4 2.5 3.0 3.5 4.0 4.5 5.0
5 3.0 3.5 4.0 4.5 5.0 5.5
6 3.5 4.0 4.5 5.0 5.5 6.0
The 36 possible outcomes are equally likely, so the probability of obtaining

Balka ISE 1.09 7.7. CONCEPTS 97
each value of X¯ is just the proportion of times that value of X

¯ occurs. The
probability distribution of X
¯ is:
x̄ 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 5.5 6.0
p(x̄) 1/36 2/36 3/36 4/36 5/36 6/36 5/36 4/36 3/36 2/36 1/36
(e) The quick way is to realize that µX¯ = µ = 3.5 (found in 17b). The slower way
would bePto use the formula for the expectation of a discrete random variable:
¯
E(X ) = xp(¯
¯ x).
√
35/12
(f) The quick way is to realize that σX̄ = n = √2 ≈ 1.2076. (In 17c we
√σ
found that σ 2 = 35
12 .) The slower way
P is to use the formula for the variance of
a discrete random variable: σX̄
2 = (x̄ − µ)2 p(x̄).
18. (a) µX¯ = µ = 3.5.

√
35/12
(b) σX¯ = √σn = √2000 ≈ 0.03818813.
(c) The central limit theorem tells us that X ¯ will be approximately normally
distributed (with a mean of 3.5 and a standard deviation of approximately
0.038188).
(d) P (X¯ < 3.5190) = P (Z < 3.5190−3.5 ) = P (Z < 0.4975368) = 0.691.
0.03818813
19. (a) If we are sampling from a normally distributed population, then the sampling
distribution of X¯ is normal for any sample size. True
(b) The standard deviation of the sampling distribution of the sample mean in-
creases as the sample size increases. False. σX̄ = √σn decreases as the sample
size increases.
(c) The central limit theorem states that the sample mean equals the population
mean, as long as the sample size is large. False. E(X ¯ ) = µ (for any sample
¯ will not be exactly
size) but there will always be some variability (the value of X
equal to µ).
(d) If the sampling distribution of X¯ is approximately normal, then the population
from which we are sampling must be approximately normal. False. For large
sample sizes the sampling distribution of X ¯ is always approximately normal,
regardless of the shape of the population from which we are sampling, and thus
an approximately normal sampling distribution of X ¯ does not say much about
the sampled population.
(e) All else being equal, the mean of the sampling distribution of X ¯ decreases as
the sample size increases. False. The mean of the sampling distribution of X ¯
is µ, regardless of the sample size.
20. (a) If we draw a very large sample from any population and plot a histogram of the
observations, the shape of the histogram will be approximately normal. False.
A histogram of a large number of observations sampled from a distribution will
Balka ISE 1.09 7.8. APPLICATIONS 98
look like the distribution from which we are sampling.

(b) In practice, we usually know the true standard deviation of the sampling dis-
tribution of X.
¯ False. The population standard deviation σ is almost always
¯
unknown, and so the true standard deviation of the sampling distribution of X
σ
(σX¯ = n ) is almost always unknown.
√
(c) In practice, we usually know the true value of µ. False. In practical problems,
the parameter µ is almost always unknown.
(d) The sample mean is an unbiased estimator of the population mean. True.
E(X ¯ ) = µ.
(e) If we were to repeatedly sample from a population, then the distribution of the
sample mean would become approximately normal as the number of samples
increases, as long as the sample size of each sample stays constant. False. The
distribution of X¯ becomes approximately normal as the sample size n increases
(in other words, as the number of values used to calculate the mean increases).
Repeatedly sampling from a population doesn’t change the sampling distribution
of X¯.
21. (a) Statistics have sampling distributions. True.

(b) The value of a parameter does not vary from sample to sample. True.
(c) The value of a statistic does not vary from sample to sample. False.
(d) All else being equal, the standard deviation of the sampling distribution of the
sample mean will be smaller for n = 10 than for n = 40. False. σX̄ = √σn
decreases for increasing sample size.
(e) The sampling distribution of µ is usually approximately normal for n > 30.
False. We view µ as a fxed, unknown quantity. As such, it does not vary from
sample to sample and does not have a sampling distribution.
7.8 Applications
22. Let X represent the amount of ribofavin (in mg) in a randomly selected large white
egg. Then X ∼ N (0.25, 0.022 ).
(a) P (X < 0.20) = P (Z < 0.20−0.25

0.02 ) = P (Z < −2.5) = 0.0062.
0.022
(b) X ∼ N (0.25, 30 ). (The mean amount of ribofavin in 30 large eggs is ap-
¯
proximately normally distributed with a mean of µX¯ = 0.25 mg and a variance
2 = 0.022 = 0.000013 mg2 .)
of σX̄ 30
(c) P (0.240 < X ¯ < 0.260) = P ( 0.240−0.25
√
0.02/ 30
< Z < 0.260−0.25
√
0.02/ 30
) = P (−2.739 < Z <
2.739) = 0.994.
(d) P (0.250 < X ¯ < 0.255) = P ( 0.250−0.25
√
0.02/ 30
< Z < 0.255−0.25
√
0.02/ 30
) = P (0 < Z <
1.369) = 0.915 − 0.5 = 0.415.
(e) 0.2469 mg. The 20th percentile of the standard normal distribution is ap-
proximately −0.84. The 20th percentile of X
¯ is approximately µ + √σ z =
n
0.25 + 0.02
√
30
(−0.84) = 0.2469 mg.
23. Let X represent the amount of protein (in grams) of a randomly selected breakfast
sandwich of this type. Then X ∼ N (17.0, 0.82 ).
(a) P (X < 16.0) = P (Z < 16.0−17.0

0.80 ) = P (Z < −1.25) = 0.106.
(b) T ∼ N (34.0, 1.28). Let T = X1 + X2 represent the total amount of protein in
2 randomly selected sandwiches. Then E(T ) = 17.0 + 17.0 = 34.0 grams, and
V ar(T ) = V ar(X1 ) + V ar(X2 ) = 0.82 + 0.82 = 1.28 grams2 . Since X1 and X2
are normal, their sum is normal.
(c) X¯ ∼ N (17.0, 0.802 ). (The mean amount of protein in 2 sandwiches is approx-
2
imately normally distributed with a mean of µX¯ = 17.0 g and a variance of
σX2 = 0.802 = 0.32 g2 .)
¯ 2
(d) 0.0385 There are 2 approaches that lead to the same result:
i. We know from 23b that T ∼ N (34.0, 1.28) (where T represents the total
amount of protein in 2 sandwiches). So P (T < 32.0) = P (Z < 32.0−34.0
√
1.28
)=
P (Z < −1.768) = 0.0385.
ii. Alternatively, we could note that the total amount of protein will be less
than 32.0 grams if the mean amount in the 2 sandwiches is less than 16.0
grams. P (X¯ < 16.0) = P (Z < 16.0−17.0
√ ) = P (Z < −1.768) = 0.0385.
0.80/ 2
(e) P (X ¯ ≥ 16.5) = P (Z ≥ 16.5−17.0
√ ) = P (Z ≥ −0.884) = 0.812.
0.80/ 2
(f) 18.109 grams. The 97.5th percentile of the standard normal distribution is
approximately 1.96. The 97.5th percentile of the distribution of X
¯ is approx-
imately µ + √n z = 17.0 + √2 · 1.96 = 18.109 grams.
σ 0.80
24. Let X represent the length (in mm) of the right ear of a randomly selected Caucasian
Italian man in this age group. Then X ∼ N (62, 42 ).
(a) P (X ≥ 60) = P (Z ≥ 60−62 4 ) = P (Z ≥ −0.5) = 0.691.

(b) 67.13 mm. (There is a little rounding error if you use a table. The 90th
percentile of the standard normal distribution is approximately 1.28. The 90th
percentile of right ear lengths is approximately µ + σz = 62 + 4 · 1.28 = 67.12
mm.)
(c) X¯ ∼ N (62, 0.16). (The mean ear length is approximately normally distributed
with a mean of µX¯ = 62 mm and a variance of σX̄ 2 = 42 = 0.16 mm2 .)
100
(d) P (X
¯ > 61.8) = P (Z > 61.8−62
√
4/ 100
) = P (Z > −0.5) = 0.691.
imately −1.28. The 10th percentile of the distribution of X
¯ is approximately
µ + √n z = 62 + √100 (−1.28) = 61.49 mm.
σ 4
25. Let X represent the length (in mm) of the carapace of a randomly selected adult
male crayfsh in this lake. Then X ∼ N (48.0, 2.52 ).
(a) P (X < 50.0) = P (Z < 50.0−48.0

2.5 ) = P (Z < 0.8) = 0.788.
(b) X ∼ N (48.0, 0.625). (The mean carapace length is approximately normally
¯
distributed with a mean of µX¯ = 48.0 mm and a variance of σX̄2 = 2.52 = 0.625
10
mm2 .)
(c) P (X¯ < 50.0) = P (Z < 50.0−48.0
√
2.5/ 10
) = P (Z < 2.530) = 0.9943.
(d) P (46.0 < X ¯ < 50.0) = P ( 46.0−48.0
√
2.5/ 10
< Z < 50.0−48.0
√ ) = P (−2.530 < Z <
2.5/ 10
2.530) = 0.9886.
imately 1.645. The 95th percentile of the distribution of X
¯ is approximately
µ + √n z = 48.0 + √10 · 1.645 = 49.30 mm.
σ 2.5
26. Let X represent the weight (in kg) of a randomly selected newborn female African
elephant that is born in captivity. Then X ∼ N (95.1, 13.72 ).
(a) 0.360. P (X > 100.0) = P (Z > 100.0−95.1 13.7 ) = P (Z > 0.358) = 0.360. (There
will be a little rounding error if you use a table: P (Z > 0.36) = 0.3594.)
(b) X¯ ∼ N (95.1, 13.72 ). (The sampling distribution of the mean weight of 3 ran-
3
domly selected newborn female elephants is approximately normal with a mean
2
3 ≈ 62.6 kg .
of 95.1 kg and a variance of 13.7 2
(c) 0.268. P (X¯ > 100.0) = P (Z > 100.0−95.1 √ ) = P (Z > 0.619) = 0.268.
13.7/ 3
(d) 0.196. P (98.0 < X ¯ < 102.0) = P ( 98.0−95.1√
13.7/ 10
< Z < 102.0−95.1
√ ) = P (0.669 <
13.7/ 10
Z < 1.593) = 0.944 − 0.748 = 0.196.
(e) 87.9 kg. Using a table, we would fnd that the 30th percentile of the stan-
dard normal distribution is approximately −0.52. The 30th percentile of birth
weight is approximately µ + σz = 95.1 + 13.7(−0.52) = 87.98 kg. (But there
is a fair bit of rounding error if you use a table. Using software, we can fnd
that the 30th percentile of the standard normal distribution is −0.5244 and
the 30th percentile of birth weights is 87.915 kg.)
(f) 92.83 kg. Using a table, we would fnd that the 30th percentile of the stan-
dard normal distribution is approximately −0.52. The 30th percentile of X ¯ is
approximately µ + √n z = 95.1 + √10 (−0.52) = 92.847 kg. (But there is a fair
σ 13.7
bit of rounding error if you use a table. Using software, we can fnd that the
30th percentile of the standard normal distribution is −0.5244 and the 30th
percentile of the sampling distribution of mean birth weight is 92.828 kg.)
27. Let X represent the weight (in grams) of the heart of a randomly selected adult
Caucasian male. Then X ∼ N (365, 712 ).
(a) 456.0 grams.

The 90th percentile of the standard normal distribution is approximately 1.28.

The 90th percentile of the distribution of heart weight is approximately µ +
σz = 365 + 71 · 1.28 = 455.88. (But there is a fair bit of rounding error
if you use a table. Using software, we can fnd that the 90th percentile of
the standard normal distribution is 1.281552 and the 90th percentile of the
sampling distribution of heart weight is 455.99 g.)
(b) X¯ ∼ N (365, 712 ). (The sample mean is approximately normally distributed
9
2
with a mean of 365 grams and a variance of 719 = 560 grams2 .)
(c) 0.737. P (X
¯ > 350.0) = P (Z > 350.0−365.0
√
71.0/ 9
) = P (Z > −0.634) = 0.737. (There
will be a little rounding error if you use a table: P (Z > −0.63) = 0.7357.)
(d) 0.591. P (340.0 < X ¯ < 380.0) = P ( 340.0−365.0
√
71.0/ 9
< Z < 380.0−365.0
√
71.0/ 9
) = P (−1.056 <
Z < 0.634) = 0.591.
proximately 1.28. The 90th percentile of X ¯ is approximately µ + √σ z =
n
365 + 71 √.0 · 1.28 = 395.29 grams. (But there is a fair bit of rounding error
9
if you use a table. Using software, we can fnd that the 90th percentile of
the standard normal distribution is 1.281552 and the 90th percentile of the
sampling distribution of the mean heart weight is 395.33 grams.)
(f) 0.664. There are 2 approaches that lead to the same result:
i. The sum of the weights of 4 randomly selected adult Caucasian male hearts
is approximately normally distributed with a mean of 4µ = 4×365
P4 = 1460
grams and a variance of 4σ = 4 × 71 = 20164 grams . P ( i=1 Xi >
2 2 2
1400) = P (Z > 1400−1460

√
20164
) = P (Z > −0.423) = 0.664.
ii. Alternatively, we could note that the sum will be greater than 1400 if the
mean of the 4 observations is greater than 1400 ¯
4 = 350. P (X > 350) =
350−365
P (Z > 71/√4 ) = P (Z > −0.423) = 0.664.
28. Let X represent the height (in cm) of a randomly selected Canadian female between
(a) P (X ≥ 170) = P (Z ≥ 170.0−163.3

6.4 ) = P (Z ≥ 1.047) = 0.148. (There will be a
little rounding error if you use a table: P (Z ≥ 1.05) = 0.1469.)
(b) X¯ ∼ N (163.3, 0.4096). (The sample mean is approximately normally dis-
2 2
tributed with a mean of 163.3 cm and a variance of σn = 6.4 100 = 0.4096 cm .)
2
(c) 0.996. P (X¯ < 165.0) = P (Z < 165.0−163.3

√
6.4/ 100
) = P (Z < 2.656) = 0.996.
(d) P (163.0 < X ¯ < 165.0) = P ( 163.0−163.3
√
6.4/ 100
< Z < 165.0−163.3
√
6.4/ 100
) = P (−0.469 < Z <
2.656) = 0.996 − 0.320 = 0.676.
(e) The 2.5th percentile of the standard normal distribution is approximately
−1.96. The 2.5th percentile of the distribution of X
¯ is approximately µ +
10 (−1.96) = 162.05 cm.
σ
√ z = 163.3 + 6.4
n
The 97.5th percentile of the standard normal distribution is approximately
1.96. The 97.5th percentile of the distribution of X

¯ is approximately µ+ √σ z =
n
163.3 + 6.4
10 · 1.96 = 164.55 cm.
29. (a) µX¯ = µ = 71.

√
(b) σX̄ = √σn = √100
5
= 4.472.
(c) X¯ ∼ N (71, 20). (The sample mean is approximately normal with a mean of
2
71 and a variance of σn = 1005 = 20.)
(d) P (X > 75) = P (Z > 10 ) = P (Z > 0.4) = 0.345.
75−71
(e) P (X¯ > 75) = P (Z > 75−71√ ) = P (Z > 0.894) = 0.186. (There will be a little
10/ 5
rounding error if you use a table: P (Z > 0.89) = 0.1867).
(f) From 29d, the probability a randomly selected student scored higher than 75
is 0.345. The probability all 5 randomly selected students scored higher than
75 is approximately 0.3455 = 0.005. (Note that for this calculation to be
reasonable, we require a very large class size.)
30. Let X represent the diameter of a randomly selected shaft produced by this process.
Then X ∼ N (4.10, 0.052 ).
(a) P (X < 4.23) = P (Z < 4.23−4.10

0.05 ) = P (Z < 2.6) = 0.9953.
0.052
(b) X ∼ N (4.10, 16 ). (The sample mean is approximately normally distributed
¯
2
with a mean of 4.10 cm and variance of 0.05
16 = 0.00015625 cm .)
2
2
(c) X
¯ ∼ N (4.10, 0.05 ). (The sample mean is approximately normally distributed
100
2
with a mean of 4.10 cm and variance of 0.05
100 = 0.000025 cm .)
2
(d) P (4.05 < X

¯ < 4.15) = P ( 4.05−4.10
√
0.05/ 5
< Z < 4.15−4.10
√ ) = P (−2.236 < Z <
0.05/ 5
2.236) = 0.975.
(e) P (X
¯ ≥ 4.15) = P (Z ≥ 4.15−4√.10
0.05/ 5
) = P (Z ≥ 2.236) = 0.013.
(f) 4.114 cm. The 90th percentile of the standard normal distribution is ap-
proximately 1.28. The 90th percentile of X¯ is approximately µ + σ z =
n
0.05
4.10 + √ 20
× 1.28 = 4.114 cm.
31. Let X represent the oxygen purity level of a randomly selected container produced
by this company. If the company’s claim is true, X ∼ N (99.50, 0.0482 ).
(a) P (X < 99.43) = P (Z < 99.43−99.50

0.048 ) = P (Z < −1.458) = 0.072.
(b) P (X¯ < 99.45) = P (Z < 99.45−99.50
√
0.048/ 10
) = P (Z < −3.294) = 0.0005.
(c) P (99.45 < X¯ < 99.55) = P ( 99.45−99.50
√
0.048/ 10
< Z < 99.55−99.50
√
0.048/ 10
) = P (−3.294 < Z <
3.294) = 0.9990.
(d) If it is reasonable to view this sample as a random sample of containers from

this company, and the company’s claim is true, the probability of observing a
mean purity level this low or lower is P (X ¯ ≤ 99.00) = P (Z ≤ 99.00−99.50
√
0.048/ 25
)=
P (Z < −52). This probability is very, very near 0. There is (essentially) no
chance of seeing what we observed if the company’s claim is true, and so there
is very strong evidence against the company’s claim.
Chapter 8
Confdence Intervals
J.B.’s strongly suggested exercises: 1, 2, 4, 5, 7, 8, 9, 10, 11, 13, 14, 24, 26, 27, 29, 31, 32
8.1 Introduction
8.2 Interval Estimation of µ when σ is Known
1. (a) z.025 = 1.96 (see Figure 8.1), and the margin of error is 1.96 × √5
22
= 2.089.
(b) z.05 = 1.645, and the margin of error is 1.645 × √1010 = 5.202.
(c) z.005 = 2.576, and the margin of error is 2.576 × √1518 = 9.108.
(d) α = 1 − 0.632 = 0.368. α2 = 0.184. z.184 = 0.90, and the margin of error is
0.90 × √1518 = 3.182.
(e) α = 1 − 0.212 = 0.788. α2 = 0.394. z.394 = 0.27, and the margin of error is
0.27 × √3550 = 1.336.
2. (a) 77.4% (since the area between −1.21 and 1.21 under the standard normal curve
is 0.774 (see Figure 8.2).
(b) 82.3% (since the area between −1.35 and 1.35 under the standard normal curve
is 0.823).
(c) 88.4% (since the area between −1.57 and 1.57 under the standard normal curve
is 0.884).
(d) 98.6% (since the area between −2.45 and 2.45 under the standard normal curve
is 0.986).
(e) 99.8% (since the area between −3.15 and 3.15 under the standard normal curve
is 0.998).
105
Balka ISE 1.09 8.2. INTERVAL ESTIMATION OF µ WHEN σ IS KNOWN 106
0.05 0.05
= 0.025 = 0.025
2 0.95 2
− z 0.025 0 z 0.025
The area to the right of this value is 0.025
The area to the left of this value is 0.975
(a) We need to fnd z0.025 . (b) R tells us that z0.025 =

1.959964 (rounded to 6 decimal
places).
(c) The standard normal table tells us that (d) The standard normal table tells us that
z0.025 = 1.96 (rounded to 2 decimal places). −z0.025 = −1.96 (rounded to 2 decimal
places).
Figure 8.1: Finding the appropriate z value for a 95% interval.
0.774
0.113 0.113
-4 -3 -2 -1 0 1 2 3 4
-1.21 1.21
Figure 8.2: The confdence level is 0.774 (77.4%) if zα/2 = 1.21.
3. (a) In statistical inference, we often construct confdence intervals for statistics.

False. We construct confdence intervals for parameters, and never for statis-
tics.
(b) A confdence interval is a range of plausible values for a parameter. True.
(c) The confdence level of an interval is determined by sample data. False. We
choose the confdence level.
(d) The point estimate of µ lies at the midpoint of the confdence interval for µ.
True. The point estimate of µ is the value of X, ¯ which always falls at the
midpoint of the interval.
(e) The confdence interval procedures of this section assume a normally dis-
tributed population, but this assumption becomes less important as the sample
size increases. True. (Because of the central limit theorem.)
Balka ISE 1.09 8.3. CONFIDENCE INTERVALS FOR µ WHEN σ IS UNKNOWN 107
8.2.1 Interpretation of the Interval
4. (a) 25.7 ± 1.96 √6.0

200
, which works out to 25.7 ± 0.832 or (24.87, 26.53).
(b) We can be 95% confdent that the true mean BMI for female students at this
university lies between 24.87 kg/m2 and 26.53 kg/m2 .
(c) 25.7 ± 2.576 √6.0
200
(d) We can be 99% confdent that the true mean BMI for female students at this
university lies between 24.61 kg/m2 and 26.79 kg/m2 .
5. We can be 95% confdent that the mean right ear length of all Caucasian Italian
females between the ages of 18 and 30 lies between 60.91 mm and 62.89 mm. Or:
We can be 95% confdent that the true mean right ear length of Caucasian Italian
females between the ages of 18 and 30 lies between 60.91 mm and 62.89 mm.
(It can also be said that in repeated sampling, 95% of the 95% confdence intervals
calculated in this manner would capture the true mean right ear length of Caucasian
Italian females between the ages of 18 and 30.)
6. Statement (c) is a correct interpretation of the interval, and the rest of the state-
ments are false.
8.2.2 What Factors A˙ect the Margin of Error?
7. (a) An increase in the sample size decreases the margin of error.

(b) An increases in the variance increases the margin of error.
(c) An increase in X
¯ has no e˙ect on the margin of error of the confdence interval.
(d) An increase in the confdence level increases zα/2 , which in turn increases the
margin of error.
8.2.3 An Example
8.3 Confdence Intervals for µ When σ is Unknown
8.3.1 Introduction
8. (a) With 13 degrees of freedom, t.10 = 1.350 (see Figure 8.3). SE(X
¯) = √10
14
=
2.673. M OE = 1.350 × 2.673 = 3.609.
t distribution with n − 1 = 13
degrees of freedom
0.20 0.20
= 0.10 = 0.10
2 0.80 2
− t 0.10 0 t 0.10
The area to the right of this value is 0.10
(a) We need to fnd t0.10 .
(b) R tells us that (with 13 DF) (c) The t table tells us that (with 13 DF) t0.10 = 1.350
t0.10 = 1.350171 (rounded to 6 (rounded to 3 decimal places).
decimal places).
Figure 8.3: Finding the appropriate t value for an 80% interval.
(b) With 21 degrees of freedom, t.10 = 1.323. SE(X ¯ ) = √10 = 2.132. M OE

22
=
1.323 × 2.132 = 2.821.
(c) With 7 degrees of freedom, t.05 = 1.895. SE(X¯ ) = 2500
√ = 883.8835. M OE
8
=
1.895 × 883.8835 = 1674.606.
(d) With 7 degrees of freedom, t.025 = 2.365. SE(X
¯ ) = 2500
√ = 883.8835. M OE
8
=
2.365 × 883.8835 = 2090.031.
(e) With 7 degrees of freedom, t.005 = 3.499. SE(X ¯) = √ 3.2
8
= 1.131. M OE =
3.499 × 1.131 = 3.959.
9. (a) The variance of the t distribution is greater than the variance of the standard
normal distribution. True.
(b) The mean of the t distribution is equal to the mean of the standard normal
distribution. True. (Both distributions have a mean of 0.)
(c) The t distribution has more area in the tails and a lower peak than the standard
(d) As the degrees of freedom increase, the t distribution tends toward the standard
(e) The t distribution is equivalent to the standard normal distribution if the
degrees of freedom are at least 30. False. Although the t distribution tends
toward the standard normal distribution as the degrees of freedom increase,
there are still meaningful di˙erences at 30 degrees of freedom and above.
10. The interval found using zα/2 would be narrower. Although narrower intervals are
preferred, had we made this mistake we would be reporting incorrect results. (But
if the sample size is large the mistake would have only a very small e˙ect.)
8.3.2 Examples
11. (a) No, there is no indication of a violation of the normality assumption. The
boxplot looks roughly symmetric, and there is no systematic curvature or
outliers evident in the normal QQ plot.
(b) The sample mean is the point estimate of the population mean: X ¯ = 50.91
kgf. SE(X ¯ ) = √ = √ = 1.3683.
s
n
7.74
32
(c) Using software or a t table, we can fnd that with n − 1 = 32 − 1 = 31 degrees
of freedom, t0.025 = 2.0395. The 95% interval is given by:
¯ ± t0.025 SE(X
X ¯)
50.91 ± 2.0395 × 1.3683
50.91 ± 2.791
which works out to (48.12, 53.70).

(d) We can be 95% confdent that the population mean grip strength of male
students at this German university lies between 48.12 kgf and 53.70 kgf.
The population is made up of all male students at this university at the time
of the study. It wasn’t clearly stated in the study how the volunteers were
recruited, and the method of recruited could introduce bias. As an extreme
example, if the volunteers were chosen from varsity sports teams, then that
might severely bias the results (since athletes would likely tend to have a
greater grip strength than non-athletes). We would need to know more infor-
mation about the volunteer recruitment protocol to make an informed decision
about whether this is a concern. In any event, the fact that the students vol-
unteered for the study could possibly introduce bias, as volunteers might di˙er
in a meaningful way from the general student population.
8.3.3 Assumptions of the One-Sample t Procedures
12. If the population is not normally distributed, then the true coverage probability
of the confdence interval method can be very di˙erent from the nominal (stated)
confdence level. In short, the reported results might be very misleading.
Balka ISE 1.09 8.4. DETERMINING THE MINIMUM SAMPLE SIZE N 110
13. Strong skewness and outliers pose the biggest problems for the t procedure (and
they are especially problematic if the sample size is small). If a distribution is
roughly symmetric and unimodal (the distribution is “mound shaped”), then this
type of violation of the normality assumption is not a big problem, even for small
sample sizes. The e˙ect of any violation of the normality assumption becomes less
of a problem as the sample size increases.
8.4 Determining the Minimum Sample Size n
1.645 × 10 2
14. (a) n ≥ ( ) = 10.8. We need a sample size of at least 11.
5
1.96 × 10 2
(b) n≥( ) = 15.4. We need a sample size of at least 16. (Note that
5
in minimum sample size calculations, we round up to the next largest whole
number.)
1.96 × 30 2
(c) n≥( ) = 138.3. We need a sample size of at least 139.
5
2.576 × 10 2
(d) n≥( ) = 66357.8. We need a sample size of at least 66,358. (Note
0.10
that the required sample size might be far larger than what is possible in
practice. We may need to adjust our plans or abandon them if the calculation
tells us that our wants are not practical.)
1.96 × 4 2
(e) n≥( ) = 61.47. We need a sample size of at least 62.
1
15. σX̄ = √σ
n
= √10
16
= 2.5.
16. SE(X
¯) = √s
n
= 2.0
√
16
= 0.5.
17. The population is assumed to be normally distributed, and the standard deviation
is based on sample data, so to calculate the intervals we will use: X
¯ ± tα/2 √s .
n
(a) For the sample 3, 5, 12, X¯ = 6.667 and s = 4.725816. There are n − 1 =
3 − 1 = 2 degrees of freedom, and with 2 degrees of freedom, t.025 = 4.303.
The 95% interval is given by:
√ , which works out to 6.667 ± 11.74.
6.667 ± 4.303 × 4.7258
3
(b) 9.0 ± 4.303 × 13.5277

√
3
, which works out to 9.0 ± 33.61.
(c) 2027.0 ± 3.182 × 20√
.5589
4
, which works out to 2027.0 ± 32.71.
(d) −47.2 ± 2.776 × 3.633180
√
5
, which works out to −47.2 ± 4.51.
8.5.2 Concepts
18. D is the best interpretation. (The rest of the statements are false.)
19. (a) The number of intervals that contain µ will have a binomial distribution with
n = 100 and p = 0.95.
(b) P (X ≥ 1) = 1 − P (X = 0) = 1 − 0.95100 = 0.994.
20. (a) The central limit theorem tells us that the sampling distribution of X
¯ is ap-
proximately normal for large sample sizes, so in most cases when n = 100,
methods based on the normal distribution are reasonable (even when there is
some skewness). But there may be better methods of analysis, such as carrying
out an inference procedure on a transformation of the data.
(b) σX̄ = √1.0
100
= 0.10.
(c) We will use X¯ ± zα/2 √σ to calculate the interval. Using a standard normal
n
table or software, we can fnd the appropriate z value: z0.04 = 1.75. (The area
to the right of 1.75 under the standard normal curve is 0.04, so using 1.75
in the interval formula will result in a 92% confdence level.) The interval is:
1.4 ± 1.75 × 0.10, which works out to 1.4 ± 0.175.
(d) The true value of µ is unknown. (We are constructing a confdence interval in
order to estimate it.)
(e) The parameter µ represents the true mean time to death for this type of rat
after being infected with this virus (under the conditions of this experiment).
21. True. (This can be seen in the t table. For example, t0.025 → 1.96 as the degrees of
freedom increase.)
Sample Size (n) Degrees of Freedom t.025

2 1 12.706
6 5 2.571
11 10 2.228
31 30 2.042
51 50 2.009
101 100 1.984
∞ ∞ 1.960
¯ 1 represent the mean of Sample 1 and X

22. If we let X ¯ 2 represent the mean of Sample
¯ 1 − 1.96 √σ < X
2, then we need to fnd P (X ¯ 1 + 1.96 √σ ).
¯2 < X
n n
¯ 1 − 1.96 √σ < X
P (X ¯ 1 + 1.96 √σ ) = P (−1.96 √σ < X
¯2 < X ¯ 1 < 1.96 √σ )
¯2 − X
n n n n
r
σ2 σ2 √ σ
Note that X ¯ 1 has a mean of µ − µ = 0 and a standard deviation of
¯2 − X + = 2√ .
n n n
Standardizing:
σ σ
−1.96 √ − 0 1.96 √ − 0
¯ 1 − 1.96 √σ < X
P (X ¯ 1 + 1.96 √σ ) = P (
¯2 < X √ σ
n
<Z< √ σ
n
)
n n 2√ 2√
n n
1.96 1.96
= P (− √ < Z < √ )
2 2
= 0.834
The probability that the sample mean of one sample falls within the 95% confdence
interval for µ found from another sample of the same size is only 0.834.
23. We are given that θ̂ = 2.00 and σθˆ = 3.00 (the standard deviation of the sampling
distribution of θ̂ is 3.00). Since the estimator θ̂ is normally distributed, and its
standard deviation is known, the margin of error is zα/2 × σθ̂ = 1.96 × 3 = 5.88,
and the 95% interval for θ is 2.00 ± 5.88.
24. Test your conceptual understanding: Which of the following statements are true?
You should be able to explain why a statement is true or why a statement is false.
(a) The standard error of a statistic decreases as the sample size decreases. False.
The standard error increases as the sample size decreases.
(b) The one-sample t procedures are robust to violations of the normality as-
sumption. True. The t procedures work quite well in a variety of non-normal
situations.
(c) The t procedures do not perform well when there is strong skewness or outliers
in the data, especially for small sample sizes. True.
(d) The width of a confdence interval for µ decreases as the sample size decreases.
False. The width of a confdence interval increases as the sample size decreases.
(e) The width of a confdence interval for µ decreases as the variance increases.
False. The width of a confdence interval increases as the variance increases.
25. Test your conceptual understanding: Which of the following statements are true?
You should be able to explain why a statement is true or why a statement is false.
(a) In practical problems, σ is usually known and thus we usually use z procedures
instead of t procedures. False. The vast majority of the time σ is an unknown
quantity.
(b) Standard errors are the estimated standard deviations of parameters in re-
peated sampling. False. A standard error is the estimated standard deviation
of a statistic in repeated sampling.
(c) If the assumptions of a procedure used to calculate a 90% confdence interval
are violated, the true confdence level of the interval may be very di˙erent
from 90%. True. If the assumptions are false the true confdence level may be
very di˙erent from what is stated.
(d) All else being equal, the width of a confdence interval decreases as the con-
fdence level increases. False. If we want greater confdence that the interval
captures the parameter, we must cover more ground and have a wider interval.
(e) All else being equal, the margin of error decreases as the confdence level in-
creases. False. The margin of error increases as the confdence level increases.
(f) All else being equal, the standard error of the sample mean decreases as the
confdence level of the interval increases. False. SE(X ¯ ) = √s , which does not
n
involve the confdence level.
(g) All else being equal, the confdence level of an interval increases as the sample
size increases. False. The confdence level (1 − α) and sample size (n) are not
related.
8.5.3 Applications
26. (a) Both the boxplot and the normal QQ plot show indicate right-skewness (which
is typical of BMI data). This skewness is not a serious problem in this scenario,
as the skewness is not very strong, and the sample size is fairly large (94). We
can use the t procedures. (We will use inference procedures on the mean in
this question, which is perfectly valid. But because of the skewness, in real
world scenarios the median BMI is often reported instead of the mean BMI.)
(b) The sample mean is the point estimate of the population mean: X ¯ = 23.3.
SE(X ¯ ) = √ = √ = 0.5570.
s 5.4
n 94
(c) Using software, we can fnd that with 93 degrees of freedom, t0.025 = 1.9858.
(If we must use a table, we may have to look up 90 df and fnd the t value of
1.987.)
¯ ± t0.025 SE(X
X ¯)
23.3 ± 1.9858 × 0.5570
23.3 ± 1.106
which works out to (22.2, 24.4). (Had we used the t value of 1.987 from the
table, we would have gotten the same answer to 1 decimal place.)
(d) The interpretation of the interval is a little tricky here, due to the complexities
of the sampling design. The sample is not truly a simple random sample from
any population, so one could argue that we don’t really have a confdence
interval for anything. But sampling designs like this are very common in
practice (when human consent is required, there are always going to be biases
present). So I’ll give a not-quite-true-but-somewhat-reasonable interpretation,
then discuss the biases:
We can be 95% confdent that µ, the true mean BMI for 14-year old boys in
this area of rural Texas, lies between 22.2 and 24.4 kg/m2 .
The sample is drawn from a school district in north central Texas, so, loosely,
the population is all 14 year-old boys in this area of north central Texas.
However, there are some sampling biases that make this interpretation not
quite true:
• The sample was drawn from only 6 schools in that area. Students in these
schools may di˙er in a meaningful way from students at other schools in
the area (and from children that are not enrolled at a school).
• Parents needed to provide informed consent in order for students to par-
ticipate. Students whose parents provide informed consent may di˙er in
a meaningful way from those whose parents do not.
• Students needed to provide self-assent (the student had to agree to be
part of the study). Students who agree to participate may di˙er in a
meaningful way from those who do not.
• Students who participated were given a t-shirt. This could introduce bias,
as it is conceivable that there is a relationship between desire for the t-shirt
and BMI.
27. (a) No, there is no indication of a violation of the normality assumption. The box-
plot looks symmetric, and there is no systematic curvature or outliers evident
in the normal QQ plot. A minor violation of the normality assumption would
not be a big deal in this scenario, as the sample size is not small (n = 51). We
can use the t procedures.
mm/s. SE(X ¯ ) = √ = √ = 0.0868.
s
n
0.62
51
of freedom, t0.025 = 2.009. (If we must use a table, we may have to look up 90
df and fnd a t value of 1.987.)

¯ ± t0.025 SE(X
X ¯)
1.65 ± 2.009 × 0.0868
1.65 ± 0.174

(d) We can be 95% confdent that the population mean speed of tandem running
lies between 1.48 and 1.82 mm/s.
It wasn’t clearly stated how the tandem running pairs were chosen, but it is
reasonable to think that these results apply to Temnothorax albipennis ants
(with a young experienced ant leading a young inexperienced ant) under the
conditions of the study. The true mean may be di˙erent if the conditions
change, or if a di˙erent type of ant is studied.
28. (a) We need a simple random sample from a normally distributed population. The
skeletons in the Athens Collection do not truly represent a simple random from
any population, and to come up with an informed opinion we would need to
look in greater detail at how this sample was gathered. However, in a loose
way, we can think of this sample of skeletons as a random sample of skeletons
from men who lived near Athens near the end of the 20th century.
As far as normality is concerned, we don’t have the actual data to investigate
that, but it is likely that the diameters are approximately normally distributed.
The sample size (n = 24) is not very large, so normality is a concern. If we
had access to the data, we should investigate the normality assumption with
a normal QQ plot.
mm/s. SE(X ) = √n = √24 = 0.1286.
¯ s 0.63

of freedom, t0.025 = 2.069.
¯ ± t0.025 SE(X
X ¯)
10.38 ± 2.069 × 0.1286
10.38 ± 0.266

(d) We can be 95% confdent that the true mean crown diameter of upper M1 mo-
lars lies between 10.11 and 10.65 mm. The population to which this conclusion
applies depends on how this sample of skeletons was drawn. In a loose way,
we can think of the population as adult males who lived near Athens near the
end of the 20th century, but we would need to investigate the sampling design
in greater detail to reach an informed opinion.
29. (a) We would need the sample to be a simple random sample from a normally dis-
tributed population. The sample of female blowfies is not truly a simple ran-
dom sample of blowfies from the area (they were captured in traps, and there
could conceivably be factors infuencing the types of fy that get trapped). We
also need the number of larvae in fies of this type to be normally distributed.
(But since the sample size is not small (n = 49), the normality assumption is
not very important.)
(b) The largest observation is roughly 1.5 standard deviations greater than the
mean, and the smallest is roughly 2 standard deviations below the mean. So
there are not any extreme values (outliers) that would a˙ect the t procedures.
This information makes the use of the t procedures more reasonable.
(c) The sample mean is the point estimate of the population mean: X ¯ = 33.4
mm/s. SE(X ) = √n = √49 = 1.0.
¯ s 7.0
(d) Using software, we can fnd that with 48 degrees of freedom, t0.025 = 2.0106.
2.014.)
¯ ± t0.025 SE(X
X ¯)
33.4 ± 2.0106 × 1.0
33.4 ± 2.01
table, we would have gotten the same answer to 1 decimal place.)
(e) We can be 95% confdent that the true mean number of larvae carried by
female C. varifrons blowfies in this feld lies between 31.4 and 35.4.
There are possible biases introduced by the sampling mechanism. Flies with
certain characteristics (old or large fies, for example) may be more or less likely
to be captured. These characteristics may also be related to the number of
larvae that they carry. (For example, it is conceivable that older fies would be
less likely to be captured. If older fies also tend to carry fewer larvae, then the
interval would tend to be too large (since older fies would be underrepresented
in the sample).)
Also, the interval only applies to this type of fy under the conditions of the
study. If we were to use this interval to estimate the mean in other situations
(di˙erent locations, temperatures, humidity, or for a di˙erent type of fy), then
the interval may be very far o˙ and very misleading.
30. (a) We would need the sample of fies to behave like a simple random sample
of C. varifrons blowfies. We would also need the rate of deposition to be
approximately normally distributed. We would need to investigate the study
in greater detail to form an opinion about the sample of fies that they used.
Since the sample size is not small (n = 42), it would be reasonable to use the
t procedures, unless the data showed strong skewness or outliers.
mm/s. SE(X ¯ ) = √ = √ = 0.049.
s
n
0.32
42
(c) Using software, we can fnd that with 41 degrees of freedom, t0.025 = 2.0195.
2.021.)
¯ ± t0.025 SE(X
X ¯)
0.46 ± 2.0195 × 0.049
0.46 ± 0.100
table, we would have gotten the same answer to 2 decimal places.)
(d) We can be 95% confdent that the population mean rate of larviposition lies
between 0.36 and 0.46 larvae per second.
This interval applies to this type of blowfy under the conditions of this ex-
periment. If the conditions of the experiment were to change, then the true
mean rate of larviposition may change. Biases may have been introduced by
the choice of C. varifrons blowfy used in this experiment. (It is not clear how
the C. varifrons blowfies used in the experiment were chosen.)
31. (a) To draw conclusions about all hash browns made in U.S. locations of this fast-
food chain, we need the hash browns in the sample to be a random sample
from the population of hash browns in these U.S. locations. Little is said in the
U.S. Nutrient Database about how the sample was collected, but the people
that compile this database are professionals, and it is unlikely that this sample
has obvious biases. (Discussion of possible bias will be discussed below in the
answer to Question 31d.)
For the t procedure to be reasonable, we need the amount of sodium in hash
browns of this type to be normally distributed. This is especially true here,
since the sample size is so small (n = 6). It is very likely that the amount
of sodium in hash browns of this type is approximately normally distributed,
but we cannot tell from the given information.
(b) The sample mean is the point estimate of the population mean: X ¯ = 324.8mg.
SE(X ¯ ) = √ = √ = 16.330.
s 40.0
n 6
(c) Using software or a t table, we can fnd that with 5 degrees of freedom, t0.025 =
2.571.
¯ ± t0.025 SE(X
X ¯)
324.8 ± 2.571 × 16.330
324.8 ± 41.98

(d) Assuming normality, and assuming that the sample of hash browns can be
thought of as a random sample of hash browns from U.S. locations of this
chain:
We can be 95% confdent that the true mean sodium content in servings of
hash browns from this fast-food chain lies between 282.82 mg and 366.78 mg.
There are many sources of possible bias, depending on how the sample was
drawn. The distribution of sodium may di˙er from location to location, as
preparation methods might di˙er slightly between locations. (The fast-food
chain in question is a very large chain with common preparation methods, so
this is unlikely to be major infuence, but there still could be an e˙ect.) If the
sample was drawn from an individual location, and we were trying to draw
inferences about all U.S. restaurants, this could easily result in bias. If the
sample of 6 hash browns was drawn from a one batch of hash browns, then it
would be very unwise to try to draw conclusions about all hash browns from
this type of sample.
(e) Yes, the company’s claimed value of 310 mg is a plausible value of the popu-
lation mean, since it falls within the 95% confdence interval.
32. (a) To draw conclusions about black women in the Manchester area, we need the
women in the sample to be a random sample from the population of black
women in the Manchester area. The sample was drawn from women attending
a sexual health clinic, and these women might di˙er in a meaningful way from
the general population of black women in the area. (Sources of possible bias
will be discussed below in the answer to Question 32d.)
For the t procedure to be reasonable, we also need the 2D:4D ratios to be
normally distributed. The article stated that the distribution was close to
normal, so the normality assumption is reasonable. The sample size is 46,
which is not small, so the t procedures would still be reasonable even if there
were a little skewness or other issues with mild non-normality.
(b) The sample mean is the point estimate of the population mean: X ¯ = 0.963.
SE(X ¯ ) = √ = √ = 0.0050.
s 0.034
n 46
(c) Using software or a t table, we can fnd that with 45 degrees of freedom,
t0.025 = 2.014.
¯ ± t0.025 SE(X
X ¯)
0.963 ± 2.014 × 0.0050
0.963 ± 0.010

(d) We can be 95% confdent that the true mean 2D:4D ratio (of the left hand)
of black women attending this sexual health clinic in Manchester lies between
0.953 and 0.973.
There are many sources of possible bias, depending on the conclusion one
wishes to make. Some things to consider:
• The sample of women was drawn from black women attending a sexual
health clinic in Manchester, UK. It is reasonable to think that the conclu-
sions apply to black women who attend this clinic.
• It is likely reasonable to think that these results apply to all black women
in the Manchester, UK area, but we need to be cautious with a conclusion
like this, since black women attending this sexual health clinic might di˙er
in a meaningful way from the general population of black women in the
area.
• Although these results give some indication of the 2D:4D ratio for black
women in general, generalizing beyond the area from which the sample
was drawn is always dubious and may lead to misleading conclusions.
• As is often the case involving studies with human participants, individuals
were included in this study only if they provided written consent. Women
who consent to studies like this might di˙er in a meaningful way from
those who do not.
33. (a) We would need these intimate partner homicides to be a simple random sam-
ple from a (theoretical) population of intimate partner homicides. We also
need the ages of the perpetrators of intimate partner homicide to be normally
distributed. These assumptions will not be perfectly justifed in this case.
There are some sources of possible bias (discussed below in the answer to
Question 33d). The ages will not be perfectly normally distributed, but the
sample size is not too small (n = 36), so unless there is strong skewness or
outliers, the t procedure should work reasonably well.
years. SE(X ¯ ) = √ = √ = 2.583.
s
n
15.5
36
(c) Using software or a t table, we can fnd that with 41 degrees of freedom,
t0.025 = 2.030.
¯ ± t0.025 SE(X
X ¯)
43.8 ± 2.030 × 2.583
43.8 ± 5.24

(d) We can be 95% confdent that the true mean age of perpetrators of intimate
partner homicides in Denmark during the time period 1983–2007 lies between
38.6 to 49.0 years.
One source of bias is that the sample of homicides included only those scenarios
that were determined to be intimate partner homicides. (In reality, not all
homicides are correctly labelled as such, and not all intimate partner homicide
perpetrators are properly labelled as such.) It is conceivable that the age of
perpetrators in which the homicide is not classifed properly tends to be older
or younger than the age of perpetrators in homicides classifed as intimate
partner homicides. So the interval applies only to the population of scenarios in
which the death was labelled as an intimate partner homicide. This population
would di˙er (in a small but meaningful way) from the population of scenarios
that were truly intimate partner homicides.
34. (a) There are no obvious problems here. The points fall close to a straight line,
with no obvious curvature or other problems (indicating that the data is ap-
proximately normally distributed). It is reasonable to assume normality and
use the t procedure.
(b) The standard deviation is based on sample data, so the interval is found using
the formula X¯ ± tα/2 √s . With 15 − 1 = 14 degrees of freedom, t.025 = 2.145.
n
The endpoints of the interval are: 295.6 ± 2.145 × √ 10.2
15
, which works out to
295.6 ± 5.65 or (289.95, 301.25).
(c) The second statement (ii) is correct and the rest of the statements are false.
35. Of the given options, D is the best one. (The rest of the statements are false.)
36. We can be 95% confdent that the true mean wait time for all calls to this call
centre lies between 2.88 and 3.12 minutes.
37. (a) Plotting a normal quantile-quantile plot can give a good indication of whether
or not the data is approximately normal.
(b) The interval is 6.8±2.365× 0√.32
8
, which works out to 6.8±0.2676, or (6.53, 7.07).
(c) We can be 95% confdent that the true mean fuel consumption of this type of
car under these conditions lies between 6.53 and 7.07 l/100k.
38. (a) The value of X

¯ is the point estimate of µ. Here, X
¯ = 29.3.
(b) SE(X ) = √n = √26 = 2.3338.
¯ s 11.9
(c) 29.3 ± 2.060 × 2.3338, which works out to 29.3 ± 4.81 or (24.49, 34.11).
(d) 1.708 × √
11.9
26
= 3.986.
(e) 29.3 ± 3.986 or (25.31, 33.29).
(f) We can be 90% confdent that the true mean number of facial tissues used per
day by this type of cold su˙erer under these conditions lies between 25.31 and
33.29.
39. We can be 95% confdent that the true mean melting temperature of this new type
of oil lies between 34.34 and 34.71 degrees Celsius.
40. (a) 3.2 ± 2.262 × 0.6

√
10
(b) We can be 95% confdent that the true mean cadmium concentration in mush-
rooms of this type in this area lies between 2.77 and 3.63 mg/kg of dry weight.
(c) We should never simply assume the data is normally distributed; this assump-
tion should be investigated. We can investigate the normality assumption
with a normal quantile-quantile plot. If the cadmium concentrations are not
normally distributed, then what we are calling a 95% confdence interval may
have a true coverage probability that is far di˙erent from 95%. (Our stated
confdence level may be very di˙erent from the true confdence level.)
41. (a) 9.2 ± 2.110 × 0.38

√
18
(b) The frst statement is true and the rest are false.
42. (a) 61.1. (The value of X ¯ is a point estimate of µ.)

(b) 61.8−60.4
2 = 0.7. (This is half the width of the interval.)
(c) The 95% margin of error was found to be 0.7. Since the 95% margin of error
is 1.96σX̄ , this implies σX¯ = 1.96
0.7
= 0.3571. To calculate the 93% interval, we
frst need to fnd the appropriate zα/2 value: z0.035 = 1.81 (the area to the
right of 1.81 under the standard normal curve is 0.035). The 93% interval is
given by 61.1 ± 1.81 × 0.3571, which works out to 61.1 ± 0.646.
Chapter 9
Hypothesis Testing
J.B.’s strongly suggested exercises: 1, 5, 6, 10, 12, 13, 15, 19, 21, 33, 37, 38, 41, 48, 59,
61, 62, 63, 64, 65, 66
9.1 Introduction
1. (a) H0 : µM = µF (the true mean resting heart rate is the same for male and female Olympic athletes).
Ha : µM = 6 µF (the true mean resting heart rate is di˙erent between male and female Olympic athlet
(b) H0 : p1 = p2 = p3 = p4 (the six month survival rate is the same for all four treatments).
Ha : pi 6= pj for at least one i, j combination (the six month survival rates are not all equal).
(c) H0 : µ = 4.0 (the true mean saturated fat content is 4.0 grams).
Ha : µ > 4.0 (the true mean saturated fat content is greater than 4.0 grams).
It would also be acceptable to express the hypotheses this way:
H0 : µ ≤ 4.0 (the true mean saturated fat content is no more than 4.0 grams).
Ha : µ > 4.0 (the true mean saturated fat content is greater than 4.0 grams).
But when carrying out the test we will need to test a specifc value, so in this
text we will always express the null hypothesis as an equality.
(d) H0 : p = 0.98 (the true proportion of satisfed customers is 0.98).
Ha : p < 0.98 (the true proportion of satisfed customers is less than 0.98).
(e) H0 : σ12 = σ22 = σ32 (the true variances are all equal).
Ha : σi2 6= σj2 for at least one i, j combination (the true variances are not all equal).
123
Balka ISE 1.09 9.2. THE LOGIC OF HYPOTHESIS TESTING 124
9.2 The Logic of Hypothesis Testing
9.3 Hypothesis Tests for µ when σ is Known
9.3.1 Constructing Appropriate Hypotheses
2. We would never test H0 : X ¯ = 2. (It is not an appropriate hypothesis, since

hypotheses never involve statistics.) We test hypotheses about parameters (µ, for
example).
9.3.2 The Test Statistic
3. In this question we are sampling from a normally distributed population where

the population standard deviation σ is known, so we should use the Z statistic:
Z = X̄−µ
√0.
σ/ n
(a) Z = 15.5−20
√
5/ 10
= −2.85.
(b) Z = 15.5−20
√
5/ 40
= −5.69.
(c) Z = 35.5−30
√
5/ 20
= 4.91.
(d) Z = 65.5−30
√
5/ 20
= 31.75.
¯ 12 X̄ −µ
4. X−√
6/ 75
is the Z statistic: Z = σ/ √ . When we are sampling from a normally dis-
n
tributed population and the null hypothesis is true, the Z statistic has the standard
normal distribution.
9.3.3 The Rejection Region Approach to Hypothesis Testing
5. (a) Reject H0 if |Z| ≥ 1.645.

(b) Reject H0 if Z ≤ −1.28.
(c) Reject H0 if Z ≥ 1.28.
(d) Reject H0 if |Z| ≥ 2.88.
(e) Reject H0 if Z ≤ −2.65.
(f) Reject H0 if Z ≥ 2.65.
6. (a) The signifcance level of a test is the probability of rejecting the null hypothesis,
given the null hypothesis is true. True.
Balka ISE 1.09 9.3. HYPOTHESIS TESTS FOR µ WHEN σ IS KNOWN 125
(b) If we reject the null hypothesis, then we know that the null hypothesis is false.
False. There is an important di˙erence between the result of the test and the
(unknown) underlying reality. Whatever the result of the hypothesis test, we
still do not know the underlying reality.
(c) If we reject the null hypothesis, then we know that the alternative hypothesis
is false. False.
(d) If we do not reject the null hypothesis, then we know that the null hypothesis
is true. False.
(e) If we do not reject the null hypothesis, then we know that the alternative
hypothesis is true. False.
9.3.4 P-values
7. (a) p-value = 0.0058 (twice the area to the right of 2.76 under the standard normal
curve).
(b) p-value = 0.9971 (the area to the left of 2.76 under the standard normal curve).
(c) p-value = 0.0029 (the area to the right of 2.76 under the standard normal
curve).
(d) p-value = 0.0818 (twice the area to the left of −1.74 under the standard normal
curve).
(e) p-value = 0.0409 (the area to the left of −1.74 under the standard normal
curve).
(f) p-value = 0.9591 (the area to the right of −1.74 under the standard normal
curve).
8. (a) Reject H0 , since p-value < α.

(b) Reject H0 , since p-value < α.
(c) Do not reject H0 , since p-value > α.
(d) Do not reject H0 , since p-value > α.
(e) Reject H0 , since p-value < α.
9. A is correct, and the rest of the statements are false. The p-value is the probability,
given the null hypothesis is true, of obtaining a test statistic with at least as much
evidence against the null hypothesis as what was observed in the sample.
10. (a) A p-value is a probability. True.

(b) We reject the null hypothesis when the p-value is less than or equal to the
signifcance level. True.
(c) The p-value of a test is the probability, given H0 is true, of obtaining the
observed value of the test statistic or a value with even greater evidence against
H0 and in favour of Ha . True.
(d) The p-value of a test depends on sample data and on the null and alternative
hypotheses. True.
(e) The p-value of a two-sided test will be greater than 1 if the null hypothesis is
true. False. A p-value is a probability and as such cannot be greater than 1.
11. (a) The p-value of a test will be less than 0.05 if the null hypothesis is false.
False. The p-value can be any value between 0 and 1, whether or not the null
hypothesis is true. Keep in mind that there is a fundamental di˙erence between
the conclusion from a test and the (unknown) underlying reality.
(b) The p-value of a test is the probability that the null hypothesis is true. False.
(c) If the p-value of a hypothesis test with a two-sided alternative is equal to
exactly 1, then we can be certain the null hypothesis is true. False.
(d) If the p-value of a hypothesis test with a two-sided alternative is equal to
exactly 1, then we can be certain the null hypothesis is false. False.
(e) If the p-value of a test is less than 0 that means there is extremely strong
evidence against the null hypothesis. False. A p-value is a probability and as
such it cannot be less than 0. If we fnd a negative p-value, that means we
made a mistake.
9.4 Examples
12. (a) H0 : µ = 3.0 mg/kg dw. (The true mean cadmium concentration in caps of
Boletus edulis in this area is 3.0 mg/kg of dry weight. (The true mean cadmium
concentration is at the tolerable value.).)
Ha : µ > 3.0 mg/kg dw. (The true mean cadmium concentration in caps of
Boletus edulis in this area is greater than 3.0 mg/kg of dry weight. (The true
mean cadmium concentration exceeds the tolerable value.).)
It would also be acceptable to express the hypotheses this way:
H0 : µ ≤ 3.0. (The true mean cadmium concentration in caps of Boletus edulis
in this area is no more than 3.0 mg/kg of dry weight. (The true mean cadmium
concentration is no more than the tolerable value.).)
Ha : µ > 3.0 (The true mean cadmium concentration in caps of Boletus edulis
in this area is greater than 3.0 mg/kg of dry weight. (The true mean cadmium
concentration exceeds the tolerable value.).)
In some ways this set of hypotheses makes more sense, and is more ftting with
the spirit of the test. But when carrying out the test we will need to test a
specifc value, so in this text we will always express the null hypothesis as an
equality.
¯ ¯
(b) Z = X−µσ
0
= X−µ √ 0 = 9.26−3
σ/ n
√
5.00/ 29
= 6.74.
X̄
(c) Since the alternative hypothesis is Ha : µ > 3.0, the p-value is the area to the
right of 6.74 under the standard normal curve. This area is very, very small
(see Figure 9.1).
-6 -4 -2 0 2 4 6
6.74
Figure 9.1: The p-value is the area to the right of 6.74 under the standard normal curve.
Using software: p-value = 7.8 × 10−8 . Using a standard normal table, we could
say only that the p-value is very close to 0 (or we may give a range, such as
p-value < 0.0000003).
(d) The p-value is less than the given signifcance level of α = 0.05, so we can
reject the null hypothesis at α = 0.05.
(e) There is statistically signifcant evidence (at α = 0.05) against the null hy-
pothesis that µ = 3.0 mg/kg dw. So there is very strong evidence that the
population mean µ is greater than 3.0 mg/kg dw. This means that there is
statistically signifcant evidence that the true mean concentration of cadmium
in caps of Boletus edulis mushrooms in this area is greater than the Slovenian
tolerable value of 3.0 mg/kg dw.
13. (a) H0 : µ = 3.35 kg. (The true mean weight of full-term Hispanic newborns in
the U.S. in 2013 is 3.35 kg – the same mean as for full-term newborns of the
general population.)
Ha : µ 6= 3.35 kg. (The true mean weight of full-term Hispanic newborns in
the U.S. in 2013 di˙ers from 3.35 kg. That is, the true mean weight di˙ers
from that of full-term newborns of the general population.)
¯ ¯
(b) Z = X−µσ
0
= X−µ√ 0 = 3.318−3.35
σ/ n
√
0.47/ 100
= −0.681.
X̄
(c) 0.496. The alternative hypothesis is two-sided, so the p-value is double the
area in the tail, beyond the observed value of the test statistic. In this case,
the p-value is double the area to the left of −0.681 under the standard normal
curve (see Figure 9.2).
The area to the left of −0.681 under the standard normal curve is 0.248, and
so the p-value is 0.496.
(d) The p-value is greater than the given signifcance level of α = 0.05, so we
do not reject the null hypothesis at this level of signifcance. (The p-value is
in fact much larger than 0.05, so we would not reject the null hypothesis at
Balka ISE 1.09 9.5. INTERPRETING THE P -VALUE 128
The p-value is double this area
-6 -4 -2 0 2 4 6
-0.681
Figure 9.2: The p-value is double the area to the left of −0.681 under the standard normal
curve.
any reasonable signifcance level, but we used α = 0.05 since it was the given
signifcance level for this problem.)
(e) Based on this sample, there is no evidence that the true mean weight of full-
term Hispanic newborns in 2013 di˙ers from 3.35 kg (the mean weight of all
full-term U.S. newborns that year). We can say that the data is consistent
with full-term Hispanic newborns having the same mean weight as that of
full-term newborns in the general population. (N.B. We cannot say that they
are equal, just that we do not have any evidence of a di˙erence.)
9.5 Interpreting the p-value
9.5.1 The Distribution of the p-value When H0 is True
14. (a) The p-value will have a uniform distribution on the interval (0,1).
(b) If the null hypothesis is true, on average the p-value will equal 0.5.
9.5.2 The Distribution of the p-value When H0 is False
15. (a) A p-value of 2.9 × 10−48 gives extremely strong evidence against the null hy-
pothesis. There is almost no chance of observing what was observed if the null
hypothesis were true.
(b) A p-value of 0.0008 gives very strong evidence against the null hypothesis.
There is less than a one-in-a-thousand chance of observing what was observed
if the null hypothesis were true.
(c) A p-value of 0.02 gives strong, but not overwhelming, evidence against the null
hypothesis.
Balka ISE 1.09 9.6. TYPE I ERRORS, TYPE II ERRORS, AND THE POWER OF A TEST 129
(d) A p-value of 0.07 gives a little evidence against the null hypothesis, but it is
not strong (and not signifcant at the commonly used α level of 0.05).
(e) A p-value of 0.12 gives possibly a hint of evidence against the null hypothesis,
but nothing that is signifcant at any of the usual signifcance levels.
(f) A p-value of 0.45 gives no evidence against the null hypothesis.
(g) A p-value of 0.88 gives no evidence against the null hypothesis.
9.6 Type I Errors, Type II Errors, and the Power of a

Test
16. (a) Here the null hypothesis is false and we will reject it, so we will make neither
error. (Our decision is consistent with the underlying reality.)
(b) Here the null hypothesis is false and we will not reject it, so we will make a
Type II error.
(c) Here the null hypothesis is true and we will reject it, so we will make a Type
I error.
(d) Here the null hypothesis is true and we will not reject it, so we will make
neither error. (Our decision is consistent with the underlying reality.)
17. It all boils down to the balance between Type I and Type II errors. The signifcance
level α is the probability of a Type I error (given the null hypothesis is true). If a
Type I error has major repercussions (thousands of people will die, for example),
but a Type II error is not as costly, then we should choose a very small value for α.
If a Type I error is not very costly relative to a Type II error, then a larger choice
of α is reasonable.
9.6.1 Calculating Power and the Probability of a Type II Error
18. (a) Reject H0 if X¯ ≤ 26.710. With the alternative hypothesis of Ha : µ < 30 and
α = 0.05, we reject the null hypothesis if Z ≤ −1.645. Expressing the rejection
region in terms of X:
¯
Z ≤ −1.645
¯
X − 30
√ ≤ −1.645
20/ 100
X¯ ≤ 30 + √20 (−1.645)
100
¯
X ≤ 26.710
(b) 0.259. The power of the test is the probability of rejecting the null hypothesis,
given µ = 28:
26.710 − 28
Power = P (X
¯ ≤ 26.710|µ = 28) = P (Z ≤ √ )
20/ 100
= P (Z ≤ −0.645)
= 0.259
(c) 0.639. The power of the test is the probability of rejecting the null hypothesis,
given µ = 26:
¯ ≤ 26.710|µ = 26) = P (Z ≤ 26.710 − 26

Power = P (X √ )
20/ 100
= P (Z ≤ 0.355)
= 0.639
19. (a) Reject H0 if X

¯ ≥ 818.61. With the alternative hypothesis of Ha : µ > 800 and
α = 0.01, we reject the null hypothesis if Z ≥ 2.326.. (If you use a table to
fnd the z value, you will only be able to fnd it to 2 decimal places: Z ≥ 2.33.
There will be a little rounding error.) Expressing the rejection region in terms
of X
¯:
Z ≥ 2.326
¯
X − 800
√ ≥ 2.326
160/ 400
¯ ≥ 800 + √160 (2.326)
X
400
¯ ≥ 818.61
X
(b) 0.140. The power of the test is the probability of rejecting the null hypothesis,
given µ = 810:
¯ ≥ 818.61|µ = 810) = P (Z ≥ 818.61√− 810 )

Power = P (X
160/ 400
= P (Z ≥ 1.076)
= 0.140
(c) 0.996. The power of the test is the probability of rejecting the null hypothesis,
given µ = 840:
818.61 − 840
Power = P (X
¯ ≥ 818.61|µ = 840) = P (Z ≥ √ )
160/ 400
= P (Z ≥ −2.674)
= 0.996
20. (a) Reject H0 if X¯ ≤ 11.674 or X¯ ≥ 18.326.

Recall that for a two-sided alternative hypothesis and α = 0.05, we reject H0
if Z ≤ −1.96 or Z ≥ 1.96. We reject H0 if:
Z ≤ −1.96 or Z ≥ 1.96
¯ − µ0
X ¯ − µ0
X
√ ≤ −1.96 √ ≥ 1.96
σ/ n σ/ n
X¯ − 15 X¯ − 15
√ ≤ −1.96 √ ≥ 1.96
12/ 50 12/ 50
¯ ≤ 11.674
X ¯ ≥ 18.326
X
(b) The power of the test if µ = 16 is the total area in these two regions under
the sampling distribution of X.
¯
Power = P (X
¯ ≤ 11.674|µ = 16) + P (X
¯ ≥ 18.326|µ = 16)
11.674 − 16 18.326 − 16
= P (Z ≤ √ ) + P (Z ≥ √ )
12/ 50 12/ 50
= P (Z ≤ −2.549) + P (Z ≥ 1.371)
= 0.0054 + 0.0852
= 0.0906
(c) If µ = 17:
Power = P (X
¯ ≤ 11.674|µ = 17) + P (X
¯ ≥ 18.326|µ = 17)
11.674 − 17 18.326 − 17
= P (Z ≤ √ ) + P (Z ≥ √ )
12/ 50 12/ 50
= P (Z ≤ −3.138) + P (Z ≥ 0.781)
= 0.0008 + 0.2173
= 0.2181
Balka ISE 1.09 9.7. ONE-SIDED TEST OR TWO-SIDED TEST? 132
(d) If µ = 13:
Power = P (X
¯ ≤ 11.674|µ = 13) + P (X
¯ ≥ 18.326|µ = 13)
11.674 − 13 18.326 − 13
= P (Z ≤ √ ) + P (Z ≥ √ )
12/ 50 12/ 50
= P (Z ≤ −0.781) + P (Z ≥ 3.138)
= 0.2173 + 0.0008
= 0.2181
21. (a) P (X
¯ > 10|µ = 0) = P (Z > 10−0
√ )
100/ 50
= P (Z > 0.707) = 0.24.
(b) 0. A Type II error is possible only when the null hypothesis is false.
(c) 0. A Type I error is possible only when the null hypothesis is true.
(d) P (X
¯ > 10|µ = 5) = P (Z > 10−5 √ ) = P (Z > 0.353) = 0.36.
100/ 50
(e) P (X
¯ > 10|µ = 10) = P (Z > 10−10
√ ) = P (Z > 0) = 0.5.
100/ 50
9.6.2 What Factors A˙ect the Power of the Test?
22. (a) An increase in the sample size will increase the power of the test.
(b) A decrease in the signifcance level will decrease the power of the test.
(c) An increase in the distance between the true mean and the hypothesized mean
will increase the power of the test.
(d) A decease in the population variance will increase the power of the test.
9.7 One-Sided Test or Two-Sided Test?
23. If we choose a one-sided alternative hypothesis, then the test will have greater
power than its two-sided counterpart if the true mean lies in the direction specifed
by the alternative hypothesis. The downside of the one-sided alternative is that if
the true mean lies in the opposite direction, then it will be impossible to detect
this di˙erence from the hypothesized value. If we choose a two-sided alternative
hypothesis, then the test will be able to detect a di˙erence from the hypothesized
value in either direction.
24. This is poor statistical practice. We should not let the sample data infuence our
choice of hypothesis. If we are interested in a di˙erence from the null hypothesis
in either direction, then a two-sided alternative should be chosen at the outset.
Balka ISE 1.09 9.8. STATISTICAL SIGNIFICANCE AND PRACTICAL SIGNIFICANCE 133
Choosing a one-sided alternative because it fts the observed data is cheating. If

we used the approach given in this question, we would be reporting a p-value that
is smaller than it should be and our stated conclusions would be misleading.
9.8 Statistical Signifcance and Practical Signifcance
25. (a) If the p-value of a hypothesis test is 0.0000001, then there is very strong
evidence that the results of the test have practical importance. False. A
very small p-value means that there is very strong evidence against the null
hypothesis. Whether or not these results have any practical importance is a
di˙erent question (one that cannot be answered with a hypothesis test).
(b) If the sample size is extremely large, then even tiny, meaningless di˙erences
from the hypothesized value will be found statistically signifcant. True.
(c) If the results of a hypothesis test are reported, then there is no need to also
report the associated confdence interval for the parameter. False. It is good
practice to also report a confdence interval for the parameter of interest, to
give an idea of the size of the e˙ect.
9.9 The Relationship Between Hypothesis Tests and Con-

fdence Intervals
26. (a) No. The p-value is less than 0.05, implying 50 would fall outside of the 95%
confdence interval.
(b) Yes. The p-value is greater than 0.01, implying 50 would be contained in the
99% confdence interval.
27. B (the p-value is less than 0.05). Since the 95% confdence interval for µ does not
contain 0, a two-sided test of H0 : µ = 0 would result in a p-value that is less than
0.05.
9.10 Hypothesis Tests for µ When σ is Unknown
28. (a) Reject H0 if |t| ≥ 1.761.

(b) Reject H0 if t ≤ −1.345.
(c) Reject H0 if t ≥ 1.345.
(d) Reject H0 if |t| ≥ 2.056.
(e) Reject H0 if t ≤ −3.183.
Balka ISE 1.09 9.10. HYPOTHESIS TESTS FOR µ WHEN σ IS UNKNOWN 134
(f) Reject H0 if t ≥ 1.038.
29. (a) Using software: p-value = 0.0508. From the table: 0.05 < p-value < 0.10.
(b) Using software: p-value = 0.9745. From the table: 0.95 < p-value < 0.975.
(c) Using software: p-value = 0.0254. From the table: 0.025 < p-value < 0.05.
(d) Using software p-value = 0.0833. From the table: 0.05 < p-value < 0.10.
(e) Using software: p-value = 0.0416. From the table: 0.025 < p-value < 0.05.
(f) Using software: p-value = 0.9584. From the table: 0.95 < p-value < 0.975.
9.10.1 Examples of Hypothesis Tests Using the t Statistic
30. (a) The hypotheses are:
H0 : µ = 11.0 (This type of burrito has a true mean fat content of 11.0 g.)
Ha : µ 6= 11.0 (This type of burrito has a true mean fat content that di˙ers from 11.0 g.)
(b) SE(X
¯) = √s
n
= 1√
.45
6
= 0.59196.
X¯ − µ0
t= ¯)
SE(X
11.49 − 11.0
=
0.59196
= 0.828
(c) Since the alternative hypothesis is two-sided, the p-value is double the area
in the tail, beyond the observed value of the test statistic. In this case, the
p-value is double the area to the right of 0.828 under a t distribution with
n − 1 = 6 − 1 = 5 degrees of freedom (see Figure 9.3).
-6 -4 -2 0 2 4 6
0.828
Figure 9.3: The p-value is double the area to the left of 0.828 under a t distribution with
5 degrees of freedom.
Balka ISE 1.09 9.11. MORE ON ASSUMPTIONS 135
Using software, p-value = 0.45. Using a t table, we could fnd only a range:
p-value > 0.40.
(d) Since the p-value is large (0.45), there is no evidence against the null hypoth-
esis. There is no evidence that the true mean fat content in this type of bean
burrito di˙ers from the stated value of 11 grams. (The observed data is con-
sistent with the true mean equalling 11 grams.) N.B. We cannot say that the
true mean does equal 11.0 g, we can only that we have no evidence that it
di˙ers from 11 g.
(e) In order for the t procedure to be valid, we need the sample to be a simple
random sample from the population of interest, and we need the population
to be normally distributed. The sampling design was not clearly stated (we do
not know how the sample was obtained), so we cannot make a truly informed
decision about any possible biases that may have been introduced. While fast-
food restaurants strive for consistency, it is likely that the amount of fat in a
burrito tends to change a little from preparer to preparer, and possibly from
restaurant to restaurant. But given the nature of fast food it is probably not
too far o˙ the mark to think of the sample as a random sample from all bean
burritos from this fast-food chain.
The sample size is small, so the normality assumption is important. We were
not given the data, but we were told that there were no outliers, which is
somewhat comforting. Using the t procedure here is a little questionable, but
still reasonable.
9.11 More on Assumptions
31. No, it would not be reasonable to use the t procedures on this raw data. This plot
shows some right skewness and a large outlier. (Outliers and skewness negatively
a˙ect the t procedures.) The negative impact of skewness and/or outliers can be
bad, especially when the sample size is not large. (The sample size here is 20, which
is not very large.) There are other analysis options, such as using the t procedure
on a transformation of the data (such as the log of the data values), or using a
distribution-free procedure.
32. (a) The t test performs poorly when we are sampling from a strongly skewed
distribution, especially if the sample size is small.
(b) If we use the t test when sampling from skewed distributions, then the true
signifcance level may be quite di˙erent from the reported signifcance level,
and the true p-value may be quite di˙erent from the stated p-value. In short,
the reported results may be very misleading. This is especially true for small
sample sizes.
Balka ISE 1.09 9.12. CHAPTER SUMMARY 136
9.12 Chapter Summary
X̄−µ0
33. (a) t = SE(X̄) = 15.5−20
√
2/ 10
= −7.12. If the null hypothesis is true, the t test statistic
will have a t distribution with n − 1 = 10 − 1 = 9 degrees of freedom.
X̄−µ0
(b) t = SE(X̄) = 15.5−20
√
20/ 10
= −0.712. If the null hypothesis is true, the t test statistic
(c) Z = X̄−µσ
0
= 35.5−30
√
2/ 40
= 17.39. If the null hypothesis is true, the Z test
X̄
statistic will have the standard normal distribution. (Since we are sampling
from a normally distributed population and σ is known.)
X̄−µ0
(d) t = SE(X̄) = 45.5−40
√
20/ 50
= 1.94. If the null hypothesis is true, the t test statistic
will have a t distribution with n − 1 = 50 − 1 = 49 degrees of freedom. (With
these degrees of freedom, the t distribution will be similar to the standard
normal distribution.)
X̄−µ0
34. (a) t = SE(X̄) = 35.5−30
√
2/ 20
(b) Z = X̄−µσ
0
= 35.5−40
√
20/ 25
= −1.125. If the null hypothesis is true, the Z test
X̄
statistic will have the standard normal distribution. (Since we are sampling
from a normally distributed population and σ is known.)
X̄−µ0
(c) t = SE(X̄) = 65.5−30
√
2/ 200
(With these degrees of freedom, the t distribution is very close to the standard
normal distribution.)
X̄−µ0
(d) t = SE(X̄) = 10/32.1−32
√
60000
(With these degrees of freedom, the t distribution is very, very close to the
standard normal distribution.)
35. (a) The null hypothesis is X¯ = 50. False. We make hypotheses about parameters,
not statistics.
(b) The sample size is 24. False. n = 25.
(c) The area to the left of −4.1027 under a t distribution with 24 degrees of
freedom is 2 × 0.0004 = 0.0008. False. Twice the area to the left of −4.1027 is
given in the output as the p-value (0.0004), so the area to the left of −4.1027
is 0.0002.
(d) If we used this data to test the null hypothesis that the population mean is
40.0, against a two-sided alternative hypothesis, the p-value would be less than
0.05. True, since the 95% confdence interval does not contain 40.
(e) The 95% margin of error is greater than 20.0. False. The margin of error is
the half-width of the interval: 48.94322−46.80365
2 = 1.069785.
9.12.2 Concepts
36. For a one-sample Z test about a population mean µ to be valid, the sample must
be a simple random sample from a normally distributed population. (We also need
to know the true value of σ.) The assumption of a normally distributed population
is not very important if the sample size is large.
37. For a one-sample t test about a population mean µ to be valid, the sample must be
a simple random sample from a normally distributed population. The assumption
of a normally distributed population is not very important if the sample size is
large.
38. If the assumptions of a test are violated, then the reported signifcance level and
p-value will be di˙erent from their “true” values, and the reported results may be
misleading.
39. Both tests require the assumption of a normally distributed population, and for
both tests this assumption becomes less important as the sample size increases. We
use a Z test when the population standard deviation σ is known. We use a t test
when the population standard deviation σ is unknown, and must be estimated by
the sample standard deviation s.
40. The signifcance level, the hypothesized mean, and the choice of alternative hy-
pothesis should not be based on the current sample’s data. (One should be able to
arrive at the appropriate choices before looking at the sample data.) The sample
mean, the value of the test statistic, and the p-value do involve the current sample
data (the test statistic and p-value also involve the appropriate hypotheses of the
test).
41. (a) Here the normality assumption of the t test is true. If the null hypothesis is
true, the t statistic will have a t distribution, and will therefore have a mean
of 0.
(b) Here the normality assumption of the t test is true. If the null hypothesis is
true, the p-value will have a continuous uniform distribution on the interval
(0,1). As such, the p-value will have a mean of 0.5.
42. (a) We are carrying out 25 independent tests, and the probability that we make a
Type I error in any individual test is 0.05, so the number of Type I errors we
make will have a binomial distribution with n = 25 and p = 0.05.
(b) Since the mean of a binomial random variable is np, on average we will make
np = 25 · 0.05 = 1.25 Type I errors.
(c) P (At least one Type I error) = 1−P (No Type I errors) = 1−0.9525 = 0.7226.
43. B (0.0358). Since the hypothesized value of 29 is not contained in the 95% conf-
dence interval, the p-value of the two-sided test would be less than 0.05. Since the
hypothesized value of 29 is contained in the 99% confdence interval, the p-value
of the two-sided test would be greater than 0.01. So, 0.01 < p-value < 0.05. Of
the 5 numbers listed, the only one that lies between 0.01 and 0.05 is 0.0358. (It is
possible to verify that this is the correct p-value by carrying out the appropriate
calculations.)
44. (a) X¯ = −0.0377 (the midpoint of the confdence interval).

(b) The margin of error is the half-width of the interval: 0.2656748−(−0.3412277)
2 =
0.30345.
(c) The p-value given in the output (0.8025) is double the area to the left of
−0.2518 under a t distribution with 39 degrees of freedom. For an alternative
of µ < 0, the p-value is the area to the left of the test statistic, which is
0.8025
2 = 0.40125.
(d) Here we require the area to the right of the observed value of the test statistic,
which is 1 − 0.8025
2 = 1 − 0.40125 = 0.59875.
(e) A p-value of 0.008025 gives strong evidence against the null hypothesis. Here,
there is strong evidence that the true population mean µ di˙ers from 0.
45. All 3 values are possible values of µ. Although there is strong evidence against
the null hypothesis, indicating that the population mean is likely less than 32, any
value of µ is still possible.
46. Whether or not it is reasonable to use the t procedures is debatable in both cases.
The plot on the left shows strong right skewness (a clear violation of the normality
assumption). Skewness negatively impacts the performance of the t procedures, but
in this situation the large sample size of n = 100 would lessen the impact. In this
situation, we would likely be better o˙ doing an analysis involving a transformation
of the data.
The plot on the right shows no clear violation of the normality assumption, but
the sample size is very small. In order to use the t procedures when the sample
size is this small, we need to be very confdent that the normality assumption is
reasonable. (Use of the t procedures is dubious when the small sample size is so
small.)
47. (a) SE(X

¯) = √s
40
= 0.591.
(b) There is very strong evidence (two-sided p-value = 0.00017) against the null
hypothesis. There is very strong evidence that the true population mean di˙ers
from 50. The point estimate of µ is 52.46, with an associated 95% confdence
interval of (51.26, 53.66), giving strong evidence that the population mean is
in fact greater than 50.
(c) 0.05. The signifcance level of a test is the probability of a Type I error (the
probability of rejecting the null hypothesis, given the null hypothesis is true).
(d) The p-value is less than 0.05, so we would reject the null hypothesis at α = 0.05.
But if µ = 50, the null hypothesis is true. We would have rejected a true null
hypothesis, and thus we would have committed a Type I error.
(e) No. The plots give no indication of a violation of the normality assumption,
so it is reasonable to carry out a t test.
48. (a) Sample B would have the largest p-value of this test (the least evidence that
its population mean di˙ers from 18).
(b) Sample C would have the smallest p-value of this test (the greatest evidence
that its population mean di˙ers from 18).
49. (a) No, the null hypothesis is not rejected (since the p-value is greater than the
signifcance level).
(b) Yes, the null hypothesis is rejected (since the p-value is less than the signif-
cance level).
(c) C (but opinions may di˙er here). Since the p-values are nearly identical in
the two tests, the amount of evidence against the null hypothesis is nearly
identical.
50. A. If a Type I error is much, much worse than a Type II error, we should choose a
very small value for α.
51. (a) A Type I error is rejecting the null hypothesis when it is true. If the null
hypothesis is true, we will get tails (and reject the null hypothesis) with prob-
ability 0.5. The probability of a Type I error is thus 0.5.
(b) For it to be impossible to make a Type II error, it must be impossible to
not reject the null hypothesis when it is false. This is true here only if the
probability of tails is 1.
52. The null hypothesis H0 : X¯ = 10 is inappropriate, since we make hypotheses about

parameters and not about statistics. The given p-value must be incorrect, as p-
values are probabilities and cannot be greater than 1.
¯ = µ0 .
53. Since the one-sided p-value is 0.5, this implies Z = 0, which in turn implies X
In this case, X
¯ = 50.
54.
P (Type II error) = P (Do not reject H0 when it is false)
¯ ≤ 20.1|µ = 20.3)
= P (X
20.1 − 20.3
= P (Z ≤ √ )
0.5/ 100
= P (Z ≤ −4.0)
= 0.00003167
55. (a) P (Z > 3) + P (Z < −3) = 0.0013 + 0.0013 = 0.0026.

(b) The probability of a Type I error will stay constant. (Due to the nature of
her rejection rule, the probability of a Type I error for any sample size is what
was calculated in 55a.)
(c) 0. Since a Type I error is rejecting the null hypothesis when it is true, if the
null hypothesis is false we cannot make a Type I error.
(d) The power of her test will increase as the sample size increases.
56. B (the reported p-value will be less than it should be). The standard normal
distribution has less area in the tails than the t distribution, and the two-sided p-
value is double the area in the tail, beyond the observed value of the test statistic.
57. (a) All else being equal, the power of a test will increase as the population variance
decreases. True. With lower variability there is less uncertainty and a greater
ability to detect deviations from the null hypothesis.
(b) If the signifcance level of a test is decreased (from 0.10 to 0.05, say), then the
probability of a Type II error will increase. True.
(c) The probability of a Type II error decreases the farther the true population
mean is from the hypothesized mean. True (for two-sided alternatives), and
true in spirit for one-sided alternatives.
(d) If the null hypothesis is true, we cannot make a Type II error. True.
58. (a) Power = 1 − P (Type I error). False.

(b) Power = 1 − P (Type II error). True.
(c) P (Type I error) = 1 − P (Type II error). False.
(d) All else being equal, the power of a test will decrease the closer the true value
of the population mean is to the hypothesized value of the population mean.
True.
(e) All else being equal, the power of a test will increase as the sample size in-
creases. True.
59. (a) If a null hypothesis is rejected at the 5% signifcance level, then it would also
be rejected at the 10% signifcance level. True. If we reject H0 at α = 0.05,
this implies p-value ≤ 0.05, and we would also reject at α = 0.10.
(b) If a null hypothesis is rejected at α = 0.05, then it would also be rejected at

α = 0.01. False. If we reject H0 at α = 0.05, then p-value ≤ 0.05. We cannot
tell if the p-value is less than 0.01.
(c) If a null hypothesis is not rejected at the 5% signifcance level, then it will
defnitely not be rejected at the 10% signifcance level. False. If we do not
reject at α = 0.05, we know that p-value > 0.05, and it may or may not be
greater than 0.10.
(d) The signifcance level of a test is 1 − p-value. False.
(e) The signifcance level of a test is the probability the null hypothesis is true.
False.
60. (a) If the p-value of a hypothesis test with a two-sided alternative is equal to
exactly 0, we can be certain the null hypothesis is false. (This is a conceptual
question—for all the tests we have done, it is not possible to get a p-value of
exactly 0.) True, since this would mean it was impossible to observe what was
observed if the null hypothesis were true.
(b) The p-value of a test is the probability of getting an insignifcant result, given
the null hypothesis is true. False.
(c) The p-value of a test is less than the signifcance level whenever the null hy-
pothesis is false. False.
(d) The p-value of a test is the probability of rejecting the null hypothesis when
it is false. False.
(e) If the p-value of a test is large (0.99, say), then we can be almost positive that
the null hypothesis is true. False.
61. (a) For a two-sided alternative hypothesis, the p-value of the test will equal 1 if
the sample mean equals the hypothesized mean. True.
(b) If we are sampling from a normally distributed population and conducting
an ordinary t test, then if the null hypothesis is true the p-value will have a
uniform distribution between 0 and 1. True.
(c) If the null hypothesis is true, then on average the p-value will equal 0.5. True.
(d) If the null hypothesis is false, then on average the p-value will equal 0.5. False.
9.12.3 Applications
62. (a) The boxplot and the normal QQ plot both show indications of right-skewness
(which is typical of BMI data). This skewness is not a serious problem in this
scenario, as the skewness is not very strong, and the sample size is fairly large
(94). We can use the t procedures.
(b) The hypotheses are:
H0 : µ = 22.4 (The true mean BMI equals 22.4.)

Ha : µ 6= 22.4 (The true mean BMI di˙ers from 22.4.)
(c) SE(X
¯) = √s
n
= 5.4
√
94
= 0.5570.
¯ − µ0
X
t= ¯)
SE(X
23.3 − 22.4
=
0.5570
= 1.616
(d) Since the alternative hypothesis is two-sided, the p-value is double the area to
the right of 1.616 under a t distribution with n − 1 = 94 − 1 = 93 degrees of
freedom (see Figure 9.4). Using software, we can fnd that the p-value is 0.1095.
0 1.616
Figure 9.4: The p-value is double the area to the right of 1.616 under a t distribution
with 93 degrees of freedom.
Using a table, we could fnd only a range of values: 0.10 < p-value < 0.20.
(e) Because the p-value is not small (p-value = 0.1095), there is not strong ev-
idence against the null hypothesis (The evidence against the null hypothesis
would not be signifcant at α = 0.05 or even α = 0.10.) There is not strong
evidence that the population mean BMI for 14 year-old boys in rural Texas
di˙ers from that of 14 year-old boys in the United States as a whole. (It is
plausible that they are the same.)
(f) There is little or no evidence (p-value = 0.1095) against the null hypothesis
that the true mean BMI for 14 year-old boys in rural Texas di˙ers from that
of 14 year-old boys in the United States as a whole. We can be 95% confdent
that the population mean BMI for 14 year-old boys in rural Texas lies between
22.2 and 24.4 kg/m2 . (Note that the 95% confdence interval captures 22.4,
the hypothesized value of the population mean.)
The sample is drawn from a school district in North Central Texas, so very
loosely speaking the population is all 14 year-old boys in north central Texas.
However, there are some sampling biases that make this assessment not quite
true:
• The sample was drawn from only 6 schools in the area. Students in these
schools may di˙er in a meaningful way from those at other schools in the
area (or from children that don’t go to school).
• For students to participate, parents needed to provide informed consent.
Students whose parents provide informed consent may di˙er in a mean-
ingful way from those whose parents do not.
• Students needed to provide self-assent (the student had to agree to be
part of the study). Students who agree to be part of a study like this may
di˙er in a meaningful way from those who do not.
• Students who participated were given a T-shirt. This could introduce
bias, as it is conceivable that there is a relationship between desire for the
t-shirt and BMI.
63. (a) The plots look good, and show no indication of a violation of the normality
assumption. There are no outliers, obvious skewness, or other problems. While
we would prefer a larger sample size (n = 11 here), it is reasonable to use the
t procedures.
(b) H0 : µ = 0. (The true mean di˙erence is 0. On average, the BR estimation
method is equal to the true age at death.)
Ha : µ 6= 0. (The true mean di˙erence is not 0. On average, the BR estimation
method does not equal the true age at death.)
(c) SE(X ¯ ) = √s = 7√.784 = 2.3470.
n 11
X¯ − µ0
t= ¯)
SE(X
11.073 − 0
=
2.3470
= 4.718
freedom. (see Figure 9.6).
Using software, we can fnd that the p-value is 0.0008. Using a table, we could
fnd only a range: p-value < 0.001.
(e) There is very strong evidence against the null hypothesis (p-value = 0.0008).
There is very strong evidence that the true mean di˙erence between the BR
estimate and the true age di˙ers from 0.
(f) There is very strong evidence against the null hypothesis (p-value = 0.0008).
There is very strong evidence that the true mean di˙erence between the BR
-6 -4 -2 0 2 4 6
4.718
estimate and the true age di˙ers from 0. We can be 95% confdent that the true
mean di˙erence between the BR estimate and the true age lies between 5.84
years and 16.302 years. (Note that the hypothesized value of 0 falls outiside
of the 95% interval.) It appears as though the true mean di˙erence is in fact
greater than the hypothesized value of 0, and therefore the BR estimate tends
to be too large for individuals of this age.
The conclusions apply only to individuals that die at the age of 22 (but they
would likely also hold for people of similar age). The sample is not truly a
simple random sample from any population, and there may be strong sampling
biases present, depending on the specifc nature of the sample. While we are
attempting to use these results to draw inferences about all people that die
at 22, the BR method may have di˙erent performance for di˙erent subgroups.
For example, it is possible that the BR aging method performs di˙erently for
di˙erent ethic groups. It is also conceivable that people who donate their
bodies for educational or medical research di˙er in a meaningful way from
those that do not. Experts in the feld of dental aging would likely be able to
provide guidance about the possible impacts of this type of bias.
64. (a) H0 : µ = 1. (The true mean 2D:4D ratio equals 1.)

Ha : µ =
6 1. (The true mean 2D:4D ratio di˙ers from 1.)
(b) SE(X ) = √sn = √
¯ 0.035
135
= 0.00301.
¯ − µ0
X
t= ¯)
SE(X
0.994 − 1
=
0.00301
= −1.992
(c) Since the alternative hypothesis is two-sided, the p-value is double the area to
the left of -1.992 under a t distribution with n − 1 = 135 − 1 = 134 degrees of
freedom (see Figure 9.6).
-6 -4 -2 0 2 4 6
-1.992
Figure 9.6: The p-value is double the area to the left of −1.992 under a t distribution
Using software, we can fnd that the p-value is 0.048. Using a table, we could
fnd only a range: 0.02 < p-value < 0.05.
(d) The p-value is less than the given signifcance level (0.048 < 0.05), so there is
statistically signifcant evidence against the null hypothesis. There is strong
evidence that the true mean 2D:4D ratio (on the left hand) for women of
European descent at Western Washington University di˙ers from 1.
(e) The p-value is less than the given signifcance level (0.048 < 0.05), so there is
statistically signifcant evidence against the null hypothesis. There is strong
evidence that the true mean 2D:4D ratio (on the left hand) for women of
European descent at Western Washington University di˙ers from 1. We can
be 95% confdent that the true mean 2D:4D ratio lies between 0.98804 and
0.99996. (Note that the hypothesized value of 1 falls outiside of the 95%
interval.) It appears as though the true mean 2D:4D ratio is in fact less than
the hypothesized value of 1. (Female students of European descent tended to
have an index fnger that is shorter than the ring fnger.)
The sample was of female students of European descent in an introductory
biology course at Western Washington University. So our conclusions apply
to women of European descent enrolled in biology courses at this university,
and to a lesser extent, women of European descent at the university. While we
may look to studies like this to provide information about women of European
descent in general, there are many possible biases to keep in mind, including:
• Students enrolled in an introductory biology course may tend to have
di˙erent physical characteristics than other students.
• Students had to consent to being part of this study, and students who
consent may possibly di˙er in a meaningful way from those that do not.
• Most importantly, women of European descent at this university may
di˙er in a meaningful way from women of European descent in general.
While it is seemingly unlikely that these factors would have a major impact on
the 2D:4D ratio, it is very possible that they would have a meaningful e˙ect.
University students are not a random sample from the general population.
65. (a) The hypotheses are:
H0 : µ = 30.0 (This type of cheeseburger has a true mean protein content of 30 g.)
Ha : µ 6= 30.0 (This type of cheeseburger has a true mean protein content that di˙ers from 30 g.)
(b) SE(X
¯) = √s
n
= 2√
.72
6
= 1.1104.
X¯ − µ0
t= ¯)
SE(X
33.81 − 30.0
=
1.1104
= 3.431
(c) Since the alternative hypothesis is two-sided, the p-value is double the area
in the tail, beyond the observed value of the test statistic. In this case, the
p-value is double the area to the right of 3.431 under a t distribution with
n − 1 = 6 − 1 = 5 degrees of freedom (see Figure 9.7).
-6 -4 -2 0 2 4 6
3.413
Figure 9.7: The p-value is double the area to the left of 3.431 under a t distribution with
5 degrees of freedom.
Using software, p-value = 0.019. Using a t table, we could fnd only a range:
0.01 < p-value < 0.02.
(d) The p-value is less than the given signifcance level (0.019 < 0.05), so there is
statistically signifcant evidence against the null hypothesis. There is statis-
tically signifcant evidence that the true mean protein content in this type of
cheeseburger di˙ers from the stated value of 30 grams. Since the sample mean
is greater than 30 (33.81), this implies there is strong evidence that the true
mean protein content in this type of cheeseburger is in fact greater than 30
grams.
(e) In order for the t procedure to be valid, we need the sample to be a simple
random sample from the population of interest, and we need the population
to be normally distributed. The sampling design was not clearly stated (we do
not know how the sample was obtained), so we cannot make a truly informed
decision about any possible biases that may have been introduced. While
fast-food restaurants strive for consistency, it is possible that the amount of
protein in a cheeseburger tends to change a little from preparer to preparer,
and from restaurant to restaurant. But given the nature of fast food it is
probably not too far o˙ the mark to think of the sample as a random sample
from all cheeseburgers of this type from this fast-food chain.
The sample size is small, so the normality assumption is important. We were
not given the data, but we were told that there were no outliers, which is
somewhat comforting. Using the t procedure here is a little questionable, but
still reasonable.
66. (a) H0 : µ = 5.0 mg/kg dw. (The true mean lead concentration in caps of Laccaria
amethystina in this area is 5.0 mg/kg of dry weight. (The true mean lead
concentration is at the tolerable value.).)
Ha : µ > 5.0 mg/kg dw. (The true mean lead concentration in caps of Laccaria
amethystina in this area is greater than 5.0 mg/kg of dry weight. (The true
mean lead concentration exceeds the tolerable value.).)
Note that the choice of alternative hypothesis is based on the problem at hand,
and does not depend on the value of the sample mean.
(b) SE(X ¯ ) = √s = √
n
0.25
21
= 0.05455.
X¯ − µ0
t= ¯)
SE(X
0.94 − 5.0
=
0.05455
= −74.421
(c) Since the alternative hypothesis is µ > 5.0, the p-value is the area to the right
of the observed value of the test statistic. In this case, the p-value is the area
to the right of −74.421 under a t distribution with n − 1 = 21 − 1 = 20 degrees
of freedom (see Figure 9.8).
For all intents and purposes, this area is 1. Using software: p-value = 1 −
3.2 × 10−26 . Using a t table, we could fnd only a range: p-value > 0.40. (But
we should have a pretty good idea that the p-value is near 1 just by looking
at the value of the test statistic.)
(d) A p-value this large gives absolutely no evidence against the null hypothesis.
There is no evidence whatsoever that the true mean lead concentration in this
type of mushroom in this area is greater than the tolerable value of 5.0 mg/kg
dw.
The p-value is the area to the right of -74.42
-80 -60 -40 -20 0

-74.42
Figure 9.8: The p-value is the area to the right of −74.42 under a t distribution with 20
degrees of freedom. This area is very, very close to 1.
67. (a) No, there is no indication of a violation of the normality assumption. The
boxplot looks roughly symmetric, and there is no systematic curvature or
outliers evident in the normal QQ plot. A minor violation of the normality
assumption would not be a big deal in this scenario, as the sample size is not
small (n = 51). We can use the t procedures.
H0 : µ = 1.0 (The true mean speed of tandem running is 1.0 mm/s.)

Ha : µ 6= 1.0 (The true mean speed of tandem running di˙ers from 1.0 mm/s.)
(c) SE(X
¯) = √s
n
= 0.60
√
51
= 0.0840.
X¯ − µ0
t= ¯)
SE(X
1.80 − 1.0
=
0.0840
= 9.522
freedom (see Figure 9.9).
(e) There is very, very strong evidence against the null hypothesis (p-value =
8.1 × 10−13 ). There is very strong evidence that the true mean speed of
tandem running (under the conditions of the study) di˙ers from 1.0 mm/s.
(f) There is very, very strong evidence against the null hypothesis (p-value =
8.1 × 10−13 ). There is very strong evidence that the true mean speed of
tandem running (under the conditions of the study) di˙ers from 1.0 mm/s.
We can be 95% confdent that the true mean speed of tandem running (under
the conditions of the study) lies between 1.63 mm/s and 1.97 mm/s. (Note
0 9.522
that the hypothesized value of 1.0 mm/s falls outiside of the 95% interval.) It
appears as though the true mean speed of tandem running is in fact greater
than the hypothesized value of 1.0 mm/s.
Although it wasn’t perfectly clear from the article how the tandem running
pairs were chosen, it is reasonable to think that these results apply to Tem-
nothorax albipennis ants (with a young experienced ant leading a young in-
experienced ant) under the conditions of the study. The true mean may be
di˙erent if the conditions change, or if a di˙erent type of ant is studied.
68. (a) The normality assumption looks reasonable, and we can go ahead with the t
procedures on this data.
(b) The 90% interval is given by 1082 ± 1.771 × √6014 , which works out to 1082 ±
28.399, or (1053.6, 1110.4).
(c) H0 : µ = 920, Ha : µ > 920.
(d) The t procedure is appropriate. Both the t and Z procedures assume a nor-
mally distributed population, which is a reasonable assumption here. Since the
standard deviation is based on sample data (the population standard deviation
is unknown), we would use the t procedure.
(e) t = 1082−920
√
60/ 14
= 10.10.
(f) The p-value of the test is the area to the right of 10.10 under a t distribution
with n − 1 = 14 − 1 = 13 degrees of freedom. Using software, the p-value can
be found to be 8.0 × 10−8 . Using the t table, we could say only that the p-value
is very close to 0 (depending on the specifc table, we might say something like:
p-value < 0.0005).
(g) There is very strong evidence (one-sided p-value = 8.0 × 10−8 ) that the popu-
lation mean calorie content is greater than what the company claims. A 90%
confdence interval for the population mean is found to be (1053.6, 1110.4),

giving strong evidence that the true mean calorie content is more than 100
calories greater than what the company claims.
N.B. These conclusions rely heavily on the assumption of a simple random
sample from the population of interest, and that assumption may not be rea-
sonable. (For example, if all 14 bags were obtained from a new employee who
was putting too much oil in the popcorn, that might seriously bias the results.)
69. (a) (ii) is correct (H0 : µ = 12.0, Ha : µ > 12.0). (Note once again that we make
hypotheses about parameters, and never about statistics.)
(b) t = X̄−µ√ 0 = 16.7−12.0
s/ n
√
4.8/ 9
= 2.9375. The p-value is the area to the right of 2.9375
under a t distribution with n − 1 = 9 − 1 = 8 degrees of freedom. Using
software: p-value = 0.0188. Using a t table, we could only give a range:
0.01 < p-value < 0.02.
(c) Statement 2 (ii) is correct (and the rest of the statements are incorrect). Since
the p-value is less than the given signifcance level of 0.05, the evidence against
the null hypothesis is statistically signifcant at α = 0.05. There is very strong
evidence that the population mean arsenic level is greater than 12.0 ppm (the
Environment Canada guideline).
(d) This might not be the most appropriate test for a number of reasons. An
important reason is that the average arsenic concentration may not be the
best summary of the data, since a high arsenic level in one school is very bad,
regardless of whether it is o˙set by low levels in other schools. Also, we may
wish to look at the hypotheses from the other perspective, and test to see if
there is strong evidence that the population mean arsenic level is less than the
Environment Canada guideline.
(e) We rejected a true null hypothesis, and therefore we made a Type I error.
70. (a) There is extremely strong evidence (p-value = 7.7 × 10−11 ) that the mean
response time for this type of rat under these conditions di˙ers from 1.25
seconds. The point estimate of the mean response time is 1.476, with an
associated confdence interval of (1.42, 1.53). This gives strong evidence that
the true mean response time is greater than 1.25 seconds.
(b) Although it is necessary from time to time, it is unwise to compare sample data
to a “historical control”, as it opens up a host of problems. Perhaps the rats
in this experiment have slower response times due to characteristics of the lab
in which the experiment took place, and not because of an e˙ect of the drug.
(The lab might have a poor measuring device, or low oxygen levels, or some
other type of improper setup.) It is much better to conduct a case-control
experiment, comparing two groups of rats under similar conditions.
71. (a) n = 45.

(b) 0.1622. Since the alternative hypothesis in the output is two-sided, the p-
value given in the output is double the area to the right of 0.9966 under a t
distribution with 44 degrees of freedom. For the alternative Ha : µ > 250, the
p-value is the area to the right of 0.9966, so for this alternative: p-value =
0.3244
2 = 0.1622.
(c) 0.8378. Since the alternative hypothesis in the output is two-sided, the p-
value given in the output is double the area to the right of 0.9966 under a
t distribution with 44 degrees of freedom. For the alternative Ha : µ < 250,
the p-value is the area to the left of 0.9966, so for this alternative: p-value =
1 − 0.3244
2 = 0.8378.
(d) There is no evidence (p-value = 0.32) that the true mean weight of nuts in
packages of this type di˙ers from 250 grams.
(e) The third statement (iii) is a reasonable interpretation of the confdence inter-
val, and the rest of the statements are incorrect.
72. (a) The appropriate hypotheses are given in the frst option: H0 : µ = 0.50, Ha : µ >
0.50.
(b) t = 0.538−0.50
√
0.14/ 150
= 3.32.
(c) The p-value is the area to the right 3.32 under a t distribution with n − 1 =
150 − 1 = 149 degrees of freedom. Using software: p-value = 0.00056. Using
the t table: 0.0005 < p-value < 0.001.
(d) Since the p-value is less than the given signifcance level (0.00056 < 0.05) there
is statistically signifcant evidence against the null hypothesis. There is very
strong evidence that the population mean trans-fat content exceeds 0.50 grams
per serving.
(e) Yes, this hypothesis test would still be informative. The procedure assumes a
normally distributed population, but this assumption becomes less important
as the sample size increases. For a sample size of n = 150, the central limit
theorem tells us that these methods should still work reasonably well. (But
there may be better methods of analysis. For example, we might wish to use
the t procedures on a transformation of the data.)
73. (a) A t statistic is appropriate, since the observations were roughly normally dis-
tributed and the standard deviation is based on sample data.
(b) t = 84.14−82.9
√
0.12/ 100
= 103.33.
(c) The p-value is very close to 0. The p-value is twice the area to right of 103.33
under a t distribution with 100 − 1 = 99 degrees of freedom. This area is very,
very near zero. (While conceptually there is some non-zero area this far out
in the tail, for all intents and purposes this area is 0.)
(d) Since the p-value is less than the given signifcance level of 0.01, we can say
there is statistically signifcant evidence against the null hypothesis. There is
extremely strong evidence that the population mean weight of bolts of this
type exceeds 82.9 grams.
(e) We would have rejected a true null hypothesis, and therefore we would have
made a Type I error.
74. (a) H0 : µ = 800 (the true mean head injury rating for drivers is 800).
Ha : µ < 800 (the true mean head injury rating for drivers is less than 800).
X̄−µ
(b) t = s/ √ = 750−800
n
√
50/ 24
= −4.90.
(c) The p-value is the area to the left of −4.90 under a t distribution with n − 1 =
24 − 1 = 23 degrees of freedom. Using software: p-value = 3.0 × 10−5 . Using
the t table: p-value < 0.0005.
(d) There is very strong evidence (p-value = 3.0 × 10−5 , signifcant at α = 0.05)
that the population mean driver head injury rating for this model of car under
these conditions is less than 800.
75. (a) H0 : µ = 7, Ha : µ 6= 7. t = 0.32/ √ = −1.768. Using software: p-value =

6.8−7
8
0.1204. Using the t table: 0.10 < p-value < 0.20. There is perhaps a tiny hint
of evidence (p-value = 0.12) that the true mean fuel consumption under the
driving conditions of the test di˙ers from 7.0 l/100km, but it is not statistically
signifcant at any of the usual levels.
(b) We need the sample to be a simple random sample from a normally distributed
population. (Can we view the observed values as a random sample from a
population of fuel consumption values for this model of car under these driving
conditions?) The assumption of a random sample from the population could
be violated here (for example, the manufacturer might have hand-picked the
most recently tuned-up cars, or drivers might alter their driving habits in an
e˙ort to improve fuel eÿciency). The normality assumption could be violated
(the distribution of fuel consumption might be skewed, or have other types
of non-normality.) If the assumptions are violated then the stated p-value
will di˙er from the true p-value, and thus the stated conclusions might be
misleading.
Chapter 10
Inference for Two Means
J.B.’s strongly suggested exercises: 4, 6, 7, 9, 10, 11, 17, 23, 26, 27, 30, 31, 33
10.1 Introduction
10.2 The Sampling Distribution of the Di˙erence in Sam-

ple Means
1. (a) µX̄1 −X̄2 = µ1 − µ2 = 15 − 8 = 7.

q 2
σ σ2
q
2 2 √
(b) σX¯ 1 −X̄2 = n11 + n22 = 24 + 64 = 10.
(c) It is impossible to know the shape of the distribution without further infor-
mation. We do not know the shape of the distributions from which we are
sampling, and the sample sizes are not large enough for us to count on the
central limit theorem to provide approximate normality.
(d) µX̄1 −X¯ 2 = µ1 − µ2 = 15 − 8 = 7.
q 2
σ σ2
q
22 62
√
(e) σX¯ 1 −X̄2 = n11 + n22 = 400 + 400 = 0.1.
(f) Although we cannot know the exact shape of the distribution without further
information, the sample sizes are large enough for the distribution of X
¯1 − X
¯2
to be approximately normal (due to the central limit theorem).
2. (a) Let T be a random variable representing Tom’s time on the frst run, and P
be a random variable representing Pete’s time on the frst run. We are given
that T ∼ N (4.612, 0.048) and P ∼ N (4.528, 0.044).
153
10.3. HYPOTHESIS TESTS AND CONFIDENCE INTERVALS FOR TWO INDEPENDENT SAMPLES (WHEN σ1 AND σ2 ARE KNOWN)
Balka ISE 1.09 154
T − P has aqmean of µT − µP = 4.612 − 4.528 = 0.084 and a standard

√
deviation of σT2 + σP2 = 0.0482 + 0.0442 = 0.0651. Since both T and P
are approximately normal, T − P is approximately normal. P (T > P ) =
0 − 0.084
P (T − P > 0) = P (Z > ) = P (Z > −1.290) = 0.901.
0.0651
(b) Let X¯ T represent the mean of Tom’s 3 times and X ¯ P represent the mean of
Pete’s 3 times. X¯T − X¯ P has a mean of µT − µP = 4.612 − 4.528 = 0.084 and
q q
2
σT σ2 0.0482 2
a standard deviation of 3 + P =
3 + 0.044 = 0.03759. P (X
3 3
¯T >
X ¯T − X
¯ P ) = P (X ¯ P > 0) = P (Z > 0−0.084
= P (Z > −2.234) = 0.987.
0.03759 )
10.3 Hypothesis Tests and Confdence Intervals for Two

Independent Samples (When σ1 and σ2 are known)
10.4 Hypothesis Tests and Confdence Intervals for µ1 −

µ2 (When σ1 and σ2 are unknown)
10.4.1 Pooled-Variance t Tests and Confdence Intervals
3. (a) The point estimate of the di˙erence in the population means (µ1 − µ2 ) is the
value of the di˙erence in the sample means: X¯1 − X
¯ 2 = 8.8 − 17.2 = −8.4.
2 2
(b) s2p = (10−1)1.42 +(5−1)2.61
10+5−2 = 3.492.
√
(c) SE(X ¯1 − X ¯ 2 ) = 3.492 1/10 + 1/5 = 1.0235.
p
(d) −8.4 ± 2.160 × 1.0235, which works out to −8.4 ± 2.211 or (−10.61, −6.19).
(e) H0 : µ1 = µ2 , Ha :√µ1 = 2.
6 µp
SE(X ¯1 − X¯ 2 ) = 3.492 1/10 + 1/5 = 1.023523. t = 8.8−17.2 = −8.21.
1.0235
The p-value is double the area to the left of −8.21 under a t distribution with
10+5−2 = 13 degrees of freedom. Using software: p-value = 1.7×10−6 . Using
a t table: p-value < 0.001. There is statistically signifcant evidence against
the null hypothesis (p-value = 1.7 × 10−6 ) at α = 0.05. There is statistically
signifcant evidence that the population means are not equal.
4. (a) H0 : µ1 = µ2 (Vitamin D has no e˙ect on parathyroid hormone levels).

Ha : µ1 =
6 µ2 (Vitamin D has an e˙ect on parathyroid hormone levels).
2 (14−1)37.52 +(12−1)34.62
sp = 14+12−2 = 1310.417
√
¯1 − X ¯ 2 ) = 1310.417 1/14 + 1/12 = 14.24088. t = −9.0−(−1.6) =
p
SE(X 14.24088
−0.52.
The p-value is double the area to the left of −0.52 under a t distribution with
14 + 12 − 2 = 24 degrees of freedom. Using software: p-value = 0.61. Using a
10.4.
Balka HYPOTHESIS
ISE 1.09 TESTS AND CONFIDENCE INTERVALS FOR µ1 − µ2 (WHEN σ1 AND σ2 ARE UNKNOWN) 155
t table: p-value > 0.40.

There is no evidence (p-value = 0.61) that vitamin D has an e˙ect on PTH
levels (there is no evidence of a di˙erence in population mean PTH levels
between the groups).
(b) −9.0−(−1.6)±2.064×14.24088, which works out to −7.4±29.39, or (−36.79, 21.99).
We can be 95% confdent that the di˙erence in the population mean change
in PTH lies between −36.79 pg/mL and 21.99 pg/mL. (It should come as no
surprise that this interval contains 0, as the hypothesis test in part a) showed
no evidence of a di˙erence in means.)
(c) There is no evidence of an e˙ect of vitamin D on PTH levels in the blood
(p-value = 0.61). The point estimate of the di˙erence in means (Vitamin D
vs control) was −7.4 pg/mL, with a corresponding 95% confdence interval of
(−36.79, 21.99).
10.4.2 Welch (Unpooled Variance) t Tests and Confdence Inter-

vals
5. (a) X¯1 − X¯ 2 = 8.8 − 17.2 = −8.4.

q
(b) SEW (X ¯1 − X¯ 2 ) = 1.422 + 2.612 = 1.2506. (Note that this is a little di˙erent
10 5
from the standard error of the pooled-variance procedure.)
(c) To construct the interval, we need to fnd t0.025 with 5.221 degrees of freedom.
Using software, we can fnd that with 5.221 degrees of freedom, t0.025 = 2.538.
Using a t table, we would round the degrees of freedom down to 5, and fnd
that t0.025 = 2.571.
Using software, the interval is: −8.4 ± 2.538 × 1.2506, which works out to
−8.4 ± 3.174 or (−11.57, −5.23).
Using a t table (with 5 DF), the interval is: −8.4±2.571×1.2506, which works
out to −8.4 ± 3.215 or (−11.62, −5.18). (There is, of course, a little rounding
error here.)
(d) H0 : µ1 = µ2 , Ha : µq 6 µ2 .
1 =
2 2.612
¯ ¯
SEW (X1 − X2 ) = 1.42 8.8−17.2
10 + 5 = 1.2506. t = 1.2506 = −6.72.
The p-value is double the area to the left of −6.72 under a t distribution
with 5.221 degrees of freedom. Using software: p-value = 0.0009. Using a t
table (with 5 DF): 0.001 < p-value < 0.002. There is statistically signifcant
evidence against the null hypothesis (p-value = 0.0009) at α = 0.05. There is
statistically signifcant evidence that the population means are not equal.
q
6. (a) SEW (X ¯1 − X¯ 2 ) = 18.72 + 1.72 = 5.0218.
14 12
(b) t = −6.3−(−0.2)
5.0218 = −1.215
(c) There is little or no evidence (p-value = 0.245) of an e˙ect of vitamin D on
Balka ISE 1.09 10.5. PAIRED-DIFFERENCE PROCEDURES 156
phosphorous levels (little or no evidence of a di˙erence in population mean

phosphorous levels between the groups). The point estimate of the di˙erence
in change in phosphorous levels (Vitamin D vs control) was −6.1 mg/dL, with
a corresponding 95% confdence interval of (−16.9, 4.7).
10.4.3 Guidelines for Choosing the Appropriate Two-Sample t Pro-

cedure
7. Opinions di˙er a great deal on the choice between the Welch procedure and the
pooled-variance t procedure. Both procedures assume normally distributed pop-
ulations. The pooled-variance procedure also assumes that the populations have
equal variances. If the sample variances are very di˙erent, then this gives some
indication that the assumption of equal population variances is not reasonable, and
the pooled-variance procedure may not be performing well. The e˙ect of di˙erent
population variances is exaggerated if the sample sizes are also very di˙erent. So
if the sample variances are very di˙erent, and the sample sizes are very di˙erent,
then the pooled-variance procedure is not an appropriate method of analysis. If the
sample variances are similar, and the sample sizes are similar, then the choice be-
tween the Welch procedure and the pooled-variance t procedure is largely a matter
of personal preference.
8. D, since the pooled-variance procedure performs poorly in this situation.
10.5 Paired-Di˙erence Procedures
10.5.1 Paired-Di˙erence t Tests and Confdence Intervals
9. (a) We should use a paired-di˙erence procedure, since the before and after mea-
surements represent two measurements on the same pig. (The before and after
measurements do not represent independent samples.)
(b) The 9 After − Before measurements are:
935 − 685 = 250

700 − 545 = 155
770 − 480 = 290
640 − 475 = 165
800 − 680 = 120
955 − 685 = 270
780 − 590 = 190
790 − 600 = 190
830 − 630 = 200
(c) X
¯ = 250+155+290+165+120+270+190+190+200 = 203.333 mL. s = 56.1805 mL.
9
(d) SE(X
¯ ) = 56.1805
√
9
= 18.7268 mL.
(e) The boxplot and normal QQ plot look good. There are no obvious outliers,
skewness, or other major problems. The sample size is quite small (n = 9),
but it is not unreasonable to use the t procedures here.
(f) Using software, we can fnd that with 8 degrees of freedom, t0.025 = 2.306004.
(If we must use a t table, we can fnd the rounded t value of 2.306.)
¯ ± t0.025 SE(X
X ¯)
203.333 ± 2.306 × 18.7268
203.333 ± 43.184
We can be 95% confdent that the true mean increase in leg volume, for this
type of pig under the conditions of this experiment, lies between 160.15 mL
and 246.52 mL.
(g) The hypotheses are:
H0 : µ = 0 (The true mean change in leg volume (under the conditions of this experiment) is 0.)
Ha : µ > 0 (The true mean change in leg volume is greater than 0.)
¯ − µ0
X
t= ¯)
SE(X
203.333 − 0
=
18.7268
= 10.8579
The p-value is the area to the right of 10.8579 under a t distribution with
n − 1 = 9 − 1 = 8 degrees of freedom. Using software, we can fnd: p-value =
2.3 × 10−6 . Using a t table, we could fnd only a range: p-value < 0.0005.
A p-value that is this small gives very strong evidence against the null hy-
pothesis. (The evidence would be statistically signifcant at any of the usual
signifcance levels.) There is very strong evidence that, on average, pigs treated
with a dose of venom and antivenom experience swelling in the a˙ected area.
10. (a) The data should be plotted. Plotting the data helps us get a feel for what the
numbers are telling us, and can help us to determine whether the normality
assumption is reasonable.
H0 : µ = 0 (The true mean change in discrimination accuracy is 0.)

Ha : µ > 0 (The true mean change in discrimination accuracy is greater than 0.)
(In loose terms, the null hypothesis is that the short training program did not
help people discriminate between genuine and fake smiles, and the alternative
hypothesis is that the training program (on average) helped people to discrim-
inate between genuine and fake smiles.)
The sample mean fell in the direction given in the alternative hypothesis, so
the one-sided p-value is half of the reported two-sided p-value. One-sided
p-value = 0.0348
2 = 0.0174.
(c) There is statistically signifcant evidence at α = 0.05 (one-sided p-value =
0.0174) that the training program results in improved discrimination accuracy
scores on average. The point estimate of the true mean change in discrimina-
tion accuracy is 0.618, with an associated 95% confdence interval of 0.048 to
1.188.
Balka ISE 1.09 10.6. INVESTIGATING THE NORMALITY ASSUMPTION 159
10.6 Investigating the Normality Assumption
10.7.2 Concepts
11. SE(X ¯1 − X¯ 2 ) is the estimate of the standard deviation of the sampling distribution
of X
¯1 − X¯ 2 . (It estimates σ ¯ 2 ). It is a measure of the dispersion of X
X̄1 −X
¯ 2 in
¯1 − X
repeated sampling.
12. No, this would not make sense. Statistical hypotheses always involve parameters
and never statistics.
13. (a) Any hypothesized value outside of the 95% interval would be rejected at α =
0.05. For example, H0 : µ1 − µ2 = 42 would be rejected.
(b) Any hypothesized value within the 95% interval would not be rejected at α =
0.05. For example, H0 : µ1 − µ2 = 20 would not be rejected.
14. Yes, of course X¯ 1 does not equal X¯ 2 . The output shows that X¯ 1 = −0.127 and
X2 = 0.082, so we know the sample means are not equal. However, in the test
¯
of H0 : µ1 = µ2 , there is not any evidence against the null hypothesis (p-value =
0.5763).
15. (a) All else being equal, as µ1 − µ2 gets closer to 0, the power of the test will
decrease (it will be harder to detect a di˙erence if the di˙erence is small).
(b) All else being equal, as the sample size increases, the power of the test will
increase.
16. (a) If the sample means are equal (X ¯1 = X¯ 2 ), then t = 0.

(b) The p-value would equal 1 if t = 0 (which occurs if X ¯ 2 ). If the sample
¯1 = X
means are equal, then one could not possibly get a sample with less evidence
against the null hypothesis, with a resulting p-value of 1.
(c) The p-value would be equal to exactly 0 if the test statistic was infnite (which
would only occur if the standard error was 0, implying no variance within each
sample). In practice, this will never occur, although the p-value at times may
be very, very, very close to 0.
17. (a) H0 : µA = µC (II) would have the smallest p-value, since it shows the greatest
evidence against the null hypothesis.
(b) H0 : µA = µB (I) would have the largest p-value, since it shows the least
18. (a) The boxplots do not give any indication that the pooled-variance t procedure
is inappropriate. (The pooled-variance t procedure is a reasonable choice.)
There is one small outlier in Sample B, which may distort things a little bit,
but it is not very extreme and it is not likely to have a major impact. The
variances of the two samples are similar, so the equal variance assumption may
not be too far o˙ the mark.
(b) No, there is not strong evidence (p-value = 0.29) of a di˙erence in population
means.
(c) Yes. We can be 95% confdent that µ1 − µ2 lies between −1.69 and 0.52.
Since 10 falls far outside of this interval, there is very strong evidence that the
di˙erence in population means does not equal 10.
(d) For the pooled variance procedure, df = n1 + n2 − 2 = 43 (from the output).
This implies that n1 + n2 = 45.
19. (a) Output 1 is from the Welch procedure and Output 2 is from the pooled-
variance procedure. (One can tell by looking at the degrees of freedom.)
(b) The conclusions are nearly identical, so it matters little which procedure is
used. The sample standard deviations are not given in the output, but they
were likely similar.
20. (a) Of the given choices, the paired-di˙erence procedure is the most appropriate
one. We have two measurements on each individual, so in this situation it
would not be appropriate to use Welch’s t or the pooled-variance t (since
they assume that the samples are independent). We need to use the paired-
di˙erence approach.
(b) The fact that the placebo injection was given frst, then the drug two days
later, is a cause for concern. It is conceivable (and not a major stretch), to
think that the individuals may be more nervous on the frst day of the study,
at the frst injection. Perhaps by the second day, the participants were more
comfortable with the study and more relaxed. This may very well infuence
the results. It would be much better to randomize the order in which they
received each treatment (the drug or the placebo).
21. We should use the paired-di˙erence approach, since there is a dependence within
each pair of twins. (Twins are more similar than non-twins.)
q q
2 2
22. Here we require 1.96 3n + 3n ≤ 0.5. This implies that 1.96 18 n ≤ 0.5 =⇒ n ≥
√
( 1.960.5 18 )2 =⇒ n ≥ 276.5. We would need to sample at least 277 members from
each population.
23. (a) When we use the pooled-variance t procedure, it is because we know the pop-
ulations have the same variance. False. We are assuming that the (unknown)
population variances are equal.
(b) The pooled-variance t procedure works well, even when the population vari-
ances are a little di˙erent. This is especially true if the sample sizes are similar.
True.
(c) It would be most appropriate to use the Welch procedure instead of the pooled-
variance t procedure if the sample variances are very di˙erent and the sample
sizes are very di˙erent. True.
(d) If the conclusions from the Welch procedure and the pooled-variance t proce-
dure are very similar, then it does not matter much which procedure is used.
True.
(e) The Welch procedure is an exact procedure, as long as X ¯1 = X ¯ 2 . False. The
Welch procedure is an approximate procedure.
24. (a) If µ1 = µ2 , the sampling distribution of X ¯ 2 is approximately symmetric

¯1 − X
about 0 for large sample sizes. True. The sampling distribution of X ¯1 − X¯2
has a mean of µ1 − µ2 , and is approximately normal for large sample sizes.
(b) If X
¯1 = X ¯ 2 , the sampling distribution of µ1 − µ2 is approximately symmetric
about 0 for large sample sizes. False. µ1 − µ2 is a fxed quantity, and does not
have a sampling distribution.
(c) SE(X ¯1 − X ¯ 2 ) is the true standard deviation of the sampling distribution of
X¯1 − X¯ 2 . False. It is the estimate of this quantity.
(d) The pooled-variance t procedures work well, even when the variances for the
two populations are very di˙erent, as long as the sample sizes are very di˙erent
as well. False. The pooled-variance procedure performs poorly in this situation.
(e) Suppose we are interested in testing the null hypothesis µ1 = µ2 , against a
two-sided alternative. All else being equal, the greater the di˙erence between
µ1 and µ2 , the greater the power of the test. True. The greater the di˙erence,
the easier it is to detect that di˙erence.
25. (a) Suppose we are constructing a confdence interval for µ1 − µ2 . All else being
equal, the greater the di˙erence between µ1 and µ2 , the wider the interval.
False. The width does not depend on µ1 − µ2 .
(b) Suppose we are constructing a confdence interval for µ1 − µ2 . All else being
equal, the greater the di˙erence between X ¯ 1 and X
¯ 2 , the wider the interval.
¯ ¯ ¯
False. The width does not depend on X1 − X2 . (X1 − X ¯ 2 is the midpoint of
the interval.)
(c) Suppose we are constructing a confdence interval for µ1 − µ2 . All else being
equal, the greater the sample sizes, the narrower the interval. True.
(d) Suppose we wish to test H0 : µ1 = µ2 . We obtain random samples from the
respective populations, run the appropriate test, and fnd that the p-value is
0.00000032. We can be very confdent that our results have important practical
implications. False. There is strong evidence of a di˙erence in population
means, but whether this is important in a practical sense is an entirely di˙erent
question.
(e) If we test H0 : µ1 = µ2 against a two-sided alternative and fnd a p-value of
0.32, then we know that µ1 = µ2 . False. We do not have any evidence of a
di˙erence in population means, but we do not know the population means, and
we do not know if they are equal.
10.7.3 Applications
26. (a) The t procedures should be reasonable here. The boxplots appear to show
that the distributions are roughly symmetric, but there is one (small) outlier
in the OE group. The normal QQ plots are very roughly linear, although
the OE group shows hints of possible left skewness (mainly due to the one
small outlier). Overall the plots look reasonable, and since the two-sample t
procedures are quite robust to violations of the normality assumption, the t
procedures will likely perform reasonably well in this scenario.
(b) The sample standard deviations are not that similar (0.37 vs 0.60), and the
sample sizes are quite di˙erent (15 vs 51), so it is probably best to use the
Welch (unpooled variance) approach. (Opinions may di˙er on this matter.)
(c) Using software, we can fnd that with 37.714 degrees of freedom t0.025 = 2.0249.
(If we used the table or software with 37 degrees of freedom, t0.025 = 2.026.)
¯1 − X
X ¯ 2 ± t0.025 SEW (X ¯2)
¯1 − X
1.14 − 1.80 ± 2.0249 · 0.1272
−0.66 ± 0.258
which works out to (−0.92, −0.40).

(d) H0 : µ1 = µ2 (The true mean running speed for pairs led by an OE ant is equal
to the true mean running speed for pairs led by a YE ant.)
Ha : µ 1 =
6 µ2 (The true mean running speed for pairs led by an OE ant is not
equal to the true mean running speed for pairs led by a YE ant.)
(e)
X¯1 − X¯2
t= ¯1 − X¯2)
SEW (X
1.14 − 1.80
=
0.1272
= −5.19
(f) The p-value is double the area to the left of −5.19 under a t distribution
with 37.714 degrees of freedom. Using software, we can fnd that the p-value
is 7.5 × 10−6 . Using a table, we could fnd only a range for the p-value:
p-value < 0.001. This is a very small p-value, implying very strong evidence
against the null hypothesis.
(g) There is very strong evidence (p-value = 7.5 × 10−6 ) of a di˙erence in the
true mean speed of YE and OE leaders in tandem running for Temnothorax
albipennis under the conditions of the study. The 95% confdence interval
indicates that we can be 95% confdent that the di˙erence (µOE − µY E ) in
true mean speed lies between −0.92 and −0.40 mm/s. In short, there is very
strong evidence that, on average, tandem running pairs led by an old and
experienced ant move more slowly than pairs lead by a young and experienced
ant.
Although it wasn’t clear from the question how the tandem running pairs were
chosen, it is reasonable to think that these results apply to Temnothorax al-
bipennis ants under the conditions of the study. The means of tandem running
speed may very well change if the conditions change, or if a di˙erent type of
ant is studied.
27. (a) There are a variety of options, but we should (at least) plot boxplots and nor-
mal quantile-quantile plots. This would help us visualize possible di˙erences
between the groups, and investigate whether the normality assumption is rea-
sonable. (The sample sizes are fairly large here (246 and 46), so the normality
assumption is not overly important, but we should still investigate the shape
of the distribution and ensure there are no major outliers.)
(b) The sample standard deviations are very similar here (0.035 vs 0.034), so the
pooled-variance procedure would be a reasonable choice. The Welch (unpooled
variance) approach would also be reasonable. (Opinions may di˙er on this
matter.)
(c) There are n1 + n2 − 2 = 246 + 46 − 2 = 290 degrees of freedom. Using software
or a t table, we can fnd that with 290 degrees of freedom t0.025 = 1.968.
¯1 − X
X ¯1 − X
¯ 2 ± t0.025 SE(X ¯2)
0.994 − 0.963 ± 1.968 · 0.005598
0.031 ± 0.0110

(d) H0 : µ1 = µ2 (The true mean 2D:4D ratio for white women is equal to the true
mean 2D:4D ratio for black women.)
Ha : µ1 =
6 µ2 (The true mean 2D:4D ratio for white women is not equal to the
true mean 2D:4D ratio for black women.)
(e)
X¯1 − X¯2
t= ¯1 − X
¯2)
SE(X
0.994 − 0.963
=
0.005598
= 5.538
(f) The p-value is double the area to the right of 5.538 under a t distribution
with 290 degrees of freedom. Using software, we can fnd that the p-value
is 6.8 × 10−8 . Using a table, we could fnd only a range for the p-value:
p-value < 0.001. This is a very small p-value, implying very strong evidence
(g) There is very strong evidence (p-value = 6.8 × 10−8 ) of a di˙erence in the
true mean 2D:4D ratio for white and black women in Manchester. The 95%
confdence interval indicates that we can be 95% confdent that the di˙erence
in true mean 2D:4D ratio (µW − µB ) lies between 0.020 and 0.042. In short,
there is very strong evidence that, on average, white women have a greater
mean 2D:4D ratio than that of black women.
The women were drawn from people attending a sexual health clinic in Manch-
ester, UK. It is reasonable to think that these results would apply to women
who attend this clinic. Extrapolating to an even larger population is some-
what dubious, but it is likely reasonable to think that these results apply to
white and black women in the Manchester, UK area. Although these results
give some indication of the di˙erence in this ratio for white and black women
in general, generalizing beyond the area from which the sample was drawn is
always dubious, and may lead to incorrect conclusions.
28. (a) There are a variety of options, but we should (at least) plot boxplots and
normal quantile-quantile plots. This would help us visualize possible di˙er-
ences between the groups, and investigate whether the normality assumption
is reasonable. (The sample sizes are small here (8 and 7) so the normality
assumption is important. Also, outliers could have a strong e˙ect on the re-
sults.)
(b) The sample standard deviations are very di˙erent here (25.4 vs 7.9), so the
Welch (unpooled variance) approach would likely be the best option. (Opinions
may di˙er on this matter.)
(c) The degrees of freedom are given as 8.512. Using software, we can fnd that
with 8.512 degrees of freedom t0.025 = 2.2821. (If we had to use a t table with
8 degrees of freedom, t0.025 = 2.306.

¯1 − X
X ¯1 − X
¯ 2 ± t0.025 SE(X ¯2)
48.9 − 23.8 ± 2.2821 · 9.4637
25.1 ± 21.597
which works out to (3.5, 46.7). (Had we used the table value of t0.025 = 2.306,
we’d have found an interval of (3.3,46.9).)
(d) H0 : µ1 = µ2 (The true mean number of lever presses for the restricted food
group is equal to the true mean number of lever presses for the sated group.)
Ha : µ1 6= µ2 (The true mean number of lever presses for the restricted food
group is not equal to the true mean number of lever presses for the sated
group.)
(e)
X¯1 − X¯2
t= ¯ ¯2)
SE(X1 − X
48.9 − 23.8
=
9.4637
= 2.652
(f) The p-value is double the area to the right of 2.652 under a t distribution with
8.512 degrees of freedom. Using software, we can fnd that the p-value is 0.0276.
Using a table, we could fnd only a range for the p-value: 0.02 < p-value < 0.05.
There is strong, but not overwhelming evidence against the null hypothesis.
The evidence against the null hypothesis would be signifcant at the commonly
chosen signifcance level of 0.05.
(g) There is strong evidence (p-value = 0.0276) of a di˙erence in the true mean
number of lever presses under the conditions of this experiment. The 95%
confdence interval indicates that we can be 95% confdent that the di˙erence
in true mean number of lever presses (µF R − µS ) lies between 3.5 and 46.7.
In short, there is strong evidence that, on average, the food restricted group
presses the heroin lever more often.
The conclusions apply to the type of rat used in the study, under the condi-
tions of the experiment. Using a di˙erent type of rat (or another animal), or
changing the conditions of the study, could easily change the means a great
deal.
sonable. (The sample sizes are not small here (66 and 28), so the normality
of the distribution.)
(b) The sample standard deviations are very similar here (1.50 and 1.64), so the
variance) approach would also be reasonable. (Opinions may di˙er on this
matter.)
(c) There are n1 + n2 − 2 = 66 + 28 − 2 = 92 degrees of freedom. Using software,
we can fnd that with 92 degrees of freedom t0.025 = 1.9861. (If we had to use
a t table, we may have to look up 90 DF, and fnd t0.025 = 1.987.)
¯1 − X
X ¯ 2 ± t0.025 SE(X ¯2)
¯1 − X
95.71 − 95.41 ± 1.9861 · 0.3479
0.30 ± 0.691
which works out to (−0.39, 0.99). (Had we used t0.025 = 1.987 from the t table,
we’d have gotten the same interval to 2 decimal places.)
(d) H0 : µ1 = µ2 (The true mean 3D symmetry for men is equal to the true mean
3D symmetry for women.)
Ha : µ1 6= µ2 (The true mean 3D symmetry for men is not equal to the true
mean 3D symmetry for women.)
(e)
¯1 − X
X ¯2
t= ¯ ¯2)
SE(X1 − X
95.71 − 95.41
=
0.3479
= 0.862
92 degrees of freedom. Using software, we can fnd that the p-value is 0.39.
Using a table, we could fnd only a range for the p-value: p-value > 0.20. This
p-value is large, yielding no evidence against the null hypothesis.
(g) There is no evidence (p-value = 0.39) of a di˙erence in the true mean 3D
symmetry for men and women. The 95% confdence interval indicates that we
can be 95% confdent that the di˙erence in true mean 3D symmetry (µM −µW )
lies between −0.39 and 0.99. There is no evidence that, on average, men have
di˙erent mean 3D symmetry than that of women.
The people in the sample were drawn from healthy Italian Caucasians between
31 and 40 years of age. Depending on the sampling design, it is likely reason-
able to think that these results apply to healthy Italian Caucasians between
31 and 40 years old.
sonable. (The sample sizes are not small here (47 and 27), so the normality
of the distribution.)
(b) The sample standard deviations are similar here (33.53 and 37.55), so the
variance) approach would also be reasonable.
(c) There are n1 + n2 − 2 = 47 + 27 − 2 = 72 degrees of freedom. Using software,
we can fnd that with 72 degrees of freedom t0.025 = 1.9935. (If we had to use
a t table, we may have to look up 70 DF, and fnd t0.025 = 1.994.)
¯1 − X
X ¯1 − X
¯ 2 ± t0.025 SE(X ¯2)
256.12 − 245.82 ± 1.9935 · 8.4603
10.3 ± 16.867
which works out to (−6.6, 27.2). (Had we used t0.025 = 1.994 from the t table,
we’d have gotten the same margin of error and interval to 2 decimal places.)
(d) H0 : µ1 = µ2 (The true mean of the BEST index for psychopaths is equal to
the true mean of the BEST index for nonpsychopaths.)
Ha : µ1 6= µ2 (The true mean of the BEST index for psychopaths is not equal
to the true mean of the BEST index for nonpsychopaths.)
(e)
¯1 − X
X ¯2
t= ¯1 − X
¯2)
SE(X
256.12 − 245.82
=
8.4603
= 1.217
72 degrees of freedom. Using software, we can fnd that the p-value is 0.2274.
Using a table, we could fnd only a range for the p-value: p-value > 0.20. This
p-value is large, yielding little or no evidence against the null hypothesis.
(g) There is no evidence (p-value = 0.22) of a di˙erence in the true mean BEST
index of psychopaths and nonpsychopaths. The 95% confdence interval in-
dicates that we can be 95% confdent that the di˙erence in true mean BEST
index (µP − µN P ) lies between −6.6 and 27.2. There is no evidence that, on
average, psychopaths have a di˙erent mean BEST index than that of nonpsy-
chopaths.
The people in the sample were males drawn from an inpatient psychiatric
treatment facility in Holland. Technically the conclusions apply only to male
psychiatric patients from this facility. (The results give some indication of the
reality of the situation for male inpatient psychiatric patients in general, but
generalizing to a population outside of the one sampled is always dubious.)
31. (a) SE(X ¯ ) = √s = √3.7 = 1.11559.

n 11
Using software, we can fnd that with 10 degrees of freedom, t0.025 = 2.228139.
(If we must use a t table, we can fnd the rounded t value of 2.228.) The
endpoints of a 95% confdence interval for µ are given by:
¯ ± t0.025 SE(X
X ¯)
3.3 ± 2.228 × 1.11559
3.3 ± 2.486
We can be 95% confdent that, when this type of pig is Tasered, the true mean
change in mean corpuscular volume lies between 0.814 fL and 5.786 fL.
H0 : µ = 0 (The true mean change in mean corpuscular volume is 0.)

Ha : µ 6= 0 (The true mean change in mean corpuscular volume di˙ers from 0.)
X¯ − µ0
t= ¯)
SE(X
3.3 − 0
=
1.11559
= 2.958
The p-value is double the area to the right of 2.958 under a t distribution with
n − 1 = 11 − 1 = 10 degrees of freedom. Using software, we can fnd: p-value =
0.0143. Using a t table, we could fnd only a range: 0.01 < p-value < 0.02.
A p-value of this size gives moderately strong evidence against the null hypoth-
esis. (The evidence would be statistically signifcant at the commonly chosen
signifcance level of 0.05.) There is moderately strong evidence that, on aver-
age, Tasered pigs of this type experience an increase in their mean corpuscular
volume.
32. (a) 3468−3200±1.684×240.8803, which works out to 268±405.6 or (−137.6, 673.6).

(b) ii) is the most appropriate conclusion.
33. (a) This is an experiment (since the researchers imposed a treatment, the exercise
wheel, on the rats).
(b) H0 : µ1 = µ2 (the tumour size is the same on average for both groups).
Ha : µ1 < µ2 (the exercise group has a lower mean tumour size). (Here the
researchers are only interested if the wheels tend to reduce tumour size, but
one could certainly make an argument for a two-sided alternative hypothesis).
s2p = (10−1)0.08+(20−1)0.12
10+20−2 = 0.10714.
√
¯ ¯
p
SE(X1 − X2 ) = 0.10714 1/10 + 1/20 = 0.12677.
0.12677 = −2.366. The p-value is the area to the left of −2.366 under a t
t = 1.2−1.5
distribution with 28 degrees of freedom. By computer: p-value = 0.0126. By
the table: 0.01 < p-value < 0.025.
Here we do not have a given value of α, but we should still be able to say
something reasonable. There is strong evidence (p-value = 0.0126) of a di˙er-
ence in population mean tumour size. There is strong evidence that those rats
in the exercise group have a lower population mean size of tumour.
(c) Yes, absolutely. A p-value that small gives very, very strong evidence against
the null hypothesis. Since this was a designed experiment, we can say that
there is evidence of a causal relationship (the exercised wheels caused the
observed di˙erence).
(d) 1.2−1.5±1.701×0.12677, which works out to −0.3±0.2156 or (−0.516, −0.084).
We can be 90% confdent that the true di˙erence in mean tumour size lies be-
tween −0.516 and −0.084.
(e) H0 : µ1 = µ2 (the tumour size is the same on average for both groups).
Ha : µ1 < µ2 (the exercise group has a lower mean tumour size). (Here the
researchers are only interested if the wheels tend to reduce tumour size, but
one could certainlyp make an argument for a two-sided alternative hypothesis).
SEW (X ¯ 2 ) = 0.08/10 + 0.12/20 = 0.1183.
¯1 − X
0.1183 = −2.535. The p-value is the area to the left of −2.535 under a t
t = 1.2−1.5
distribution with 21.764 degrees of freedom.
By computer: p-value = 0.0095. By the table (with 20 DF): 0.005 < p-value <
0.01.
Here we do not have a given value of α, but we should still be able to say
something reasonable. There is strong evidence (p-value = 0.0095) of a di˙er-
ence in population mean tumour size. There is strong evidence that those rats
in the exercise group have a lower population mean size of tumour.
(f) By computer (with 21.764 DF): 1.2 − 1.5 ± 1.718 × 0.1183, which works out
to −0.3 ± 0.2032 or (−0.503, −0.097).
By the table (with 21 DF): 1.2 − 1.5 ± 1.721 × 0.1183, which works out to
−0.3 ± 0.2036 or (−0.504, −0.096).
(g) The choice of appropriate procedure here would be debatable. There is not a
large di˙erence between the variances, so the pooled-variance procedure would
not be a terrible choice. But one could make an argument for each one, so the
choice comes down largely to a matter of personal preference.
34. (a) The last statement (v) is the best conclusion. The other statements are not
reasonable.
(b) We can be 95% confdent that the di˙erence in population mean cholesterol
levels between males and females (µM − µF , for students at this university)
lies between −21.9 and 17.1 mg/dl.
35. (a) s2p = (12−1)16+(6−1)9

12+6−2 √ = 13.8125.
¯1 − X ¯ 2 ) = 13.8125 1/12 + 1/6 = 1.8583.
p
SE(X
A 95% confdence interval is 63 − 57 ± 2.120 × 1.8583, which works out to
6 ± 3.94 or (2.06, 9.94). We can be 95% confdent that the di˙erence in
population mean contaminant levels (µA − µB ) lies between 2.06 and 9.94.
(b) H0 : µ1 = µ2 (the mean contaminant level is the same for both groups).
Ha : µ1 =6 µ2 (there is a di˙erence in the mean contaminant level between the
groups). √
¯1 − X ¯ 2 ) = 13.8125 1/12 + 1/6 = 1.8583.
p
SE(X
1.8583 = 3.23. The p-value is double the area to the right of 3.23 under a t
t = 63−57
distribution with 16 degrees of freedom. By computer: p-value = 0.0053. By
(c) There is strong evidence (p-value = 0.0053) of a di˙erence in populate mean
contaminant levels. The point estimate of the di˙erence in means is 6, with
a corresponding 95% confdence interval of (2.06, 9.94). It appears as though
process B results in a lower contaminant level on average.
36. (a) H0 : µ1 = µ2 (the population mean “fear of negative evaluation” is the same
for bulimics and non-bulimics).
Ha : µ1 6= µ2 (the population mean “fear of negative evaluation” is di˙erent
between bulimics and non-bulimics).
(b) There is signifcant evidence against the null hypothesis. There is signifcant
evidence of a di˙erence in population mean “fear of negative evaluation” score.
It appears as though bulimics tend to have a higher fear of negative evaluation
score.
(c) We can be 95% confdent that the di˙erence in population mean “fear of eval-
uation” scores lies between 0.94 and 8.80.
37. (a) No, these plots give no indication of a violation of the normality assumption.
Although the sample sizes are small, the use of the t procedures is reasonable
here.
2 +(16−1)0.422
(b) s2p = (12−1)0.65
12+16−2 = 0.2805.
(c) H0 : µ1 = µ2 (The population mean resistance is the same for both types of
resistor.)
Ha : µ1 6= µ2 (The population mean resistance di˙ers between the two types
of resistor.)
√
(d) ¯1 − X ¯ 2 ) = 0.2805 1/12 + 1/16 = 0.2023.
p
SE(X
t = 11.6−12.2
0.2023 = −2.967.
(e) The p-value is double the area to the left of −2.967 under a t distribution
with 26 degrees of freedom. By computer: p-value = 0.0064. By the table:
0.002 < p-value < 0.010. There is signifcant evidence at the 5% signifcance
level (p-value < 0.05) of a di˙erence in population mean resistance. It appears
as though the resistors from Manufacturer B have a greater mean resistance.
(f) 11.60−12.20±2.056×0.2023, which works out to −0.6±0.416 or (−1.016, −0.184).
(g) We can be 95% confdent that the di˙erence in the population mean resistance
(µA − µB ) lies between −1.052 and −0.148.
(h) The output tells us that double the area to the left of −2.7904 under the t
distribution is 0.01222. Since for the one-sided alternative Ha : µA < µB we
require the area to the left of −2.7904, the p-value is 0.01222
2 = 0.0061.
2 2
38. (a) s2p = (80−1)23.3 +(75−1)19.8
80+75−2 = 469.9299.
√
¯1 − X ¯ 2 ) = 469.9299 1/80 + 1/75 = 3.4842.
p
SE(X
A 95% confdence interval is given by 664.2 − 624.3 ± 1.976 × 3.4842, which
works out to 39.9 ± 6.88 or (33.02, 46.78). We can be 95% confdent that the
di˙erence in population mean yield strength (µN ew − µStandard ) lies between
33.02 MPa and 46.78 MPa.
(b) H0 : µ1 = µ2 (the population mean yield strength is the same for both meth-
ods).
Ha : µ1 =6 µ2 (the √ populationpmean yield strength di˙ers between the methods).
¯ ¯
SE(X1 − X2 ) = 469.9299 1/80 + 1/75 = 3.4842.
t = 664.2−624.3
3.4842 = 11.45.
80 + 75 − 3 degrees of freedom. Using software: p-value = 2.6 × 10−22 . Using
a t table: We can say that the p-value is close to 0 (p-value < 0.001).
There is extremely strong evidence (p-value = 2.6 × 10−22 ) of a di˙erence in
population mean yield strength between the two procedures. In fact, there is
very strong evidence that the heat treated method results in a greater mean
yield strength.
39. (a) There is strong evidence (p-value = 0.028) of a di˙erence on average between
the stated and actual weight of the chickens. The point estimate of the dif-
ference (Stated − Actual) is 22.8, with an associated 95% confdence interval
of 4.1 to 41.5 grams. In loose terms, it appears as though the actual weight
tends to be lower than the stated weight.
(b) If the normality assumption is violated then our reported results may be mis-
leading. What we are stating is a 95% confdence interval will have a true
confdence level that di˙ers from 95%. And our reported p-value may be very
di˙erent from the “true” p-value.
40. (a) Here we have 2 measurements on each gas sample, and thus these measure-
ments cannot be considered independent. We need to use a paired-di˙erence
procedure.
(b) We could try plotting a normal quantile-quantile plot of the 6 di˙erences (but
with only 6 points, the plot may not be very informative). It will be a little
sketchy using the t procedures on such a small sample.
(c) The 6 di˙erences (Old − New) are 0.03, 0.10, −0.12, 0.05, −0.08, 0.08. These 6
di˙erences have a mean of X ¯ = 0.01 and a standard deviation of s = 0.08944.
¯
SE(X ) = 0.08944
√
6
= 0.0365. A 95% confdence interval for the population
mean di˙erence (Old − New) is 0.01 ± 2.571 × 0.0365, which works out to
0.01 ± 0.094 or (−0.084, 0.104). We can be 95% confdent that the population
mean di˙erence (Old method − New method) lies between −0.084 and 0.104.
(d) H0 : µ = 0 (There is no di˙erence between the two methods on average).
Ha : µ =6 0 (There is a di˙erence between the two methods on average).
0.01−0
t = 0.0365 = 0.274.
The p-value is double the area to the right of 0.274 under a t distribution
with 5 degrees of freedom. By computer: p-value = 0.80. By the table:
p-value > 0.40. There is no evidence (p-value = 0.80) that the population
mean di˙erence di˙ers from 0. We have no evidence of a di˙erence between
the viscosity measurements on average.
41. (a) If we take the di˙erences as After − Before:

H0 : µ = 0 (there is no change in reaction time on average).
Ha : µ > 0 (there is an increase in reaction time on average).
The 4 di˙erences are: 0.37, 0.25, 0.42, 0.06. These values have a mean of X¯ =
0.275 and a standard deviation of s = 0.1601. SE(X ¯) = √ 0.1601
4
= 0.0801.
0.0801 = 3.44. The p-value is the area to the right of 3.44 under a
t = 0.275−0
t distribution with 3 degrees of freedom. Using software: p-value = 0.021.
Using a t table: 0.01 < p-value < 0.025.
There is signifcant evidence at at the 5% level that, on average, alcohol in-
creases reaction times (under the conditions of this experiment).
(b) 0.275 ± 3.182 × 0.0801, which works out to 0.275 ± 0.255 or (0.02, 0.53). We
can be 95% confdent that the true mean increase in reaction times, under the
conditions of this experiment, lies between 0.02 and 0.53.
(c) Setting up the experiment in this way serves to reduce the variability in the
experiment. By taking two measurements on the same person, we reduce the
overall variability, and make it easier to isolate the e˙ect of the alcohol.
(d) Before-and-after experiments are not without their problems. Here it is con-
ceivable that the students would be more comfortable with the testing pro-
cedure after going through the process once. This may serve to naturally
decrease (or possibly increase) reactions times. It is better to randomly assign

which treatment (alcohol/sober) comes frst. But here, in a practical setting,
it is much easier to have the sober times be the “before” times.
Chapter 11
Inference for Proportions
J.B.’s strongly suggested exercises: 2, 5, 6, 8, 10, 12, 16, 17, 19, 20, 24, 25, 26, 27
11.1 Introduction
11.2 The Sampling Distribution of p̂
p(1−p)
1. E(p̂) = p (p̂ is an unbiased estimator of p), and V ar(p̂) = n .
2. The normal approximation is best when n is large and p = 0.5. The normal
approximation is worst when n is small and p is close to 0 or 1.
11.3 Confdence Intervals and Hypothesis Tests for the

Population Proportion p
3. Never. We make confdence intervals for parameters, and never for statistics.
4. (a) p̂ = 62
= 0.31.
200q q
p̂(1−p̂)
(b) SE(p̂) = n = 0.31(1−0.31)
200 = 0.0327.
(c)
p̂ ± zα/2 SE(p̂)
0.31 ± 1.96 × 0.0327
175
Balka ISE 1.0911.3. CONFIDENCE INTERVALS AND HYPOTHESIS TESTS FOR THE POPULATION PROPORTION P 176
which works out to 0.31 ± 0.064 or (0.245, 0.374).

q q
p0 (1−p0 )
(d) SE0 (p̂) = n = 0.5(1−0.5)
200 = 0.03536.
p̂−p0
(e) Z = SE0 (p̂) = 0.31−0.5
0.03536 = −5.374.
5. (a) np̂ = 10 × 0.70 = 7, n(1 − p̂) = 10 × 0.30 = 3. Since these quantities are
not both at least 15, we should not be using the normal approximation. (The
sampling distribution of p̂ cannot be reasonably approximated by a normal
distribution in this situation.)
(b) np̂ = 200 × 0.70 = 140, n(1 − p̂) = 200 × 0.30 = 60. Since these quantities
are both at least 15, we can use the normal approximation. (The sampling
distribution of p̂ can be reasonably approximated by a normal distribution in
this situation.)
(c) np̂ = 200 × 0.99 = 198, n(1 − p̂) = 200 × 0.01 = 2. Since these quantities are
(d) np̂ = 10,000×0.01 = 100, n(1−p̂) = 10,000×0.99 = 9,900. Since these quanti-
ties are both at least 15, we can use the normal approximation. (The sampling
this situation.)
(e) Here np̂ = 0, and we should not even think about using the normal approxi-
mation in this situation.
6. (a) p represents the true proportion of male births to heavy-smoking parents in

Liverpool, UK (in the time frame of the study).
(b) p̂ = 158
363 = 0.435.
(c) Yes, the sample size is large enough. (There are 158 successes (males) and 205
failures (females), so the at least 15 guideline is satisfed.)
q q
(d) SE(p̂) = p̂(1−p̂)
n = 0.435(1−0.435)
363 = 0.0260.
(e)
p̂ ± 1.96SE(p̂)
0.435 ± 1.96 × 0.0260
0.435 ± 0.051

(f) We can be 95% confdent that the true proportion of male births for Liverpool
couples in which both parents are heavy smokers lies between 0.384 and 0.486.
(g) H0 : p = 0.5 (The true proportion of male births is 0.50.)
Ha : p = q(The true proportion
6 0.5 q of male births is not 0.50.)
p0 (1−p0 ) 0.5(1−0.5)
SE0 (p̂) = n = 363 = 0.0262.
Balka ISE 1.09 11.4. DETERMINING THE MINIMUM SAMPLE SIZE N 177
p̂−p0
Z = SE 0 (p̂)
= 0.435−0.5
0.0262 = −2.48.
Since the alternative hypothesis is two-sided, the p-value is double the area
to the left of −2.48 under the standard normal curve. The area to the left of
−2.48 is 0.0066 (found using software or the standard normal table), and so
the p-value is about 0.013.
The p-value is less than the given signifcance level (0.013 < 0.05), so the
evidence against the null hypothesis is signifcant at α = 0.05. There is statis-
tically signifcant evidence that the true proportion of male births (for babies
born to heavy-smoking Liverpudlian parents) di˙ers from 0.5. Note that the
sample proportion of males was 0.435, with a corresponding 95% confdence
interval of (0.384, 0.486), indicating that there is strong evidence that the true
proportion of male births is in fact less than 0.5.
(h) The sample was of births to Liverpool couples in which both parents were
heavy smokers. As such, our conclusions apply only to the population of Liv-
erpool births in which both parents are heavy smokers. (Technically speaking
our conclusions apply only to the population from which we sampled, but we
may use a study like this to give us a hint of what might be the reality in other
scenarios (all UK births to heavy-smoking parents, for example).)
11.4 Determining the Minimum Sample Size n
7. (a) n = ( 1.96
0.03 ) · 0.5(1 − 0.5) = 1067.1, and we round up to 1068. 1068 is the
2
minimums sample size required.

(b) n = ( 1.96
0.03 ) · 0.20(1 − 0.20) = 682.95, and we round up to 683. 683 is the
2
minimums sample size required. Note that this is just an approximation. If

we are wrong about p being close to 0.2, and it is closer to 0.5, then we will
be underestimating the sample size required.
(c) n = ( 1.645
0.01 ) · 0.5(1 − 0.5) = 6765.1, and we round up to 6,766. 6,766 is the
2

(d) n = ( 2.576
0.01 ) · 0.5(1 − 0.5) = 16589.4, and we round up to 16,590. 16,590 is the
2

(e) The variance of p̂ is greater if p = 0.2 than if p = 0.1, and thus we would need
a larger sample size if p = 0.2 than if p = 0.1. So, n = ( 2.576 2
0.01 ) · 0.2(1 − 0.2) =
10617.2, and we round up to 10,618. 10,618 is the minimums sample size
required. (Had we used p = 0.1 in the formula instead, we would have arrived
at a minimum sample size of 5,973, but this would not have been large enough
if p is actually closer to 0.2).
11.5.
Balka ISE 1.09 INFERENCE PROCEDURES FOR THE DIFFERENCE BETWEEN TWO POPULATION PROPORTIONS 178
11.5 Inference Procedures for the Di˙erence Between

Two Population Proportions
11.5.1 The Sampling Distribution of p̂1 − p̂2
11.5.2 Confdence Intervals and Hypothesis Tests for p1 − p2
8. SE(p̂1 − p̂2 ) is the standard error of the di˙erence in sample proportions, which is
an estimate of the standard deviation of the sampling distribution of p̂1 − p̂2 . It is
a measure of the variability in p̂1 − p̂2 if samples were to be repeatedly drawn from
the populations.
9. (a) p̂1 = 82
= 0.205, p̂2 = 104
400 400 = 0.260.
q
(b) SE(p̂1 − p̂2 ) = 0.205(1−0.205)
400 + 0.260(1−0.260)
400 = 0.02981.
(c) 0.205 − 0.260 ± 1.96 × 0.02981, which works out to −0.055 ± 0.0584 or (−0.113,
0.003).
(d) H0 : p1 = p2 , Ha : p1 =
6 p2 . The pooled proportion p̂ = 400+400
82+104
= 0.2325.
q
1 1
SE0 (p̂1 − p̂2 ) = 0.2325(1 − 0.2325)( 400 + 400 ) = 0.02987.
Z = 0.205−0.260
0.02987 = −1.841. The resulting p-value is 0.066, which does not give
signifcant evidence against the null hypothesis at α = 0.05.
10. (a) p̂M = 21

39 = 0.5385, p̂C = 22
38 = 0.5789.
s
p̂M (1 − p̂M ) p̂C (1 − p̂C )
SE(p̂M − p̂C ) = +
nM nC
r
0.5385(1 − 0.5385) 0.5789(1 − 0.5789)
= +
39 38
= 0.1131
The 95% confdence interval for pM − pC is:
p̂M − p̂C ± 1.96 × SE(p̂M − p̂C )
0.5385 − 0.5789 ± 1.96 × 0.1131
−0.0404 ± 0.2216
which works out to(−0.26, 0.18).
(b) We can be 95% confdent that the di˙erence in the true proportion of pigeons
that will navigate to the home loft under the conditions of the experiment
(pM − pC ) lies between −0.26 and 0.18. (Note that 0 is contained within this
interval and thus it is plausible that pM = pC . The data is consistent with the
magnetic pulse having no e˙ect.)
(c) H0 : pM = pC (The true proportion of pigeons that will navigate to the home
loft under the conditions of this experiment is the same for the magnetic pulse
group and the control group. In short, the magnetic pulse has no e˙ect on the
likelihood of a pigeon arriving home.)
Ha : pM 6= pC (The true proportion of pigeons that will navigate to the home
loft under the conditions of this experiment di˙ers between the magnetic pulse
group and the control group. In short, the magnetic pulse has an e˙ect on the
likelihood of a pigeon arriving home.)
The pooled proportion is p̂ = 21+22

39+38 = 43
77 = 0.5584.
r
1 1
SE0 (p̂M − p̂C ) = p̂(1 − p̂)( + )
nM nC
r
1 1
= 0.5584(1 − 0.5584)( + )
39 38
= 0.1132
p̂M − p̂C
Z=
SE0 (p̂M − p̂C )
0.5385 − 0.5789
=
0.1132
= −0.357
to the left of −0.357 under the standard normal curve. The area to the left
of −0.357 under the standard normal curve is approximately 0.360, resulting
in a p-value of approximately 0.72. This is a large p-value, and it yields
absolutely no evidence against the null hypothesis. There is no evidence that
the proportion of pigeons that successfully navigate home di˙ers between the
magnetic pulse group and the control group. (There is no evidence that the
magnetic pulse has any e˙ect on the likelihood of a pigeon navigating back to
the loft.)
11. (a) 0.919 (the area to the right of −1.40 under the standard normal curve).
(b) 0.081 (the area to the left of −1.40 under the standard normal curve).
(c) 0.162 (twice the area to the left of −1.40 under the standard normal curve).
(d) 0.030 (the area to the left of −1.88 under the standard normal curve).
(e) 0.060 (twice the area to the left of −1.88 under the standard normal curve).
11.6.2 Concepts
12. p̂ is the sample proportion, whereas p represents the true proportion for the entire
population. Once the sample is drawn, we will know the value of p̂, but p is typically
an unknown value that we are trying to estimate.
13. No, this would never make sense. We test hypotheses about parameters, and not
about statistics.
14. SE(p̂) is the standard error of p̂, the estimate of the standard deviation of the
sampling distribution of p̂. It is a measure of the variability in p̂, if we were to
repeatedly draw samples from the population. Since we regard p as a fxed, unknown
quantity that we are trying to estimate, it is simply a constant and it does not have
a standard error.
15. (a) np0 = 10 × 0.20 = 2, n(1 − p0 ) = 10 × 0.80 = 8. Since these quantities are
(b) np0 = 200 × 0.20 = 40, n(1 − p0 ) = 200 × 0.80 = 160. Since these quantities
are both at least 15, we can use the normal approximation. (The sampling
this situation.)
(c) np0 = 10 × 0.998 = 9.98, n(1 − p0 ) = 10 × 0.002 = 0.02. Since these quantities
are not both at least 15, we should not be using the normal approximation.
(The sampling distribution of p̂ cannot be reasonably approximated by a nor-
mal distribution in this situation.)
(d) np0 = 1000×0.998 = 998, n(1−p0 ) = 1000×0.002 = 2. Since these quantities
are not both at least 15, we should not be using the normal approximation.
(The sampling distribution of p̂ cannot be reasonably approximated by a nor-
mal distribution in this situation.)
(e) np0 = 100,000 × 0.998 = 99,800, n(1 − p0 ) = 100,000 × 0.002 = 200. Since
these quantities are both at least 15, we can use the normal approximation.
(The sampling distribution of p̂ can be reasonably approximated by a normal
16. There are many possible sources of bias, including:
• These types of poll are often conducted by telephone, and will therefore not
include anyone who does not have a phone. In the early days of cell phones,
these types of surveys called only landline phones. People who were cell-phone-
only would be left out (and these people tended to be a younger subgroup
of the population, whose opinions may di˙er in an important way from the
population as a whole).
• If this was an online survey, there are obvious biases. People who never use a
computer (who tend to be older and poorer than those who do), would not be
included.
• The survey does not include people who hang up (refuse to answer questions).
If the opinions of these people tend to di˙er from those who do answer (which
is quite likely), then there will be bias in the survey.
• On phone surveys they typically speak to adults only, and thus the interpre-
tation of the survey must relate to adult Canadians, and not Canadians as a
whole.
• The people spoken to may have a tendency to lie. As a possible example, a

person may feel that schools are as safe as they used to be, but think that
if a high proportion say otherwise, perhaps the government will spend more
on schools. This type of bias is more obvious with potentially embarrassing
questions such as “do you have an STD?” But overall, what people say they
think isn’t necessarily what they truly think.
17. (a) p̂ is an unbiased estimator of p. True, since E(p̂) = p.

(b) When n < 30 we should use the t distribution when calculating confdence
intervals for p. False. The t distribution never arises in inference for propor-
tions.
(c) The true distribution of p̂ is based on the binomial distribution. True. p̂ = Xn,
where X has a binomial distribution with parameters n and p.
(d) The true standard deviation
q of the sampling distribution of p̂ depends on the
value of p. True. σp̂ = p(1−p)
n .
(e) The sampling distribution of p̂ is perfectly normal for large sample sizes. False.
It’s an approximation, but it can be a very good approximation.
18. (a) The normal approximation to the sampling distribution of p̂ works best when
we have a large sample size and p = 0.5. True.
(b) The sampling distribution of p̂ becomes more normal as p tends to 1. False.
The sampling distribution becomes strongly skewed as p → 1.
(c) The sampling distribution of p is approximately normal for large sample sizes.
False. p is a parameter, not a statistic, and as such does not have a sampling
distribution.
(d) All else being equal, the value of SE(p̂) decreases as the sample size increases.
True.
(e) In repeated sampling, exactly 95% of 95% confdence intervals for p will cap-
ture p. False, since we are using a normal approximation and not an exact
procedure. (But if the sample size is very large then the true percentage will be
very close to 95%.)
11.6.3 Applications
19. (a) p represents the true proportion of homing pigeons that would navigate back to
the home loft (for this type of pigeon, under the conditions of this experiment).
(b) p̂ = 22
38 = 0.5789.
(c) Yes, the sample size is large enough. There are 22 successes (pigeons that
made it back) and 16 failures (pigeons that did not make it back), so the at
least 15 guideline is satisfed (barely).
q q
(d) SE(p̂) = p̂(1−p̂)
n = 0.5789(1−0.5789)
38 = 0.0801.
(e)
p̂ ± 1.96SE(p̂)
0.5789 ± 1.96 × 0.0801
0.5789 ± 0.1570

(f) We can be 95% confdent that the true proportion of pigeons that would suc-
cessfully navigate home lies between 0.42 and 0.74 (for this type of pigeon,
under the conditions of this experiment).
(g) H0 : p = 0.25 (The true proportion of pigeons that would navigate home is
0.25.)
Ha : p > 0.25 (The true proportion of pigeons that would navigate home is
greater than 0.25. If we are trying to show strong evidence against the scien-
tist’s claim that no more than 25% would navigate home on average, we care
only aboutqa di˙erence qin this direction.)
p0 (1−p0 ) 0.25(1−0.25)
SE0 (p̂) = n = 38 = 0.0702.
p̂−p0 0.5789−0.25
Z= SE0 (p̂) = = 4.68.
0.0702
Since the alternative hypothesis is p > 0.25, the p-value is the area to the
right of 4.68 under the standard normal curve. Using software, we can fnd:
p-value = 1.4 × 10−6 . (Using the standard normal table we could say only that
the p-value is very close to 0.) This p-value is very small, yielding very strong
There is very strong evidence that, for this type of pigeon under the conditions
of this experiment, the true proportion that will successfully navigate to the
home loft is greater than 0.25. (This should not come as a great surprise, since
we found a 95% confdence interval for p of 0.42 to 0.74.)
20. (a) p represents the true proportion of seatbelt-wearing victims of fatal car acci-
dents that show seat belt marks. (Note that the study involved only accidents
in Sydney, Australia, in which there was no airbag deployment. Any conclu-
sions relate to that population of accidents.)
(b) p̂ = 27
74 = 0.3649.
(c) Yes, the sample size is large enough. (There are 27 successes (individuals with
seat belt marks) and 47 failures (individuals without seat belt marks), so the
at least 15 guideline is satisfed.)
q q
(d) SE(p̂) = p̂(1−p̂)
n = 0.3649(1−0.3649)
74 = 0.0560.
(e)
p̂ ± 1.96SE(p̂)
0.3649 ± 1.96 × 0.0560
0.3649 ± 0.1097
(f) We can be 95% confdent that p lies between 0.255 and 0.475, where p is the
true proportion of seatbelt-wearing victims that show seat belt marks.
(g) No, since the information in the question does not give any indication of a
hypothesized value to be tested. It is conceivable that a researcher may have
a hypothesis they wish to test, but based on the given information this is an
estimation problem, not a testing problem.
21. (a) p represents the true proportion of murders in which the victim was the of-
fender’s domestic partner (for Finnish murders in which the murderer was
male).
(b) p̂ = 24
91 = 0.2637.
(c) Yes, the sample size is large enough. (There are 24 successes (partners) and
67 failures (non-partners), so the at least 15 guideline is satisfed.)
q q
(d) SE(p̂) = p̂(1−p̂)
n = 0.2637(1−0.2637)
91 = 0.0462.
(e)
p̂ ± 1.96SE(p̂)
0.2637 ± 1.96 × 0.0462
0.2637 ± 0.0905
(f) We can be 95% confdent that the true proportion of murders that were of the
domestic partner of the perpetrator lies between 0.173 and 0.354. (For Finnish
murders in which the murderer was male.)
(g) H0 : p = 0.4 (The true proportion of victims that were the domestic partner of
the perpetrator is 0.40.)
Ha : p =
6 0.4 (The true proportion of victims that were the domestic partner of
the perpetrator
q is not 0.40.)
q
p0 (1−p0 ) 0.4(1−0.4)
SE0 (p̂) = n = 91 = 0.0514.
p̂−p0 0.2637−0.4
Z= SE0 (p̂) = = −2.65.
0.0514
to the left of −2.65 under the standard normal curve. The area to the left of
−2.65 is 0.0040 (found using software or a standard normal table), and so the
p-value is about 0.008.
The p-value is less than the given signifcance level (0.008 < 0.05), so the
evidence against the null hypothesis is signifcant at α = 0.05. There is statis-
tically signifcant evidence that the true proportion of victims that were the
domestic partner of the perpetrator di˙ers from 0.4. The sample proportion is
0.26, with a 95% confdence interval for p of (0.173, 0.354), indicating strong
evidence that the true proportion of victims that are domestic partners is in
fact less than 0.4.
(h) The sample was of Finnish murder victims in which the perpetrator was male.
As such, our conclusions apply only to the population of Finnish murder vic-
tims in which the perpetrator was male. (Technically speaking our conclusions
apply only to the population from which we sampled, but we may use a study
like this to give us a hint of what might be the reality in other scenarios
(murders committed by European males, for example).)
22. (a) p̂F = 32

91 = 0.3516, p̂M = 24
91 = 0.2637.
s
p̂F (1 − p̂F ) p̂M (1 − p̂M )
SE(p̂F − p̂M ) = +
nF nM
r
0.3516(1 − 0.3516) 0.2637(1 − 0.2637)
= +
91 91
= 0.0681
The 90% confdence interval for pF − pM is:
p̂F − p̂M ± 1.645 × SE(p̂F − p̂m )
0.3516 − 0.2637 ± 1.645 × 0.0681
0.0879 ± 0.112
which works out to (−0.024, 0.200).
(b) We can be 90% confdent that the di˙erence in the true proportions of victims
that were domestic partners (pF − pM ) lies between −0.024 and 0.200.
(c) H0 : pF = pM (The true proportion of murder victims that were domestic part-
ners is the same for female and male murderers.)
Ha : pF 6= pM (The true proportion of murder victims that were domestic part-

ners di˙ers between female and male murderers.)

91+91 = 56
182 = 0.3077.
r
1 1
SE0 (p̂F − p̂M ) = p̂(1 − p̂)( + )
nF nM
r
1 1
= 0.3077(1 − 0.3077)( + )
91 91
= 0.0684
p̂F − p̂M
Z=
SE0 (p̂F − p̂M )
0.3516 − 0.2637
=
0.0684
= 1.285
to the right of 1.285 under the standard normal curve. The area to the right
of 1.285 under the standard normal curve is about 0.099, resulting in a p-
value of approximately 0.20. The p-value is greater than the given signifcance
level (0.20 > 0.05), and so the evidence against the null hypothesis is not
signifcant at the 5% signifcance level. There is not signifcant evidence that
the proportion of murder victims that were domestic partners di˙ers between
male and female murderers. (The data is consistent with the two proportions
being equal.)
23. (a) p̂U = 43

60 = 0.7167, p̂C = 27
= 0.6136.
44
s
p̂U (1 − p̂U ) p̂C (1 − p̂C )
SE(p̂U − p̂C ) = +
nU nC
r
0.7167(1 − 0.7167) 0.6136(1 − 0.6136)
= +
60 44
= 0.0937
The 95% confdence interval for pU − pC is:
p̂U − p̂C ± 1.96 × SE(p̂U − p̂C )

0.7167 − 0.6136 ± 1.96 × 0.0937
0.1031 ± 0.1835

(b) We can be 95% confdent that the di˙erence in the true proportions of individ-
uals that express the Leb antigen (pU −pC ) lies between −0.08 and 0.29. (Note
that 0 is contained within this interval and thus it is plausible that pU = pC .
The data is consistent with there being no association between peptic ulcers
and the Leb antigen.)
(c) H0 : pU = pC (The true proportion of individuals expressing the Leb antigen
is the same for peptic ulcer su˙erers and healthy individuals.)
Ha : pU 6= pC (The true proportion of individuals expressing the Leb antigen
di˙ers between peptic ulcer su˙erers and healthy individuals.)

60+44 = 70
104 = 0.6731.
r
1 1
SE0 (p̂U − p̂C ) = p̂(1 − p̂)( + )
nU nC
r
1 1
= 0.6731(1 − 0.6731)( + )
60 44
= 0.0931
p̂U − p̂C
Z=
SE0 (p̂U − p̂C )
0.7167 − 0.6136
=
0.0931
= 1.107
Since the alternative hypothesis is two-sided, the p-value is double the area to
the right of 1.107 under the standard normal curve. The area to the right of
1.107 under the standard normal curve is approximately 0.134, resulting in a p-
value of approximately 0.27. The p-value is greater than the given signifcance
level (0.27 > 0.05), and so the evidence against the null hypothesis is not
signifcant at the 5% signifcance level. There is not signifcant evidence that
the proportion of individuals expressing the Leb antigen di˙ers between peptic
ulcer su˙erers and healthy individuals. (The data is consistent with the two
proportions being equal.)
24. (a) p̂H = 158

363 = 0.4353, p̂N = 2685
5045 = 0.5322.
s
p̂H (1 − p̂H ) p̂N (1 − p̂N )
SE(p̂H − p̂N ) = +
nH nN
r
0.4353(1 − 0.4353) 0.5322(1 − 0.5322)
= +
363 5045
= 0.0270
The 95% confdence interval for pH − pN is:
p̂H − p̂N ± 1.96 × SE(p̂H − p̂N )

0.4353 − 0.5322 ± 1.96 × 0.0270
−0.0969 ± 0.0528

(b) We can be 95% confdent that the di˙erence between the true proportion of
male births to heavy-smoking parents and the true proportion of male births
to non-smoking parents (pH − pN ) lies between −0.15 and −0.04. (Note that
the entire interval lies to the left of 0, thus giving some evidence that the
true proportion of male births for heavy-smoking parents is less than the true
proportion for non-smoking parents.)
(c) H0 : pH = pN (The true proportion of male births for heavy-smoking parents
is equal to the true proportion of male births for non-smoking parents.)
Ha : pH 6= pN (The true proportion of male births for heavy-smoking parents
di˙ers from the true proportion of male births for non-smoking parents.)

363+5045 = 2843
5408 = 0.5257.
r
1 1
SE0 (p̂H − p̂N ) = p̂(1 − p̂)( + )
nH nN
r
1 1
= 0.5257(1 − 0.5257)( + )
363 5045
= 0.0271
p̂H − p̂N
Z=
SE0 (p̂H − p̂N )
0.4353 − 0.5322
=
0.0271
= −3.57
to the left of −3.57 under the standard normal curve. The area to the left
of −3.57 under the standard normal curve is approximately 0.0002 (found
using software, or approximated with the standard normal table), resulting in
a p-value of approximately 0.0004.
Since the p-value is less than the given signifcance level (0.0004 < 0.05), there
is statistically signifcant evidence against the null hypothesis. There is strong
evidence that the true proportion of male births to heavy-smoking parents
di˙ers from the true proportion of male births to non-smoking parents. (By
looking at the sample proportions of p̂H = 0.4353 and p̂N = 0.5322, and the
resulting 95% confdence interval for pH − pN of (-0.15, -0.04), we can state

that there is strong evidence that the population proportion of male births to
heavy-smoking parents is in fact less than the population proportion of male
births to non-smoking parents.)
(d) The samples were drawn from births in Liverpool, UK, and as such our conclu-
sions apply only to that area in the time frame of the study. (But the results
of this study do give us a hint that the observed relationship may hold in other
areas as well.) The confdence interval and the hypothesis test give evidence
of an association between parental smoking and the sex of the child, but they
do not imply a causal link. (For the usual reasons – this was an observational
study, not an experiment, and so there are a multitude of lurking variables
that could be the cause of the observed relationship.)
25. (a) p̂F = 46

200 = 0.230, p̂C = 27
200 = 0.135.
s
p̂F (1 − p̂F ) p̂C (1 − p̂C )
SE(p̂F − p̂C ) = +
nF nC
r
0.230(1 − 0.230) 0.135(1 − 0.135)
= +
200 200
= 0.0383
The 95% confdence interval for pF − pC is:
p̂F − p̂C ± 1.96 × SE(p̂F − p̂C )

0.230 − 0.135 ± 1.96 × 0.0383
0.095 ± 0.0751

(b) We can be 95% confdent that the di˙erence between the true success rate
in pleasant aroma areas and the true success rate in areas with no aroma
(pF − pC ) lies between 0.02 and 0.17. (Note that the entire interval lies to the
right of 0, thus giving evidence that the true success rate in pleasant aroma
areas is greater than the true success rate in areas with no aroma.) There is
evidence that the pleasant aroma areas have a positive e˙ect on the success
rate. We will now investigate that a little more formally with a hypothesis
test.
(c) H0 : pF = pC (The true success rate in pleasant aroma areas is equal to the
true success rate in areas of no aroma.)
Ha : pF 6= pC (The true success rate in pleasant aroma areas di˙ers from the
true success rate in areas of no aroma.)

200+200 = 73
400 = 0.1825.
r
1 1
SE0 (p̂F − p̂C ) = p̂(1 − p̂)( + )
nF nC
r
1 1
= 0.1825(1 − 0.1825)( + )
200 200
= 0.0386
p̂F − p̂C
Z=
SE0 (p̂F − p̂C )
0.230 − 0.135
=
0.0386
= 2.46
to the right of 2.46 under the standard normal curve. The area to the right
of 2.46 under the standard normal curve is approximately 0.007 (found using
software or the standard normal table), resulting in a p-value of approximately
0.014.
Since the p-value is less than the given signifcance level (0.014 < 0.05), there
is statistically signifcant evidence against the null hypothesis. There is strong
evidence that the true success rate (of obtaining a phone number) in pleasant
aroma areas di˙ers from the true success rate in areas with no aroma. (By
looking at the sample proportions of p̂F = 0.230 and p̂C = 0.135, and the
resulting 95% confdence interval for pF − pC of (0.02, 0.17), we can state that
there is strong evidence that the true success rate in pleasant aroma areas is
greater than the true success rate in areas with no aroma.)
Note that the conclusions to this study apply only to the specifc conditions of
the experiment (good-looking young males approaching young women walking
alone in a shopping mall). If the conditions of the experiment were to change
(using males who were not quite as good-looking, for example), then the results
might possibly change dramatically.
26. (a) p̂p = 200430

160 86
= 0.0007982837, p̂v = 199747 = 0.0004305446. The point estimate
of the di˙erence in population proportions (pp − pv ) is the di˙erence in sample
proportions: p̂p − p̂v = 0.0007982837 − 0.0004305446 = 0.0003677391.
(b) The pooled proportion is p̂ = 200430+199747
160+86
= 0.000614728.
p
SE0 (p̂p −p̂v ) = 0.000614728(1 − 0.000614728)(1/200430 + 1/199747) = 0.00007836327
0.0007982837 − 0.0004305446
Z= = 4.693
0.00007836327
(c) The p-value is twice the area to the right of 4.693 under the standard normal
curve. This area is small (close to 0). By the computer, p-value = 2.7 × 10−6 .
(d) Since the p-value is very small (2.7 × 10−6 ), there is very strong evidence
of a di˙erence in population proportions. Since the sample proportion in the
placebo group is larger than that of the vaccine group, there is strong evidence
that those in the vaccine group developed polio less often. And since this was
a randomized experiment, we can say the strong conclusion: This study yields
strong evidence that use of this vaccine in children reduces the risk of developing
polio.
0.0007982837
(e) RR = = 1.854. People without the vaccine were 1.85 times
0.0004305446
more likely to develop polio.
q
0.68(1−0.68)
27. (a) p̂ = 100
68
= 0.68. SE(p̂) = 100 = 0.046648. The 95% confdence
interval is given by 0.68 ± 1.96 × 0.046647, which works out to 0.68 ± 0.0914
or (0.589, 0.771). We can be 95% confdent that the proportion of this type
of minnow that would die within 48 hours after exposure to 5 ppm of zinc lies
between 0.589 and 0.771.
(b) Yes. The entire interval lies to the right of 0.50, indicating that 5 ppm of zinc
kills more than 50% of these minnows under these conditions. It appears as
though the LD50 is less than 5 ppm.
(c) H0 : p = 0.50 (5 ppm of zinc will kill 50% of minnows within 48 hours)
Ha : p =6 0.50 (The true percentage of minnows killed within 48 hours with 5
ppm of zinc qdi˙ers from 0.50.)
SE0 (p̂) = 0.50(1−0.50)
100 = 0.05. Z = 0.68−0.50
0.05 = 3.6. The p-value is double the
area to the right of 3.6 under the standard normal curve (p-value = 0.0003).
Since the p-value is less than the given signifcance level of α = 0.05, there is
signifcant evidence against the null hypothesis. There is signifcant evidence
that, under these conditions, the true proportion of minnows killed by 5 ppm
of zinc will di˙er from 0.50. Since the sample proportion is greater than 0.50
(leading to a test statistic in the right tail of the distribution), there is strong
evidence that the true proportion killed is in fact greater than 0.50. (Which
shouldn’t come as a big surprise, since our confdence interval was found to be
(0.589, 0.771) in 27a.)
(d) n = ( 2.576
0.03 ) · 0.5(1 − 0.5) = 1843.3. We would need a sample size of at least
2
1844 minnows.
(e) No. The rough guideline tells us that we need at least 15 successes and 15
failures for the normal approximation to be reasonable. Here, depending on
how we defne success and failure, we have 198 of one and only 2 of the other.
The sampling distribution of p̂ cannot reasonably be approximated by a normal
distribution in this spot.
q
0.64179(1−0.64179)
28. (a) p̂ = 43
67 = 0.64179. The 95% confdence interval is 0.64179±1.96 67 ,
which works out to .64179 ± 0.1148, or (0.527, 0.757). We can be 95% conf-
dence that the proportion of babies born at 25 weeks that live for at least a
year lies between 0.527 and 0.757.
(b) H0 : p = 0.5 (Half of 25 week premature babies die before one year.)
Ha : p 6= 0.5 (The true proportion premature babies that die before one year
di˙ers fromq0.5.)
0.5(1−0.5)
SE0 (p̂) = 67 = 0.06108. Z = 0.64179−0.5
0.06108 = 2.321. The p-value is
twice the area to the right of 2.321 under the standard normal curve, which is
approximately 0.020. There is very strong evidence (p-value = 0.02) that the
true proportion of this type of premature baby that survives to one year di˙ers
from 0.5. In fact, since the sample proportion is greater than 0.50 (leading to
a Z statistic in the right tail of the distribution), there is strong evidence that
the true proportion that survives is in fact greater than 0.50.
q
0.425(1−0.425)
29. p̂ = 17
40 = 0.425. SE(p̂) = 40 = 0.07816. The 95% confdence interval is
given by 0.425 ± 1.96 × 0.07816, which works out to 0.425 ± 0.1532 or (0.272, 0.578).
We can be 95% confdent that proportion of all soft-serve ice cream vendors in this
city that have E. coli counts that exceed recommended guidelines lies between 0.272
and 0.578.
q
0.05(1−0.05)
30. (a) p̂ = 50025
= 0.05. SE(p̂) = 500 = 0.009746. The 90% confdence
interval is given by 0.05 ± 1.645 × 0.009746, which works out to 0.05 ± 0.016 or
(0.034, 0.066). We can be 95% confdent that the proportion of adults (in this
age group, area, and time) that have a university degree lies between 0.034
and 0.066.
(b) H0 : p = 0.23 (The proportion with a university degree was the same as for all
of Canada.)
Ha : p =
6 0.23 (The proportion with a university degree di˙ered from the rest
of Canada.)q
SE0 (p̂) = 0.23(1−0.77)
500 0.01882 = −9.564. The p-value is
= 0.01882. Z = 0.05−0.23
double the area to the left of −9.564 under a standard normal curve, which
is very near 0 (using a computer, we can fnd it to be 1.1 × 10−21 ). There is
extremely strong evidence that the proportion of people that have a university
degree in this age group and area di˙ered from that of Canada as a whole. In
fact, since the sample proportion is less than the proportion for all of Canada
(leading to a test statistic in the left tail of the distribution), there is a strong
indication that the true proportion in this area was in fact less than that of
Canadians as a whole.
31. n = ( 1.645
0.04 ) · 0.5(1 − 0.5) = 422.8. We would need a sample size of at least 423
2
individuals in order to estimate p to within 0.04 with 90% confdence.

q
32. (a) p̂F = 10058 43
= 0.58, p̂M = 100 = 0.43. SE(p̂F −p̂M ) = 0.58(1−0.58)
100 + 0.43(1−0.43)
100 =
0.0699.
A 90% confdence interval for pF − pM is given by 0.58 − 0.43 ± 1.645 × 0.0699,
which works out to 0.15 ± 0.115 or (0.035, 0.265).
(b) We can be 90% confdent that the true di˙erence between the approval ratings
of women and men (pF − pM ) lies between 0.035 and 0.265.
(c) H0 : pF = pM (The approval ratings for men and women are equal.)
Ha : pF 6= pM (The approval ratings for men and women are di˙erent.)
58+43
= 0.505.
SE0 (p̂F −p̂M ) = 0.505(1 − 0.505)(1/100 + 1/100) = 0.070707. Z = 0.58−0.43
p
0.070707 =
2.121.
The p-value is twice the area to the right of 2.121 under the standard normal
curve, which works out to 0.034. There is signifcant evidence against the null
hypothesis at the given level of α = 0.05. There is signifcant evidence of a
true di˙erence in approval ratings between men and women. It appears as
though women are more likely to approve of the way the President is handling
his job.
q
0.88(1−0.88)
33. (a) p̂1 = 176
200 = 0.88, p̂ 2 = 184
200 = 0.92. SE(p̂1 − p̂ 2 ) = 200 + 0.92(1−0.92)
200 =
0.02993.
A 95% confdence interval for p1 − p2 is given by 0.88 − 0.92 ± 1.96 × 0.02993,
which works out to −0.04 ± 0.0587 or (−0.099, 0.019).
We can be 95% confdent that the di˙erence in population proportion of bed-
bugs killed for heat treatments of 45 degrees and 47 degrees Celsius lies between
−0.099 and 0.019.
(b) H0 : p1 = p2 (The proportion of bedbugs killed is the same for both tempera-
tures),
Ha : p1 < p2 (The proportion of bedbugs killed is lower at the lower tempera-
ture).
p̂ = 176+184
200+200 = 0.90.
SE0 (p̂1 − p̂2 ) = 0.90(1 − 0.90)(1/200 + 1/200) = 0.030. Z = 0.88−0.92
p
0.030 =
−1.333.
p-value = 0.091 (the area to the left of −1.333 under the standard normal
curve).
There is not signifcant evidence (at the given signifcance level of α = 0.01)
of a di˙erence in proportion of bedbugs killed between the two temperatures.
Chapter 12
Inference for Variances
(No exercises yet.)
193
Balka ISE 1.09 194
Chapter 13
χ2 Tests for Count Data
J.B.’s strongly suggested exercises: 2, 3, 4, 6, 7, 8, 10, 11, 14, 15, 16, 17, 19, 21, 22, 23
13.1 Introduction
13.2 χ2 Tests for One-Way Tables
13.2.1 The χ2 Test Statistic
1. (a) H0 : p1 = p2 = p3 = p4 = 0.25 (The four groups are all equally likely).

Ha : pi =
6 0.25 for at least two of the groups (the four groups are not all equally
likely).
(b)
Observed 52 149 99 100
Expected .25 × 400 = 100 .25 × 400 = 100 .25 × 400 = 100 .25 × 400 = 100
(52 − 100)2 (149 − 100)2 (99 − 100)2 (100 − 100)2

(c) χ2 = + + + = 47.06.
100 100 100 100
(d) 4 − 1 = 3.
(e) The p-value is the area to the right of 47.06 under a χ2 distribution with 3
degrees of freedom. The correct area is 3.374954 × 10−10 . From a table, the
best we could say is that the area is very small (p-value < 0.0005, or something
along those lines).
(f) Yes and yes. With such a small p-value, the null hypothesis would be rejected
at any reasonable level.
195
Balka ISE 1.09 13.2. χ2 TESTS FOR ONE-WAY TABLES 196
2. (a) Yes, the observed percentage of Q1 birthdays is higher for Canadian NHL
players than for Canadians as a whole (26.5% vs 24.2%). Yes, the observed
percentage of Q4 birthdays is lower for Canadian NHL players than for Cana-
dians as a whole (20.6% vs 23.8%).
(b) H0 : p1 = 0.242, p2 = 0.260, p3 = 0.260, p4 = 0.238 (The distribution of birth
dates for Canadian NHL players is the same as for Canadians as a whole.)
Ha : The given probabilities are not all correct. (The distribution of birth dates
for Canadian NHL players is not the same as for Canadians as a whole.)
(c) 510 × 0.242 = 123.42. This is a fair bit lower than the observed count of 135.
(d) 4 − 1 = 3.
(e)
Birth date Q1 Q2 Q3 Q4 Total
Observed count 135 146 124 105 510
Expected count .242 · 510 = 123.42 .260 · 510 = 132.60 .260 · 510 = 132.60 .238 · 510 = 121.38 510
The test statistic:
(135 − 123.42)2 (146 − 132.60)2 (124 − 132.60)2 (105 − 121.38)2
χ2 = + + + = 5.209
123.42 132.60 132.60 121.38
(f) The p-value is the area to the right of 5.209 under a χ2 distribution with 3
degrees of freedom, as illustrated in Figure 13.1.
χ2 Distribution with 3 DF
The p-value is this area
0 5.209
Figure 13.1: The p-value for the NHL birthday question.
Using software, we can fnd that the p-value is 0.157. Using a χ2 table, we can
fnd only that the p-value falls in an interval, such as: p-value > 0.10.
(g) Since the p-value is greater than the given signifcance level (0.157 > 0.05),
the evidence against H0 is not signifcant at α = 0.05. There is not signifcant
evidence that the distribution of birth dates for Canadian NHL players di˙ers
from that of Canadians as a whole. (It is plausible that the null hypothesis
is true.) Note that this does not mean that Malcom Gladwell’s assertion is
wrong, we simply do not have signifcant evidence of a birthday e˙ect in this
analysis.1 The observed percentage of Canadian NHL players born in Q1
(26.5) is greater than the percentage for Canadians (24.2). But we do not
have enough evidence to say this is a signifcant di˙erence.
1
A proper attempt to to investigate Canadian NHL player birthdays would involve data over many years,
and possibly other methods of analysis. The birthday e˙ect in professional hockey is real, but we simply
have not seen much evidence of it here.
Balka ISE 1.09 13.2. χ2 TESTS FOR ONE-WAY TABLES 197
13.2.2 Testing Goodness-of-Fit for Specifc Parametric Distribu-

tions
3. (a) H0 : The number of eggs that hatch in a nest follows a binomial distribution
with n = 3 and p = 0.80. Ha : The number of eggs that hatch in a nest does
not follow a binomial distribution with n = 3 and p = 0.80.
(b) We fnd the expected probabilities using the binomial distribution with n = 3
and p = 0.80. The expected probabilities are 0.008, 0.096, 0.384, 0.512. We
multiply these by the total number of nests (500) to get the expected counts:
4, 48, 192, 256.
2 2 2 2
(c) χ2 = (80−4)
4 + (29−48)
48 + (164−192)
192 + (227−256)
256 = 1458.889.
(d) The degrees of freedom are 4 − 1 = 3.
(e) The p-value is the area to the right of 1458.889 under a χ2 distribution with
4 − 1 = 3 degrees of freedom. The p-value is close to 0.
(f) There is very strong evidence that the number of eggs that hatch per nest does
not follow a binomial distribution with p = 0.80.
4. H0 : The number of eggs that hatch in a nest follows a binomial distribution.

Ha : The number of eggs that hatch in a nest does not follow a binomial distribution.
Number of eggs that hatch 80×0+29×1+164×2+227×3
p̂ = Total number of eggs = 500×3 = 0.692.
Using this estimate of p in the binomial formula with n = 3, we fnd the “expected
probabilities” for 0, 1, 2, 3 are 0.02921811, 0.19693766, 0.44247034, 0.33137389.
Multiplying by the number of nests, we obtain the expected counts: 14.60906,
98.46883, 221.23517, 165.68694. We then use our usual χ2 statistic:
(80−14.60906)2 (29−98.46883)2 (164−221.23517)2 (227−165.68694)2
χ2 = 14.60906 + 98.46883 + 221.23517 + 165.68694 = 379.1994.
The p-value is the area to the right of 379.1994 under a χ2 distribution with 4 −
1 − 1 = 2 degrees of freedom (recall that we lose one degree of freedom for every
parameter that must be estimated from the data – hence the loss of an extra degree
of freedom here). This area is near 0.
There is very, very, very strong evidence that the number of eggs that hatch per
nest does not follow a binomial distribution.
5. H0 : p1 = p2 = p3 = p4 (The four possibilities (0 eggs hatch, 1 egg hatches, 2 eggs

hatch, 3 eggs hatch) are all equally likely).
Ha : The four possibilities (0 eggs hatch, 1 egg hatches, 2 eggs hatch, 3 eggs hatch)
are not equally likely.
The expected counts are 125, 125, 125, 125.

Balka ISE 1.09 13.3. χ2 TESTS FOR TWO-WAY TABLES 198
(80−125)2 (29−125)2 (164−125)2 (227−125)2

χ2 = 125 + 125 + 125 + 125 = 185.328.
The p-value is the area to the right of 185.328 under a χ2 distribution with 4−1 = 3
degrees of freedom. This area is very close to 0. There is extremely strong evidence
that the four possible outcomes are not equally likely.
13.3 χ2 Tests for Two-Way Tables
6. (a) H0 : There is no association (relationship) between the row and column vari-
ables (this can be phrased di˙erently, depending on the sampling design—the
null hypothesis may be that the row and column variables are independent).
C1 C2
R1 200·650
1000 = 130 200·350
1000 = 70
(b)
R2 220·650
1000 = 143 220·350
1000 = 77
R3 580·650
1000 = 377 580·350
1000 = 203
(c) χ2 = 209.5835.
(d) DF = (3 − 1)(2 − 1) = 2.
degrees of freedom. This area is near 0 (p-value < 2.2e-16, as reported by R).
(f) Yes and yes. With such a tiny p-value, the null hypothesis would be rejected
at any reasonable level. There is extremely strong evidence against the null
hypothesis.
7. (a) 76.4% of the buyback guns were of small calibre. 20.3% of the homicide guns
were of small calibre. (Note that the percentage of small calibre guns is much
higher for buyback guns than for homicide guns.)
(b) H0 : The distribution of gun calibre is the same for buyback guns, homicide
guns, and suicide guns.
Ha : The distribution of gun calibre is not the same for buyback guns, homicide
guns, and suicide guns. (In other words, it is not true that the 3 distributions
are the same.)
(c) Row total×Column
Overall total
total
= 834×941
1435 = 546.89.
(d) (Number of rows − 1)(Number of columns − 1) = (4 − 1)(3 − 1) = 6.
(e) The p-value is the area to the right of 422.48 under a χ2 distribution with
6 degrees of freedom. The p-value is minuscule (very, very close to 0). (The
statistical software R reports p-value < 2.2 × 10−16 , which is R’s way of saying
that the p-value is very, very close to 0.)
(f) Here the p-value is minuscule, indicating extremely strong evidence against the
null hypothesis. (It would be nearly impossible to observe what was observed,
if the null hypothesis were true.) There is extremely strong evidence that
the distribution of gun calibre di˙ers between the 3 types of gun (buybacks,
homicides, suicides).
(g) It appears as though buyback guns were more likely than homicide and suicide
guns to be of smaller calibre, and homicide and suicide guns were more likely
than buyback guns to be of medium or large calibre. (We could investigate
this in greater detail using statistical inference techniques.)
(h) The samples were of guns from buyback programs in Milwaukee, and guns
used in homicides and suicides in Milwaukee, so those are the populations to
which the conclusions apply.
13.4.2 Concepts
8. E. The χ2 approximation breaks down when one or more of the expected counts
is small. Depending on how small they are, and how many are small, the ap-
proximation may be very poor. (Some people use the rough guideline that the
approximation is poor if some of the expected counts are less than 5. But this
guideline is quite conservative—if one or two of the expected counts slip a little
under 5, it is not a big problem.)
9. No. Here we have 50 children, but the sum of the observed is 119. The children are
appearing in more than one cell. (In other words, these 119 observations are not
independent.) The χ2 test is not appropriate here.
10. (a) 0.005.

(b) 0.15.
(c) P (Z12 + Z22 + Z32 + Z42 > 7.779) = 0.10. Here we are summing four squared in-
dependent standard normal random variables. This sum has a χ2 distribution
with 4 degrees of freedom (we are summing 4 squared terms). P (Z12 + Z22 +
Z32 + Z42 > 7.779) = 0.10 (found from the χ2 distribution with 4 degrees of
freedom, using either software or the χ2 table).
(d) P (Z12 + Z22 + Z32 < 11.345) = 0.99. Here we are summing three squared inde-
pendent standard normal random variables. This sum has a χ2 distribution
with 3 degrees of freedom (we are summing 3 squared terms). P (Z12 +Z22 +Z32 >
11.345) = 0.01, P (Z12 + Z22 + Z32 < 11.345) = 0.99 (found from the χ2 distri-
bution with 3 degrees of freedom, using either software or the χ2 table).
11. (a) True. The p-value is the area to the right of the test statistic.
(b) False. As the degrees of freedom increase, the χ2 distribution gets closer to a
normal distribution, that part is true. But the mean of a χ2 distribution is
equal to its degrees of freedom (not 0, as for the standard normal distribution).
(c) True. (But the skewness decreases as the degrees of freedom increase.)
(d) True.
(e) False. (Remember that the χ2 test statistic is the sum of squared terms.)
12. (a) False, it is only an approximation.

(b) True.
(c) True.
(d) True.
(e) True.
(f) True.
13.4.3 Applications
13. (a) H0 : True ratio of phenotypes is 9:3:3:1

Ha : True ratio of phenotypes is not 9:3:3:1
(b)
Phenotype
Tall/cut Tall/potato Dwarf/cut Dwarf/potato
Observed count 926 288 293 104
9 3 3 1
Expected under 9:3:3:1 16 × 1611 = 906.2 16 × 1611 = 302.1 16 × 1611 = 302.1 16 × 1611 = 100.7
Table 13.1: Observed and expected counts of phenotypes in the second generation.
(The given values are rounded, but you should carry many decimal places
throughout your calculations.)
2 2 2 2
χ2 = (926−906.2)
906.2 + (288−302.1)
302.1 + (293−302.1)
302.1 + (104−100.7)
100.7 = 1.469
(c) 0.69. The p-value is the area to the right of 1.469 under the χ2 distribution
with 4−1 = 3 degrees of freedom. Using software, we can fnd that the p-value
is 0.69. (Using a χ2 table, we could fnd only a range for the p-value, such as
p-value > 0.20.)
(d) Since the p-value is large (0.69), there is no evidence against the null hypoth-
esis. The null hypothesis would not be rejected at any reasonable signifcance
level. There is no evidence that the true ratio of phenotypes di˙ers from 9:3:3:1.
This experiment gives no evidence of genetic linkage between the genes.
14. (a) H0 : p = 0.75 (The true proportion of round seeds is 0.75.)

Ha : p =
6 0.75 (The true proportion of round seeds is not 0.75.)
(b) Of the 80 plants, 72 had round seeds and 8 had wrinkled seeds (these are the
observed counts). If the null hypothesis is true, we would expect to get, on
average, 0.75 × 80 = 60 plants with round seeds, and 0.25 × 80 = 20 with
wrinkled seeds.
2 2
χ2 = (72−60)
60 + (8−20)
20 = 9.6.
(c) The p-value is the area to the right of 9.6 under the χ2 distribution with
2 − 1 = 1 degrees of freedom. Using software, the p-value can be found to
be 0.0019. (Using a χ2 table, we could give only a range, such as 0.001 <
p-value < 0.005.)
(d) The p-value is less than the given signifcance level (0.0019 < 0.05) and so
the evidence against the null hypothesis is statistically signifcant at α = 0.05.
There is very strong evidence that the true proportion of plants with round
seeds di˙ers from 0.75. (Note that the observed proportion of round seeds
( 72
80 = 0.90) is greater than the hypothesized proportion. It appears as though
the true proportion of round seeds is greater than 0.75.)
15. (a) p̂ = 72
80 = 0.90. Z = q0.90−0.75
.75(1−.75)
= 3.098.
80
(b) p-value = 0.0019. (The p-value is double the area to the right of 3.098 under
the standard normal curve, which can be found using software or the standard
normal table.)
(c) The square of the Z test statistic is exactly equal to the χ2 test statistic.
(3.0982 = 9.6 (other than a little rounding error).)
(d) The p-values are exactly equal. The two tests are equivalent.
16. (a) H0 : The frst digit has a uniform distribution.

Ha : The frst digit does not have a uniform distribution.
Alternatively:
H0 : p1 = p2 = p3 = p4 = p5 = p6 = p7 = p8 = p9 = 19 .
Ha : These probabilities are not all correct.
(b) # of categories − 1 = 9 − 1 = 8.
(c) p-value = 5.6 × 10−8 . The p-value is the area to the right of 49.29 under a χ2
distribution with 8 degrees of freedom. Using software we can fnd that this
area is 5.6 × 10−8 . (Using a χ2 table, we could say only that the p-value is
very near 0.)
(d) The p-value of 5.6 × 10−8 is tiny, indicating extremely strong evidence against
the null hypothesis. There is extremely strong evidence that the frst digit
does not follow a uniform distribution.
(e) The fgure shows that the smaller values (such as 1 and 2) are more likely than
the uniform distribution would predict, and that the larger values (such as 7,
8, and 9) are less likely than the uniform distribution would predict.
(f) H0 : The frst digit follows Benford’s law.
Ha : The frst digit does not follow Benford’s law.

Alternatively:
H0 : p1 = 0.30, p2 = 0.18, p3 = 0.12, p4 = 0.10, p5 = 0.08, p6 = 0.07, p7 = 0.06, p8 =
0.05, p9 = 0.05
Ha : These probabilities are not all correct.
(g) # of categories − 1 = 9 − 1 = 8.
(h) p-value = 0.82. The p-value is the area to the right of 4.41 under a χ2 distri-
bution with 8 degrees of freedom. Using software we can fnd that this area is
0.82. (Using a χ2 table, we could only give a range, such as p-value > 0.20.)
(i) The p-value is large (0.82), indicating no evidence against the null hypothesis.
There is no evidence that the distribution of the frst digit di˙ers from Ben-
ford’s law. (The observed data is consistent with Benford’s law being the true
distribution.)
(j) Yes. The hypothesis test showed no evidence that the distribution di˙ered from
Benford’s law, and the fgure shows that the predicted values from Benford’s
law are very close to what was observed in the sample.
17. (a) H0 : ABO blood type and Rh factor are independent.

Ha : ABO blood type and Rh factor are not independent.
(b) Row total×Column
Overall total
total
= 39×101
305 = 12.91.
(c) (Number of rows − 1)(Number of columns − 1) = (2 − 1)(4 − 1) = 3.
(d) p-value = 5.5 × 10−6 (found using software). If we used a table instead of
software we could only give a range for the p-value, such as p-value < 0.0005.
(e) Since the p-value is less than the given signifcance level (5.5×10−6 < 0.05), we
can say that the evidence against the null hypothesis is statistically signifcant
at α = 0.05. (The p-value is much less than 0.05, so the evidence against the
null hypothesis would be signifcant at much smaller values of α, but we only
consider the given α level.) This means there is very strong evidence that
ABO blood type and Rh factor are not independent (among Libyan medical
school students).
(f) The authors of the study report that this was a random sample of Libyan
medical school students from a certain university in Libya. Thus (strictly
speaking) our conclusions apply only to the population of Libyan medical
school students at this university.
(g) This is an unusual result because ABO blood type and Rh factor are controlled
by di˙erent genes that assort independently. So, at conception at least, ABO
and Rh are independent variables. If these are truly independent variables,
then it’s unusual to see such a large χ2 test statistic and resulting tiny p-value.
18. (a) 18.3% of the 60 of peptic ulcer patients have blood type A. 34.1% of the 44
healthy individuals have blood type A.
(b) H0 : The distribution of ABO blood type is the same for peptic ulcer su˙erers
and healthy individuals.

Ha : The distribution of ABO blood type is di˙erent for peptic ulcer su˙erers
and healthy individuals.
(c) Row total×Column total
Overall total = 60×26
104 = 15.
degrees of freedom. p-value = 0.07026 (found using software). If we used a
table instead of software we could only give a range for the p-value, such as
0.05 < p-value < 0.10.
(f) Since the p-value is greater than the given signifcance level of 0.05, the evi-
dence against the null hypothesis is not statistically signifcant at this level. We
do not have strong evidence of a di˙erence in the distribution of ABO blood
type between peptic ulcer su˙erers and healthy individuals. (It is plausible
that the distribution of ABO blood type is the same for the two groups.) The
smallish p-value of 0.07 gives a hint of evidence against the null hypothesis,
but it is not statistically signifcant at α = 0.05.
19. (a) The proportion of male births to non-smoking couples is 2685 5045 = 0.532. The
proportion of male births for couples where the father is a non-smoker but the
mother is a heavy smoker is 175 365 = 0.479.
(b) H0 : Sex of the baby and maternal smoking status are independent variables.
Ha : Sex of the baby and maternal smoking status are not independent vari-
ables.
Overall total
total
= 3595×5045
6853 = 2646.545.
0.05 < p-value < 0.10.
(f) Since the p-value is greater than the given signifcance level of 0.05, the evi-
dence against the null hypothesis is not statistically signifcant at this level.
We do not have strong evidence that male births and maternal smoking status
are not independent. (It is plausible that they are independent.) The smallish
p-value of 0.064 gives a hint of evidence against the null hypothesis, but it is
not statistically signifcant at α = 0.05.
(g) The sample involved births to parents in Liverpool, UK, in which the father
was a non-smoker. As such, our conclusions apply only to the population of
Liverpool births in which the father is a non-smoker. (Technically speaking
scenarios (all UK births, for example).)
20. (a) The proportion of male births to non-smoking couples is 2685 5045 = 0.532. The
proportion of male births for couples where the mother is a non-smoker but
the father is a heavy smoker is 122 245 = 0.498.
(b) H0 : Sex of the baby and paternal smoking status are independent variables.
Ha : Sex of the baby and paternal smoking status are not independent variables.
Overall total
total
= 3192×5045
6032 = 2669.7.
p-value > 0.10.
(f) Here the p-value is large (0.48), indicating absolutely no evidence against the
null hypothesis. The observed data is consistent with sex of the baby and
paternal smoking status being independent variables.
(g) The sample involved births to parents in Liverpool, UK, in which the mother
was a non-smoker. As such, our conclusions apply only to the population of
Liverpool births in which the mother is a non-smoker. (Technically speaking
scenarios (all UK births, for example).)
21. (a) For female murderers, 19.8% of the murders were of family members. For
male murderers, 4.4% of the murders were of family members. (Note that the
percentage for female murderers is much higher.)
(b) H0 : The distribution of relationship to the victim is the same for male and
female murderers.
Ha : The distribution of relationship to the victim is di˙erent for male and
female murderers.
Overall total
total
= 22×91
182 = 11.
p-value < 0.001.
(f) Here the p-value is very small (0.00083), indicating very strong evidence against
the null hypothesis. There is very strong evidence that the distribution of re-
lationship to the victim for female murderers di˙ers from the distribution of
relationship to the victim for male murderers.
(g) It appears as though females were more likely than males to murder family
members or partners, and males were more likely than females to murder
acquaintances. (We could investigate this in greater detail using plots and
statistical inference techniques.)
(h) The samples were of convicted male and female murderers in Finland, so those
are the populations to which the conclusions apply.
22. (a) H0 : pplacebo = pvaccine (the vaccine has no e˙ect).

6 pvaccine (the vaccine has an e˙ect).
Ha : pplacebo =
χ2 = 22.02188. The p-value is the area to the right of this value under a χ2
distribution with 1 degree of freedom. By computer: p-value = 2.7 × 10−6 . By
the table: p-value < 0.0005.
There is very strong evidence (p-value = 2.7 × 10−6 ) of a relationship between
vaccine use and the rate of polio. From an inspection of the values in the table,
we can see that individuals that received the vaccine were less likely to have
gotten polio. In addition, since this was a well-designed experiment, there is
very strong evidence that the polio vaccine causes a reduction in the rate of
polio.
(b) 4.69272 = 22.02188 (other than a little round-o˙ error). In general for 2 × 2
tables of count data, Z 2 = χ2 , and the Z and χ2 tests are equivalent tests.
23. (a) H0 : Text colour and ID are independent variables. (The likelihood of a stu-
dent flling out their ID number incorrectly is the same for the two colours.)
Ha : Text colour and ID not independent variables. (The likelihood of a stu-
dent flling out their ID number incorrectly is di˙erent between the colours.)
(b) Expected number of white-test writers that incorrectly fll out their ID: 17×423
842 =
8.54
Expected number of yellow-test writers that incorrectly fll out their ID: 17×419
842 =
8.46
Expected number of white-test writers that properly fll out their ID: 825×423
842 =
414.46
Expected number of yellow-test writers that properly fll out their ID: 825×419
842 =
410.54
2 2 2 2
χ2 = (1−8.54)
8.54 + (16−8.46)
8.46 + (422−414.46)
414.46 + (403−410.54)
410.54 = 13.654
(c) Using software, the p-value is 0.00022. (Using the χ table we could give only
2
a range of values, such as p-value < 0.0005.

(d) The observed data does give strong evidence against the null hypothesis (p-value =
0.00022). For a few reasons, I believe the observed data was simply a fuke
and the observed e˙ect is not a real one. Consider the following factors:
• The hypothesis test was only being carried out because the observed data
was unusual and unexpected. We should not use the same data that sug-
gests a hypothesis to test that hypothesis, as that can lead to a misleading
p-value, and misleading conclusions. (It would have been far more con-
vincing had I planned to test this hypothesis before the students wrote
the test.)
• There is no plausible explanation for why yellow-test writers would be
more likely to fll out an incorrect ID number. (It may be the case that
the observed e˙ect is a real one, but it would be much more convincing if
we had a plausible explanation for the e˙ect that was seen in the sample.)
• This e˙ect was not seen in many other tests, including tests with the same
students in the same course.
24. (a) H0 : p1 = p2 = p3 = 13 (the pictures are equally preferred). Ha : The probabil-

ities are not all equally likely (the pictures are not equally preferred).
(b) Observed counts: 24, 37, 39. Expected counts: 33.333, 33.333, 33.333.
2 2 2
χ2 = (24−33.333)
33.333 + (37−33.333)
33.333 + (39−33.333)
33.333 = 3.98.
(c) The p-value is the area to the right of 3.98 under a χ2 distribution with 2
degrees of freedom. Using a computer, p-value = 0.1367. Using a table, we
could give only a range, such as: 0.10 < p-value < 0.15.
(d) There is very little evidence (0.10 < p-value < 0.15) against the hypothesis
that the three pictures are equally preferred.
25. (a) H0 : The true ratio is 1:1:2:2. (3 and 4 stars are twice as likely as 1 and 2
stars).
Ha : The true ratio is not 1:1:2:2.
(b) Observed counts: 12, 79, 209, 100. Expected counts: 66.667, 66.667, 133.333,
133.333. 2 2 2 2
χ2 = (12−66.667)
66.667 + (79−66.667)
66.667 + (209−133.333)
133.333 + (100−133.333)
133.333 = 98.38.
(c) The p-value is the area to the right of 98.38 under a χ distribution with
2
4 − 1 = 3 degrees of freedom. This area is near 0.

(d) There is extremely strong evidence against the movie executive’s claim. The
probability of seeing what was seen in this sample, if the executive’s claim is
true, is near 0.
26. H0 : p1 = p2 = p3 = 13 (the 3 favours are equally preferred).

Ha : pi =
6 31 for at least one i (the 3 favours are not equally preferred).
If the 3 favours are equally preferred, we would expect to get 40 observations in

each cell on average.
(26 − 40)2 (67 − 40)2 (27 − 40)2

χ2 = + + = 27.35.
40 40 40
The p-value is the area to the right of 27.35 under a χ2 distribution with 3 − 1 = 2
degrees of freedom. Using software: p-value = 1.15 × 10−6 . Using the table:
p-value < 0.0005.
There is very strong evidence (p-value = 1.15 × 10−6 , signifcant at α = 0.05) that
the favours are not equally preferred.
27. The p-value is the area to the right of 377.42 under a χ2 distribution with (3 −
1)(4 − 1) = 6 degrees of freedom. This area is near 0. There is extremely strong
evidence that age and rating are not independent for viewers of this movie.
28. C is the best option. The tiny p-value gives very strong evidence of an association
between the type of concrete and the type of damage.
29. The χ2 test statistic works out to 97.4709.
The p-value is the area to the right of 97.4709 under a χ2 distribution with (2 −
1) × (3 − 1) = 2 degrees of freedom. Using software, we can fnd that the p-value is
very near 0 (so close to 0 that most software will simply report it as 0). Using the
χ2 table, we can fnd that the p-value is much less than 0.0005.
The null hypothesis is that there is no association between the row and column
variables (in other words, that the methods are all equally e˙ective at killing bed-
bugs). Since we have a tiny p-value, there is very strong evidence against the null
hypothesis. As such, there is very strong evidence that the methods are not all
equally e˙ective at killing bedbugs.
Chapter 14
One-Way ANOVA
J.B.’s strongly suggested exercises: 2, 3, 5, 8, 10, 12, 13, 16, 18, 19, 22, 23, 25, 26, 28
14.1 Introduction
14.2 One-Way ANOVA
1. (a) H0 : µ1 = µ2 = . . . = µk (The population means of the k groups are all equal.)

(b) Ha : µi =
6 µj for at least one i, j pair. (The population means of the k groups
are not all equal.)
14.3 Carrying out the One-Way ANOVA
14.3.1 The Formulas
2. (a)
Source DF SS MS F p-value
(800/3)
Groups 4−1=3 800 800
3 ≈ 266.667 (200/36) = 48 p-value < 0.001
Error 40 − 4 =36 1000 − 800 = 200 200
36 ≈ 5.556 — —
Total 40 − 1 = 39 1000 — — —
(b) M SE = sp ≈ 5.556.
2
(c) F = 48.
(d) Yes (since p-value < 0.05).
209
Balka ISE 1.09 14.3. CARRYING OUT THE ONE-WAY ANOVA 210
3. (a) s2p = M SE = SSE n−k , where SSE = (ni − 1)s2i = (10 − 1)2.12 + (10 − 1)4.22 +
P
(8 − 1)3.72 = 294.28. M SE = 294.28

28−3 = 11.7712.
(b) SST = ¯ ¯
ni (Xi − X ) = 10(3.1 − 10.11429)2 + 10(10.1 − 10.11429)2 + 8(18.9 −
2
P
2
10.11429) = 1109.514.
(c)
Groups 3−1=2 1109.514 1109.514
2
554.757
= 554.757 11.7712 = 47.12833 p-value < 0.001
Error 28 − 3 = 25 294.28 11.7712 — —
Total 28 − 1 = 27 1403.794 — — —
(d) p-value < 0.001
(e) There is very strong evidence that the three groups do not all have the same
population mean. (The given signifcance level is 0.05, and since the p-value
is less than 0.05, we can say that the evidence against H0 is signifcant at the
5% level.)
4. (a) X ¯ 2 = 12, X
¯ 1 = 5, X ¯ 3 = 6, n1 = 2, n2 = 3, n3 = 2, s2 = 2, s2 = 16, s2 = 8.
1 2 3
X¯ = 58 = 8.285714.
7
SST = 2×(5−8.285714)2 +3×(12−8.285714)2 +2×(6−8.285714)2 = 73.42857.
(2 − 1)2 + (3 − 1)16 + (2 − 1)8
(b) s2p = M SE = = 10.5.
2+3+2−3
M ST = SSTk−1 =
73.42857
2 = 36.71428.
(c) F = M SE = 10.5 = 3.497.
M ST 36.71428
14.3.2 An Example with Full Calculations
5. (a) Yes, ANOVA is appropriate here. We have a categorical explanatory variable

with 3 levels (truth telling, unrehearsed lying, and rehearsed lying), and a
quantitative response variable (eye movement). We are interested in investi-
gating possible di˙erences in the mean response between the groups, so our
goals mesh with the goals of ANOVA.
The standard deviations for the di˙erent groups are very similar (36, 44, 38),
so the common variance assumption appears to be reasonable. The sample
sizes are large, so the normality assumption is not that important. Overall it’s
very reasonable to use ANOVA here. But we should plot the data, as this can
give us an overall view of what we are dealing with, help to assess normality,
and show any outliers or other problems.
(b) It’s very reasonable to use ANOVA here, even if there is a little right skewness.
ANOVA works well, even if there is a little non-normality in the groups (espe-
cially if the sample sizes are large, and they are fairly large here). We should
Balka ISE 1.09 14.4. WHAT SHOULD WE DO AFTER A ONE-WAY ANOVA? 211
plot the data and investigate the skewness though—if the skewness is strong
we might consider using a transformation before carrying out the ANOVA.
(For example, we might consider carrying out an ANOVA on the logs of the
eye movement measurements.)
(c) There is statistically signifcant evidence (p-value = 0.019 < 0.05) that the
groups do not all have the same population mean eye movement. (In other
words, there is statistically signifcant evidence that the e˙ect on eye movement
di˙ers between the treatment groups.)
(d) No, the analysis is not yet complete. Since there is statistically signifcant
evidence that the groups do not all have the same true mean, we will investigate
the di˙erences between the groups using pairwise comparisons.
14.4 What Should We Do After a One-Way ANOVA?
14.4.1 Introduction
6. 0.9525 = 0.277.
7. (1 − 0.05)10 = 0.5987.
14.4.2 Fisher’s LSD Method
8. (a) There are 32 = 3 pairwise comparisons. We will estimate µU L − µT T , µRL −

�
µT T and µU L − µRL .
(b) A 95% confdence √ interval
p for µi − µj is Xi − Xj ± t.025 SE(Xi − Xj ), where
¯ ¯ ¯ ¯
SE(Xi − Xj ) = M SE 1/ni + 1/nj . The t has 122 degrees of freedom (the
¯ ¯
degrees of freedom for error), yielding t.025 = 1.9796 (found using software –
the value may be slightly di˙erent if you look it up in a t table).
The interval for µU L − µT T is:
¯U L − X
X ¯ T T ± t.025 SE(X ¯T T )
¯U L − X
√ p
63 − 42 ± 1.9796 × 1561 1/42 + 1/41
21 ± 17.17
The interval for µRL − µT T is:
¯ RL − X
X ¯ T T ± t.025 SE(X ¯T T )
¯ RL − X
√ p
64 − 42 ± 1.9796 × 1561 1/42 + 1/41
22 ± 17.17

The interval for µU L − µRL is:
¯U L − X
X ¯ RL − X
¯ RL ± t.025 SE(X ¯T T )
√ p
63 − 64 ± 1.9796 × 1561 1/42 + 1/42
−1 ± 17.07

(c) The interval for µU L − µRL contains 0, and so there is not statistically signif-
icant evidence (at α = 0.05) of a di˙erence in mean eye movement between
these groups.
The intervals for µU L − µT T and µRL − µT T do not contain 0, and thus there
is statistically signifcant evidence that µU L and µRL both di˙er from µT T .
Since both intervals lie entirely to the right of 0, we have strong evidence that
µU L and µRL are both greater than µT T
(d) The true family-wise confdence level is unknown, as it depends on the de-
pendency between the comparisons, but it will be less than 95%. For a little
perspective, if the comparisons were independent (they are not), the family-
wise confdence level would be 0.953 × 100% = 85.7%.
(e) From the ANOVA, there is statistically signifcant evidence (p-value = 0.019 <
0.05) that the di˙erent lie conditions do not all result in the same true mean eye
movement. 95% LSD confdence intervals showed that both the unrehearsed
lying group and the rehearsed lying group had mean eye movements that
were signifcantly greater than that of the truth tellers, but there was not a
statistically signifcant di˙erence between the rehearsed and unrehearsed liars.
14.4.3 The Bonferroni Correction
9. (a) 52 = 10.
�
(b) If we want to keep α at no more than 0.10, for the individual intervals we use
α0 = 010
.10
= 0.01. Each interval should have a 99% confdence level.
(c) 2.787. t0.01/2 = t0.005 . There are n − k = 30 − 5 = 25 degrees of freedom.
t0.005 = 2.787.
10. (a) There are 32 = 3 pairwise comparisons. We will estimate µU L − µT T , µRL −

�
µT T and µU L − µRL .
(b) The family-wise confdence level of 95% implies α = 0.05. For the 3 individual
intervals, α0 = 0.05
3 , with a resulting confdence interval of (1 − 3 )100% =
0.05
98.33%.
(c) The appropriate t value is tα0 /2 = t.05/6 = t0.00833 . With 122 degrees of freedom
(the degrees of freedom for error), t0.00833 = 2.4274 (found using software – it
cannot be found from the t table). A 95% confdence √ intervalpfor µi − µj is

X¯i − X ¯ j ), where SE(X
¯i − X
¯ j ± t0.00833 SE(X ¯ j ) = M SE 1/ni + 1/nj .
¯i − X
The interval for µU L − µT T is:
¯U L − X
X ¯U L − X
¯ T T ± t0.00833 SE(X ¯T T )
√ p
63 − 42 ± 2.4274 × 1561 1/42 + 1/41
21 ± 21.06

The interval for µRL − µT T is:
¯ RL − X
X ¯ RL − X
¯ T T ± t0.00833 SE(X ¯T T )
√ p
64 − 42 ± 2.4274 × 1561 1/42 + 1/41
22 ± 21.06

The interval for µU L − µRL is:
¯U L − X
X ¯ RL − X
¯ RL ± t0.00833 SE(X ¯T T )
√ p
63 − 64 ± 2.4274 × 1561 1/42 + 1/42
−1 ± 20.93

(d) The 95% family-wise intervals for µU L −µT T and µU L −µRL both contain 0, and
thus the observed di˙erences X ¯ T T and X
¯U L − X ¯U L − X¯ RL are not statistically
signifcant at the family-wise signifcance level of 5%. The interval for µRL −
µT T did not contain 0, and thus the X ¯ RL − X¯ T T di˙erence is statistically
signifcant. (The interval for µU L − µT T just barely contains 0. There is some
evidence of a true di˙erence here, but it is not statistically signifcant at the
given family-wise signifcance level.)
(e) The Bonferroni intervals have the same midpoints (X ¯ j ) as the LSD inter-
¯i − X
vals, but are wider (the margin of error is greater). The LSD intervals used a t
value of 1.9796, whereas the Bonferroni intervals used a t value of 2.4274. (The
1.9796 = 1.23 shows that the Bonferroni intervals are about 23% wider
ratio 2.4274
than the LSD intervals.) In this specifc example, this resulted in one of the
comparisons (unrehearsed lying vs truth telling) changing from a signifcant
di˙erence under LSD to a not signifcant di˙erence under Bonferroni.
(f) From the ANOVA, there is statistically signifcant evidence (p-value = 0.019 <
0.05) that the di˙erent lie conditions do not all result in the same true mean eye
movement. Bonferroni confdence intervals with a 95% family-wise confdence
level showed that the rehearsed lying group had a mean eye movement that was
signifcantly greater than that of the truth tellers, but the other two di˙erences
were not statistically signifcant.
14.4.4 The Tukey Procedure
11. (a) The Tukey procedure is an exact procedure if the normality assumption is
met, the common variance assumption is met, and the sample sizes are equal.
Although not an exact procedure if the sample sizes are di˙erent, it can still
be used as an approximate procedure. Here, the sample sizes are very close
(41, 42, 42), and so it is very reasonable to use the Tukeyp
procedure.
√
(b) The interval for µU L − µT T is: 63 − 42 ± 2.3725 × √1561p1/42 + 1/41.
The interval for µRL − µT T is: 64 − 42 ± 2.3725 × √1561p1/42 + 1/41.
The interval for µU L − µRL is: 63 − 64 ± 2.3725 × 1561 1/42 + 1/42.
(c) The interval for µU L − µRL contains 0, and so there is not statistically signif-
cant evidence of a di˙erence in mean eye movement between the unrehearsed
lying and the rehearsed lying groups.
The intervals for µU L − µT T and µRL − µT T do not contain 0, and thus there
is statistically signifcant evidence that µU L and µRL both di˙er from µT T .
Since both intervals lie entirely to the right of 0, we have strong evidence that
µU L and µRL are both greater than µT T .
(d) The LSD intervals used a multiplier (t0.025 ) of 1.9796, the Bonferroni intervals
used a multiplier (t0.00833 ) of 2.4274, and the Tukey intervals used a multiplier
1.9796 = 1.198 shows that the Tukey intervals are about
of 2.3725. The ratio 2.3725
20% wider than the LSD intervals. The ratio 2.3725 2.4274 = 0.977 shows that the
Tukey intervals are a little narrower (≈ 2% narrower) than the Bonferroni in-
tervals. Like the Bonferroni method, the Tukey method maintains the family-
wise confdence level of 95%, but it manages to do so with narrower intervals.
In this example, that resulted in one comparison crossing over to statistical
signifcance at the 5% level. (The unrehearsed lying − rehearsed lying di˙er-
ence was statistically signifcant using the Tukey method, but not signifcant
using the Bonferroni approach.)
(e) From the ANOVA, there is statistically signifcant evidence (p-value = 0.019 <
0.05) that the di˙erent lie conditions do not all result in the same true mean
eye movement. Tukey confdence intervals with a family-wise confdence level
of 95% showed that both the unrehearsed lying group and the rehearsed ly-
ing group had mean eye movements that were signifcantly greater than the
truth tellers, but there was not a statistically signifcant di˙erence between
the rehearsed and unrehearsed liars.
14.5 Examples
14.6 A Few More Points
12. We assume that we have simple random samples from the k populations, and that
the populations are normally distributed with common variance σ 2 . If the assump-
tions are not true, then our conclusions may be misleading (for example, the stated
Type I error may be very di˙erent from the true probability of rejecting a true null
hypothesis).
14.8 Basic Calculations
13. (a) 0.025.

(b) 0.95.
(c) 6.16.
(d) 2.93.
14.9 Concepts
14. No. In one-way ANOVA, only large values of the F test statistic give strong evidence
15. (a) The test statistic will equal 0 if M ST = 0 (provided M SE 6= 0). MST will
equal 0 when there is no variability in the sample means—when the k groups
all have the same value of the sample mean.
(b) The test statistic will equal 1 if M ST = M SE.
16. (a) If we performed a one-way ANOVA on these 3 samples, then the p-value would
be small. True. Visually there is very, very strong evidence against H0 , im-
plying a small p-value .
(b) If we carried out a t-test of H0 : µA = µC against a two-sided alternative, then
the p-value would be small. True. Visually there is very, very strong evidence
against H0 , implying a small p-value .
(c) The use of the ANOVA procedures for the test of H0 : µA = µB = µC would
be a bad idea, as the assumptions are clearly violated. False. There is no
Balka ISE 1.09 14.9. CONCEPTS 216
visual evidence against the assumptions of normality and a common population

variance.
(d) If we carried out a pooled-variance t-test of H0 : µA = µC against a two-sided
alternative, then a one-way ANOVA of the test of H0 : µA = µC , then the
test statistics would be related by: t2 = F . True. For two-sample problems
the pooled-variance t procedure and one-way ANOVA are equivalent tests, with
t2 = F .
(e) If we carried out a pooled-variance t-test of H0 : µA = µC against a two-sided
alternative, then a one-way ANOVA of the test of H0 : µA = µC , the p-values
would be exactly equal. True. For two-sample problems the pooled-variance
t procedure and one-way ANOVA are equivalent tests, resulting in the exact
same p-value.
17. (a) The test statistic will have an F distribution with 4 df in the numerator, and
45 df in the denominator. If the null hypothesis and the assumptions are true,
the F test statistic has an F distribution with k − 1 degrees of freedom in the
numerator and n − k degrees of freedom in the denominator. If there are 10
observations in each of 5 groups, there are 5 − 1 = 4 degrees of freedom in the
numerator, and 50 − 5 = 45 degrees of freedom in the denominator.
(b) If the null hypothesis and the assumptions are true, then the p-value will have
a uniform distribution between 0 and 1.
(c) Since the p-value is uniformly distributed between 0 and 1, the p-value will
equal 0.5 on average (if the null hypothesis and the assumptions are true).
18. (a) If the null hypothesis (and the assumptions) are true, then the test statistic
in one-way ANOVA has an F distribution. True.
(b) In one-way ANOVA, we assume that the observations within each group are
normally distributed, and that all groups have the same population variance.
True.
(c) In one-way ANOVA, we assume that the observations within each group are
normally distributed, and that all groups have the same population mean.
False
(d) The test statistic in one-way ANOVA can be negative. False. The F statistic
is a ratio of variances, and as such cannot be negative.
(e) If the null hypothesis is false, then MST will tend to be bigger than MSE.
True.
19. (a) If the null hypothesis is true, then the F statistic will be infnite. False (of
course).
(b) If the null hypothesis is true, then the F statistic will equal 1. False. If H0 is
true, the F statistic will have a median that is near 1, but it can take on any
value between 0 and ∞.
(c) If the null hypothesis is false, then the F statistic will sometimes be less than
one. True. The F statistic can take on any nonnegative value, regardless of
whether the null hypothesis is true or false.
(d) If the null hypothesis is false, then the F statistic will sometimes be less than
0. False. The F statistic cannot be negative.
(e) If the null hypothesis is false, then the p-value will be less than 0.05. False. If
the null hypothesis is false, we can still end up with a large p-value.
20. (a) If M SE = 0, then the p-value will equal the F statistic. False.
(b) If the population means are equal, then the F statistic will equal 1. False. If
the population means are equal, then H0 is true and the F statistic will have
an F distribution.
(c) If the sample means are all equal, then the F statistic will equal 1. False. If
the sample means are equal, the F statistic will equal 0.
(d) If the sample means are all equal, then the null hypothesis is true. False. If
the sample means are all equal, then there is absolutely no evidence whatsoever
against H0 . But that still does not mean it must be true.
(e) The mean square error is the pooled variance. True.
21. (a) If the population means are very di˙erent, we expect the F statistic to be
greater than one. True. If H0 is true (the population means are equal), then
the F statistic will have a median that is somewhere around 1. If H0 is false,
the F statistic will tend to be larger than 1.
(b) We will reject the null hypothesis at the 5% level whenever the sample means
are all equal. False. When the sample means are all equal there is absolutely
no evidence whatsoever against the null hypothesis.
(c) An assumption of one-way ANOVA is that the groups all have di˙erent pop-
ulation variances. False. One-way ANOVA assumes the population variances
are equal.
(d) If the assumptions of the pooled-variance t procedures are met, then the t test
and the F test are equivalent tests, with t2 = F . True.
(e) The equal variance assumption is not important for large sample sizes, due to
the central limit theorem. False. A violation of the equal variance assumption
can be very problematic, even for large sample sizes. The central limit theorem
can help out with a violation of the normality assumption.
14.10 Applications
22. (a) The ANOVA assumptions appear to be reasonable here. There are a few
outliers, which can cause problems sometimes, but there are not many and
they are not that extreme. The standard deviations between the groups are
very similar, so the assumption of equal population variances appears to be
reasonable. The boxplots do not give any indication of non-normality. (And

the normality assumption is not very important, since the sample sizes are
quite large.)
(b) There is extremely strong evidence (p-value < 2.2 × 10−16 ) that the di˙erent
treatments do not all result in the same mean lifetime. (In other words, there
is very strong evidence that the diets do not all have the same e˙ect.)
(c) There are 6 groups, so there are 62 = 15 pairwise comparisons (µN P − µR1 ,
�
µN P − µR2 , . . . , µR4 − µR5 ).

(d) A 95% confdence interval √ −µR1 is X
for µN P p ¯ R1 ±t.025 SE(X
¯ N P −X
√
¯ ¯
p N P −XR1 ),
where SE(X ¯ N P −X ¯ R1 ) = M SE 1/nN P + 1/nR1 = 44.5 1/49 + 1/57 =
1.2996. The t has 325 degrees of freedom (the degrees of freedom for error),
yielding t.025 = 1.9673 (found using software – the value may be slightly dif-
ferent if you look it up in a t table). The interval is thus:
¯N P − X
X ¯N P − X
¯ R1 ± t.025 SE(X ¯ R1 )
27.4 − 32.7 ± 1.9673 × 1.2996
−5.3 ± 2.56
The interval lies entirely to the left of 0, and thus there is a statistically signif-
icant di˙erence between X ¯ N P and X ¯ R1 . (There is strong evidence that µN P
is less than µR1 .)
Non-technical summary: It appears as though mice assigned to the calorie-

restricted diet R1 tend to live longer than those without calorie restriction.
(e) A 95% confdence interval √ for µR2 −pµR3 is XR2 − XR3 ±
¯ ¯
√t.025 SE(X
¯ R2 − X¯ R3 ),
where SE(XR2 − XR3 ) = M SE 1/nR2 + 1/nR3 = 44.5 1/71 + 1/56 =
¯ ¯
p
1.1922. The t has 325 degrees of freedom (the degrees of freedom for error),
yielding t.025 = 1.9673. The interval is thus:
¯ R2 − X
X ¯ R2 − X
¯ R3 ± t.025 SE(X ¯ R3 )
42.3 − 42.9 ± 1.9673 × 1.1922
−0.6 ± 2.3454
The interval contains 0, and thus there is not a statistically signifcant di˙er-
ence between X¯ R2 and X¯ R3 . (There is not signifcant evidence that µR2 di˙ers
from µR3 .)
Non-technical summary: It appears as though mice assigned to the R2 diet

may have the same or similar mean lifetime as those on the R3 diet.
23. (a) There are no outliers or major skewness (nothing indicating non-normality),
which is good. The control group has a much greater sample standard devia-
tion than the antivenom groups (113 vs 56 and 45). This is a little troubling,
as the ANOVA assumes equal population variances. However, the sample sizes
are similar, and so the e˙ect of di˙erent variability may not be that great. It
is reasonable to use ANOVA, but it is not the ideal situation.
(b) There is absolutely no evidence (p-value = 0.98) of a di˙erence in the pop-
ulation means of the change in leg volume. There is no evidence that the
antivenom had any e˙ect on the mean swelling.
(c) Since we had no evidence against the null hypothesis in the ANOVA, the
analysis is complete. (There is no need to carry out pairwise comparisons.
The researchers could, if they so desired, compare the two antivenom groups
with a t test, since they had this comparison in mind before the experiment,
but there would not be a signifcant di˙erence in that comparison.)
24. (a) The CBT group shows several outliers, which can be a concern, but they
are not that extreme and it may be the result of having a small IQR. Other
than the outliers, the ANOVA assumptions are well satisfed—the distributions
look roughly symmetric and the standard deviations are very similar. It’s
reasonable to use ANOVA here.
(b) There is very strong evidence (p-value = 0.006) that the groups do not all have
the same population mean change in weight. (There is very strong evidence
that the treatments do not all have the same e˙ect.)
(c) There are 32 = 3 pairwise comparisons (µC − µCBT , µC − µF T , µCBT − µF T ).
�
(d) A 95% confdence p for µi − µj is Xi − Xj ± t.025 SE(Xi − Xj ), where

√ interval
¯ ¯ ¯ ¯
SE(Xi − Xj ) = M SE 1/ni + 1/nj . The t has 69 degrees of freedom (the
¯ ¯
degrees of freedom for error), yielding t.025 = 1.995 (found using software –
the value may be slightly di˙erent if you look it up in a t table).
The interval for µC − µCBT is:
X¯C − X¯ CBT ± t.025 SE(X¯C − X

¯ CBT )
√ p
−0.20 − 1.36 ± 1.995 × 11.66 1/26 + 1/29
−1.56 ± 1.8399

The interval for µC − µF T is:
X¯C − X¯ F T ± t.025 SE(X¯C − X

¯F T )
√ p
−0.20 − 3.30 ± 1.995 × 11.66 1/26 + 1/17
−3.50 ± 2.1248

The interval for µCBT − µF T is:

¯ CBT − X
X ¯ F T ± t.025 SE(X¯ CBT − X
¯F T )
√ p
1.36 − 3.30 ± 1.995 × 11.66 1/29 + 1/17
−1.94 ± 2.0809

Only the X ¯C − X¯ F T comparison has a statistically signifcant di˙erence at the
5% level. (There is statistically signifcant evidence that µC and µF T di˙er.)
Non-technical summary: The women in the Family Therapy treatment group
tended to gain more weight than the women in the control group. There was
not strong evidence of a di˙erence in weight gain between the control and CBT
groups, or the CBT and FT groups.
(e) There are 3 pairwise comparisons. To ensure the family-wise confdence level
is at least 95%, each interval should be done at a confdence level of (1 −
3 )100% = 98.33% (α = 0.05/3 = 0.0167). We need to fnd t0.0167/2 = t0.0083
0.05 0
with 69 degrees of freedom. Using software, we can fnd that with 69 degrees
of freedom t0.0083 = 2.4537.
The interval for µC − µCBT is:
X¯C − X ¯C − X
¯ CBT ± t.025 SE(X ¯ CBT )
√ p
−0.20 − 1.36 ± 2.4537 × 11.66 1/26 + 1/29
−1.56 ± 2.2629

The interval for µC − µF T is:
X¯C − X ¯C − X
¯ F T ± t.025 SE(X ¯F T )
√ p
−0.20 − 3.30 ± 2.4537 × 11.66 1/26 + 1/17
−3.50 ± 2.6133

The interval for µCBT − µF T is:
¯ CBT − X
X ¯ CBT − X
¯ F T ± t.025 SE(X ¯F T )
√ p
1.36 − 3.30 ± 2.4537 × 11.66 1/29 + 1/17
−1.94 ± 2.5593

Only the X¯C − X ¯ F T comparison has a statistically signifcant di˙erence at a
family-wise signifcance level of 5%. (There is statistically signifcant evidence
that µC and µF T di˙er.) In this case the conclusions from the LSD and the
Bonferroni procedures are essentially the same (but the Bonferroni procedure
results in slightly wider intervals).
Non-technical summary: The women in the Family Therapy treatment group
tended to gain more weight than the women in the control group. There was
not strong evidence of a di˙erence in weight gain between the control and CBT
groups, or the CBT and FT groups.
25. (a) The standard deviations are very similar, so there is no issue with the common
variance assumption. Overall, ANOVA is likely reasonable here. But we should
plot the data, as this can give us an overall view of what we are dealing with,
help to assess if the normality assumption is reasonable, and show any outliers
or other problems.
(b) There is no evidence (p-value = 0.64) of a di˙erence in the true mean SDO of
the di˙erent ethnic groups. (These conclusions apply only to the male inmate
population from which the sample was drawn.)
26. (a) Any subsequent analysis would depend on the assumption that it is reasonable
to pool the individuals together. If the distribution of SDO did vary between
the ethnic groups in a meaningful way, then conclusions based on a pooled
analysis would be suspect. For example, if the pooled sample showed no
relationship between age and SDO, this might be because the e˙ect of age
di˙ered between the ethnic groups and the pooling averaged it out. If we pool
over an important variable, the resulting conclusions can sometimes be very
misleading.
(b) There is extremely strong evidence (p-value = 3.8 × 10−8 ) that the age groups
do not all have the same population mean SDO. (These conclusions apply only
to the male inmate population from which the sample was drawn.)
(c) There is extremely strong evidence that there is a relationship between age
and SDO. From the values given in the table, it appears as though younger
inmates tend to have a higher SDO, and are thus more accepting of the notion
that it is reasonable for some groups to dominate� others. (We could investigate
this a little more formally by carrying out the 42 = 6 pairwise comparisons.

And since age is a quantitative variable, we could also investigate this by using
regression techniques instead of ANOVA.)
27. (a) The ANOVA table is given in Table 14.1.
Groups 6−1=5 8117.2 − 6862.8 = 1254.4 1254.4
5 = 250.88 250.88
60.2 = 4.167 < .01
Error 120 − 6 = 114 60.2 × 114 = 6862.8 60.2 — —
Total 120 − 1 = 119 8117.2 — — —
Table 14.1: ANOVA table for the teaching method question.

(b) F = 4.167. The p-value is the area to the right of 4.167 under an F distribution
with 5 and 114 degrees of freedom. Using a computer: p-value = 0.0016. Using
(c) There is signifcant evidence that the teaching methods do not all result in the
same mean test performance.
√
(d) SE(X ¯i − X¯ j ) = 60.2 1/20 + 1/20 = 2.4536. We now need t.05 with 114
p
degrees of freedom. The correct value is approximately 1.658 (by computer,

it might be slightly di˙erent using the table), yielding a margin of error of
1.658 × 2.4536 = 4.068.
(e) We require the di˙erent groups to have normally distributed populations with
common variance σ 2 . Normal quantile-quantile plots can help us check the
normality assumption. Summary statistics and boxplots could help us deter-
mine if the equal variance assumption is reasonable.
Source DF SS MS F
Treatments 9−1=8 612 − 324 = 288 288
= 36 36
= 3.00
28. (a) 8 12
Error 36 − 9 = 27 27 ∗ 12 = 324 12
Total 36 − 1 = 35 612
(b) Here we need the area to the right of 3 under an F distribution with 8 and
27 degrees of freedom. Using a computer: p-value = 0.015. Using the table:
0.01 < p-value < 0.025.
(c) There is signifcant evidence against the null hypothesis, which means there is
signifcant evidence that not all of the seat belt types have the same population
mean breaking strength.
√ p
(d) 2.052 × 12 1/4 + 1/4 = 5.026
(e) i. The sample means were not all equal. True. If the sample means were all
equal, then MST = 0.
ii. This was a balanced experimental design (same sample size in each group).
True, there were 4 observations in each group.
iii. We know for certain the population mean breaking strengths of the di˙er-
ent groups are not all equal. False. Statistical inference is not about being
certain.
iv. The pooled sample variance is equal to 12. True (the pooled variance is
MSE).
29. (a) H0 : µ1 = µ2 = µ3 = µ4 = µ5 (the mean fuel use is the same for all 5 additives).
Ha : µi 6= µj for at least one i, j combination. (the means of the 5 additives
are not all equal).
(b) We require the di˙erent groups to have normally distributed populations with
common variance σ 2 . Normal quantile-quantile plots can help us check the
normality assumption (although these may not be very informative with the
small sample sizes). Summary statistics and boxplots could help us determine
if the equal variance assumption is reasonable.
(c) The ANOVA table is given in Table 14.2.
Groups 5−1=4 680.2 680.2
4 = 170.05 170.05
18.533 = 9.175 p-value < .001
Error 20 − 5 = 15 958.2 − 680.2 = 278 18.533 — —
Total 20 − 1 = 19 958.2 — — —
Table 14.2: ANOVA table for the fuel additives question.
(d) F = 9.175. The p-value is the area to the right of 9.175 under an F distribution
with 4 and 15 degrees of freedom. Using a computer: p-value = 0.0006. Using
the table: p-value < 0.001.
(e) There is very strong evidence (p-value = 0.0006) that the additives do not all
result in the same population mean fuel use.
30. (a) H0 : µ1 = µ2 = µ3 = µ4 (the population mean strength is the same for all 4
solder types).
Ha : µi =
6 µj for at least one i, j combination. (the population means of the 4
solder types are not all equal).
(b) There is very strong evidence (p-value = 0.00017) that the types of tin-lead
solder do not allpresult in the same mean shear strength.
√
(c) 2.120 × 4.425 1/5 + 1/5 = 2.820.
Chapter 15
Introduction to Simple Linear

Regression
J.B.’s strongly suggested exercises: 2, 4, 6, 8, 9, 12, 14, 15, 16, 18, 20, 37, 40, 41, 42, 43
15.1 Introduction
1. In regression analysis, we use one or more explanatory variables to help us predict

or estimate a quantitative response variable. For example, in regression analysis we
might predict a person’s resting heart rate based on a number of variables, such as
age, weight, sex, and various health indicators.
2. (a) Since we wish to use high school GPA to help predict university GPA, high
school GPA will be the explanatory variable and university GPA will be the
response variable.
(b) There are a number of variables that might help to predict university GPA,
including the student’s socioeconomic background, gender, and high school
they attended. Regression analysis can help us to determine which of these
are useful predictors.
3. In simple linear regression there is one explanatory variable (X). In multiple linear
regression there is more than one explanatory variable (X1 , X2 , . . .). (In both cases
there is a single response variable Y .)
225
Balka ISE 1.09 15.2. THE LINEAR REGRESSION MODEL 226
15.2 The Linear Regression Model
4. The terms in the model:
• Y is the response variable. (We may wish to predict the value of Y .)

• X is the explanatory variable. (We use the explanatory variable X to help
predict the value of the response variable Y .)
• β0 is the Y intercept. In other words, it is the theoretical mean of Y when
X = 0. Although the intercept is required for the line, it may not have much
practical meaning.
• β1 is the slope of the line. In other words, it is the change in the mean of Y
for a 1 unit increase in X. (But note that there is not necessarily a causal link
between X and Y .)
• is a random error term, representing the fact that the response variable Y
varies about the regression line.
5. β̂1 is a statistic that is calculated using sample data. It estimates β1 , the true slope
(a parameter).
β̂0 is a statistic that is calculated using sample data. It estimates β0 , the true
intercept (a parameter).
6. β̂0 ≈ 10 and β̂1 ≈ −2.
βˆ1 is the slope. We know from our grade school days that Slope = RiseRun . When
X = 2, Y is approximately 6, and when X = 10, Y is approximately −10. This
is a drop of approximately 16 units (a rise of approximately −16) over a run of 8.
βˆ1 = Rise
Run ≈ 8 = −2. (The slope of the given line is approximately −2.) β̂0 is the
−16
value of Ŷ when X = 0. At X = 2, Ŷ ≈ 6. With a slope of approximately −2, this

implies that at X = 0, Ŷ ≈ 10, so β̂0 ≈ 10.
15.3 The Least Squares Regression Model
7. We
P 2use P the least squares method—the parameter estimate are chosen such that
ei = (Yi − (β̂0 + β̂1 Xi ))2 is a minimum. In other words, the parameter esti-
mates are chosen such that the sum of the squared vertical distances from the points
to the line is a minimum.
8. (a) β̂1 = SPXY

SSXX = 10.1
20.2 = 0.5. βˆ0 = Y¯ − β̂1 X
¯ = 23.0 − 0.5 × 12.0 = 17.0.
(b) The units of β̂0 are metres (the same units as Y ). The units of β̂1 are me-
tres/second ( Units of X ).
Units of Y
Balka ISE 1.09 15.4. STATISTICAL INFERENCE IN SIMPLE LINEAR REGRESSION 227
9. (a) Ŷ = −5.73529 + 0.63554X.

(b) β̂1 = 0.63554 is the estimated change in the mean abdomen length for a 1 mm
increase in snout-vent length. It is also the predicted di˙erence in abdomen
length between two lizards that di˙er by 1 mm in their snout-vent length. For
example, we would predict that a lizard with a snout-vent length of 51 mm
would have an abdomen length that is approximately 0.64 mm greater than
that of a lizard with a snout-vent length of 50 mm.
(c) β̂0 = −5.73529 is the estimated mean abdomen length for a lizard with a
snout-vent length of 0 mm. Note that this makes absolutely no sense from a
practical point of view. The snout-vent length of 0 is far beyond the range
of the observed data, and not something of any practical interest, so in this
example the value of the intercept has no practical meaning beyond being a
necessary part of the line.
(d) Ŷ = −5.73529 + 0.63554 × 52.0 = 27.31279.
(e) In Question 9d we found that the predicted abdomen length for a lizard with
a snout-vent length of 52.0 mm is Ŷ = −5.73529 + 0.63554 × 52.0 = 27.31279
mm. This lizard has a residual of: Y − Ŷ = 27.3 − 27.31279 = −0.01279.
(f) 0, since the residuals always sum to 0 in linear regression.
15.4 Statistical Inference in Simple Linear Regression
15.4.1 Model Assumptions
10. We assume that the observations are independent, and that the term is a random
variable that is normally distributed with a mean of 0 and the same variance (σ 2 )
at every value of X. Under these assumptions, Y ∼ N (β0 + β1 X, σ 2 ).
P 2
ei 3.12 + 2.42 + (−6.1)2 + (−.5)2 + 8.92 + (−12.6)2 + (4.8)2
11. (a) s =
2 = =
n−2 7−2
62.768.
(b) The sum of the residuals is 0. The residuals always sum to 0 in a least squares
regression.
15.4.2 Statistical Inference for the Parameter β1
12. (a) Never. We carry out hypothesis tests on parameters, not statistics.
(b) Rarely. This might be a point of interest for an investigator from time to time,
but not often.
Balka ISE 1.09 15.4. STATISTICAL INFERENCE IN SIMPLE LINEAR REGRESSION 228
(c) We test H0 : β1 = 0 very frequently in linear regression. It is often one of the

main points of interest in a study (since rejecting this null hypothesis indi-
cates strong evidence of a relationship between the explanatory and response
variables).
(d) Almost never. This hypothesis has no practical meaning in essentially all
situations.
σ2
13. β̂1 ∼ N (β1 , )
SSXX
14. (a) There are n − 2 = 22 − 2 = 20 degrees of freedom. With 20 degrees of freedom,
t.025 = 2.086. A 95% confdence interval for β1 is given by
β̂1 ± t.025 SE(β̂1 )
0.63554 ± 2.086 × 0.04847
0.63554 ± 0.1011
(b) We can be 95% confdent that β1 lies between 0.534 and 0.737, where β1 is the
change in the true mean abdomen length corresponding to a 1 mm increase in
snout-vent length.
β̂1 −0
(c) H0 : β1 = 0, Ha : β1 6= 0. t = = 0.63554−0
0.04847 = 13.11. (Note that the t
SE(β̂1 )
test statistic is given in the output.) The p-value is double the area to the
right of 13.11 under a t distribution with 20 degrees of freedom. The p-value is
2.79×10−11 (given in the output). (If this was not available in the output, and
we had to use a t table, we could say only that the p-value is very, very close to
0.) There is extremely strong evidence that the true slope β1 is greater than
0. This implies that as snout-vent length increases, abdomen length tends to
increase. (Which is hardly surprising.)
β̂0 −0
(d) H0 : β0 = 0, Ha : β0 =
6 0. t = = −5.73529−0
2.50608 = −2.289. (Note that the t
SE(β̂0 )
test statistic is given in the output.) The p-value is double the area to the left
of −2.289 under a t distribution with 20 degrees of freedom. The p-value is
0.0331 (given in the output). (If this was not available in the output, and we
had to use a t table, we could give only a range, such as 0.02 < p-value < 0.05.)
The p-value is less than the given signifcance level (p-value < 0.05), and so
there is statistically signifcant evidence that the true intercept di˙ers from 0.
(It appears as though the true intercept is in fact less than 0.)
(e) There are n − 2 = 22 − 2 = 20 degrees of freedom. With 20 degrees of freedom,
t.025 = 2.086. A 95% confdence interval for β0 is given by
β̂0 ± t.025 SE(β̂0 )
−5.73529 ± 2.086 × 2.50608
−5.73529 ± 5.2277
Balka ISE 1.09 15.5. CHECKING MODEL ASSUMPTIONS WITH RESIDUAL PLOTS 229
which works out to (−10.963, −0.507). Note that this 95% confdence interval
lies entirely to the left of 0, which should not be surprising given the results
of the hypothesis test.
15.5 Checking Model Assumptions with Residual Plots
15. Plot A shows curvature in the residuals (the residuals tend to be too small for
large and small values of X, and too large near the mean of X). Our simple linear
regression model is not adequate for this data.
Plot B shows that the variance of the residuals increases as X increases. This is
a problem for our model. (In these situations, the slope and intercept estimators
are still unbiased estimators of their respective parameters, but our standard errors
will be o˙.)
Plot C shows a random scattering of points, and no problems with the assumed
linear regression model.
Plot D shows a normal quantile-quantile plot in which the points lie very close to
a straight line. It indicates that the observed residuals are approximately normally
distributed. This indicates that the normality assumption is reasonable.
16. The plots look good, and do not indicate any problems with the assumed model.
The plot of the residuals vs snout-vent length shows a random scattering of points,
without any problems like curvature or increasing variance. The points on the
normal quantile-quantile plot fall roughly on a straight line. (There are 3 residuals
that are a little smaller than would be expected (very mild outliers), but they are
unlikely to have a major impact.) The assumptions of the model are reasonable,
and the inference procedures we carried out in Question 14 are valid.
15.6 The Correlation Coeÿcient and the Coeÿcient of

Determination
P
(Xi −X̄)(Yi −Ȳ )
17. (a) r = √ SPXY
SSXX SSY Y
= √P = √ 10.1
20.2·6.5
= 0.8814.
(Xi −X̄)2 (Yi −Ȳ )2
P
(b) No units. (The correlation coeÿcient r is a unitless quantity.)
18. (a) Yes, the correlation coeÿcient would be a very reasonable measure of the
strength of the relationship in each of the four plots, since all of the relation-
ships look roughly linear.
Balka ISE 1.09 15.7. ESTIMATION AND PREDICTION USING THE FITTED LINE 230
(b) From software, the calculated values of the correlation coeÿcients: Plot A:
r = 0.523, Plot B: r = −0.988, Plot C: r = −0.884, Plot D: r = 0.082.
Ranking in terms of increasing values of the correlation coeÿcient r: Plot B <
Plot C < Plot D < Plot A.
(c) Ranking in terms of increasing strength of the linear relationship:
Plot D < Plot A < Plot C < Plot B. (The greater the magnitude of r, the
stronger the linear relationship.)
(d) A change of the scaling of X and/or Y would not change the value of the cor-
relation coeÿcient r. Multiplying X or Y by a constant, or adding a constant,
does not change r.
19. (a) Yes, the coeÿcient of determination would be a very reasonable measure of
the strength of the relationship in each of the four plots, since all of the rela-
tionships look roughly linear.
(b) From software, the calculated values of the coeÿcient of determination: Plot
A: R2 = 0.274, Plot B: R2 = 0.977, Plot C: R2 = 0.781, Plot D: R2 = 0.007.
Ranking in terms of increasing values of the coeÿcient of determination:
Plot D < Plot A < Plot C < Plot B.
20. (a) The output shows that R2 = 0.8958.

(b) Approximately 90% of the variance in abdomen length can be explained by
the linear relationship √with snout-vent length.
√
(c) The magnitude of r is R2 = 0.8958 = 0.946. r always has the same sign as
the slope β̂1 , and since β̂1 is positive for this data, r = 0.946.
15.7 Estimation and Prediction Using the Fitted Line
21. (a) µ̂Y |X ∗ = −5.73529 + 0.63554 × 50.0 = 26.042 mm.

(b) Ŷ = −5.73529 + 0.63554 × 50.0 = 26.042 mm. (The predicted value is the
same as the estimated mean.)
(c) With 20 degrees of freedom, t.025 = 2.086. The 95% interval is:
µ̂Y |X ∗ ± tα/2 SE(µ̂Y |X ∗ )

26.042 ± 2.086 × 0.15447
26.042 ± 0.322
Which works out to (25.72, 26.36).

(d) The standard error of the estimated mean when X = 45 would be greater
than the standard error when X = 50, since 45 is further from the mean of X
than 50 is (X
¯ = 51.6.) Since the standard error is greater when X = 45, the
confdence interval would be wider.
Balka ISE 1.09 15.8. TRANSFORMATIONS 231
(e) A prediction interval for a single value is wider than the corresponding conf-
dence interval for the mean. (There is greater uncertainty when we are pre-
dicting a single value than when we are estimating the mean.)
15.8 Transformations
15.9 A Complete Example
15.10 Outliers and Infuential Points
22. (a) Points A and D have the greatest leverage, since they are furthest from the
mean of X. Points B and C have very little leverage, since they are near the
mean of X.
(b) Points B and D have the residuals with the greatest magnitude. (The vertical
distance from points B and D to the line is far greater than for the other two
points.)
(c) Point D would be the most infuential, since it has both high leverage and falls
far from the overall pattern of the observations. It would defnitely have the
most infuence of any point in this data set.
(d) If point D were removed, the slope would increase.
15.11 Some Cautions about Regression and Correlation
15.11.1 Always Plot Your Data
15.11.2 Avoid Extrapolating
23. (a) No, as this would be extrapolating far beyond the range of the observed data.
(b) No, the model is not fawed. We view β̂0 as the estimated mean of Y when
X = 0 only if X = 0 is within (or very near) the range of the observed data.
When X = 0 is far beyond the range of the observed data, we typically view
β̂0 simply as the Y -intercept, a necessary part of the line without any practical
meaning beyond that.
Balka ISE 1.09 15.12. A BRIEF MULTIPLE REGRESSION EXAMPLE 232
15.11.3 Correlation Does Not Imply Causation
15.12 A Brief Multiple Regression Example
24. (a) e1 = Y1 − Ŷ1 , where Y1 = 25, Ŷ1 = 1.20 + 3.14(8) = 26.32. e1 = 25 − 26.32 =
−1.32.
(b) 3.14 ± 2.306 × 0.3436, which works out to 3.14 ± 0.7923, or (2.35, 3.93).
(c) Yes, since the interval does not contain 0. Since the entire interval lies to the
right of 0, there is strong reason to believe β1 > 0, and there is an increasing
relationship of Y with X.
(d) Ŷ = 1.20 + 3.14(120) = 378. This prediction is an extrapolation, using the
regression line to estimate Y for an X value that is beyond the range of the
observed data. It is not a reasonable use of the regression line.
(e) Ŷ = 1.20 + 3.14(20) = 64. This is still an extrapolation, and it’s still not
reasonable.
25. (a) Ŷ = 4.0571 − 1.2429(1) = 2.8142.

(b) e1 = Y1 − Ŷ1 = 3.1 − 2.8142 = 0.2858.
(c) −1.2429±2.353×.2078, which works out to −1.2429±0.4890, or (−1.73, −0.75).
(d) If there is truly no linear relationship between X and Y , then β1 = 0. In this
event, β0 would equal the true mean of Y (E(Y )).
(e) −1.2429−2
.2078 = −15.60587. The two-sided p-value is double the area to the left of
−15.60587 under a t distribution with 5 − 2 = 3 degrees of freedom. This p-
value is very small, p-value < .001. There is very strong evidence that β1 < 2.
In other words, there is very strong evidence that the true relationship between
X and Y has a slope that is less than 2.
(f) The assumptions are that the error terms () are independent, and normally
distributed with a mean of 0 and common variance. The plot doesn’t show
any evidence against these assumptions, but with a sample size of n = 5 there
isn’t a lot to go on.
26. (a) Yes. The two-sided p-value of the test of H0 : β1 = 0 is .001.

(b) 1.218 ± 2.306 × 0.248, which works out to 1.218 ± .5719, or (.646, 1.790).
15.13.2 Concepts
27. The term extrapolating refers to using the regression line to predict or estimate
values of Y for values of X that are beyond the range of the observed data. This
should be avoided.
28. (a) Plot 1 shows a strong relationship between X and Y , but it is clearly not a
straight line relationship. Plots 2 and 3 look reasonable.
(b) Plot 1 has the strongest relationship.
(c) Plot 2 shows the strongest linear relationship.
(d) In decreasing order of the strength of the linear relationship: Plot 2 > Plot
3 > Plot 1. Since r2 measures the strength of the linear relationship, .9509
corresponds to Plot 2, .5935 corresponds to Plot 3, and .04392 corresponds to
Plot 1.
29. No. If r is very close to 0, then the sample shows no linear relationship between
X and Y . There could still be a strong relationship between X and Y , but not a
simple straight line relationship. See, for example, the relationship in Figure 15.1.
Figure 15.1: A curvilinear relationship (r is close to 0).
And if r is close to 0, there is always the chance that there is truly a strong straight
line straight line relationship between X and Y , and just we happened to see a
value of r close to 0.
30. An r2 value of .95 implies a very strong linear relationship between X and Y . For
example, Figure 15.2 shows a relationship with r2 = 0.81. But even though the
relationship is strong, there is not necessarily a causal link.
31. (a) Ŷ = −1.0 + 5.1X.

(b) Ŷ = −1.0 + 5.1(2.5) = 11.75.
(c) 5.10 ± 4.303 × .7746, which works out to 5.10 ± 3.333, or (1.77, 8.43).
(d) t = 5.10−0
.7746 = 6.584.
(e) .02 < p-value < .05.
(f) The four predicted values are Ŷ = −1.0+5.1(1) = 4.1, Ŷ = −1.0+5.1(2) = 9.2,
Figure 15.2: A relationship with r2 = 0.81.
Yˆ = −1.0 + 5.1(3) = 14.3, Yˆ = −1.0 + 5.1(4) = 19.4. The four residuals are
4.1 − 4.1 = 0, 8.2 − 9.2 = −1.0, 16.3 − 14.3 = 2.0, 18.4 − 19.4 = −1.0. This
leaves a sample variance about the regression line of
02 + (−1)2 + 2.02 + (−1.0)2 6

s2 = = =3
4−2 2
e2
P
32. s2 = n−2i is the best estimate of σ 2 . We aren’t given the fourth residual, but we
do know that the residuals must sum to 0 ( e2i = 0). Since the frst 3 resid-
P
uals sum to 4.7, this implies the fourth residual must have been −4.7. s2 =
2.02 +3.42 (−0.7)2 +(−4.7)2
4−2 = 19.07.
33. (a) In reality there is absolutely no relationship between X and Y in this situation
(they are randomly generated values, randomly paired together), so β1 = 0.
The null hypothesis and the assumptions of the t test are true, and so the test
statistic will have a t distribution with n − 2 = 10 − 2 = 8 degrees of freedom.
(b) 0, since that is the mean of a t distribution with 8 degrees of freedom.
(c) In the event the null hypothesis is true (and the assumptions are true), the
p-value has a uniform distribution between 0 and 1. so here the p-value would
have a uniform distribution between 0 and 1.
(d) 0.5 (the mean of a uniform distribution between 0 and 1).
34. The test of H0 : β1 = 0 (the null hypothesis of no linear relationship between X and
Y ).
35. Since 0 lies outside of the 95% confdence interval for β1 , the two-sided p-value of
the test of H0 : β1 = 0 is less than .05.
36. (a) The method used to fnd the sample slope and intercept is called the method
of least cubes. False. We use the method of least squares.
(b) The residuals in a simple linear regression always sum to 0. True.
(c) If X¯ = 0, then β̂1 = 0. False.
(d) If Y = 0, then β̂0 = 0. False.
¯
(e) r has the same sign as β̂1 . True. The sign of r is always the same as the sign
of β̂1 .
37. (a) β̂0 has the same sign as β̂1 . False. They can be equal, but they certainly don’t
have to be.
(b) If all data points fall perfectly on a line, then r = 1 or −1. True.
(c) The least squares regression line always passes through the point (0, 0). False.
The line will only pass through the origin if β̂0 = 0.
(d) The least squares regression line always passes through the point (X ¯ , Y¯ ).
True. (To show this, try putting X ¯ and the formula for β̂0 into the least
squares line).
(e) If there is no linear relationship between Y and X, then r = 1. False. We
can’t have a stronger sample relationship than r = 1 (all points fall perfectly
on a line when r = 1).
38. (a) Tests and confdence intervals on βˆ0 are the most common inference proce-
dures in regression. False, for a couple of reasons. First, we do not carry
out inference procedures for the statistic β̂0 . We may wish to carry out infer-
ence procedures for the parameter β0 , but these procedures are not usually of
practical interest. Inference for β1 is much more common.
(b) If the sample shows no linear relationship between Y and X (a random scat-
tering of points, say), then r and r2 will be close to 0. True.
(c) i = Yi −(β̂0 +β̂1 Xi ). False. ei = Yi −(β̂0 +β̂1 Xi ), not i . i = Yi −(β0 +β1 Xi )
(d) The residuals (the ei values) are uncorrelated. False. We assume that the true
error terms (the values of ) are uncorrelated, but the observed residuals are
correlated.
(e) The sample correlation coeÿcient is the proportion of the variance in Y that
is attributable to the linear relationship with X. False. This is the meaning
of the coeÿcient of determination.
39. (a) Instead of least squares, it would be better to choose a regression line based
on minimizing the sum of the absolute value of the residuals, but statisticians
have not yet fgured out how to do so. False. We can do this, but the least
squares method has some nice mathematical properties.
(b) The estimated standard deviation in a simple linear regression is the sum of the
squared residuals, divided by the sample size. False. s2 = Sum of squared
n−2
residuals
.
(c) If the test of the null hypothesis β1 = 0 results in a very small p-value, then r2
must have been large. False. Statistical signifcance is largely tied to sample
size. A small R2 value can still result in statistical signifcance, if the sample
size is large enough.
(d) If s = 0 and β̂1 > 0, then r = 1. True. s = 0 implies that all points fall
perfectly on a line (there is no variability about the regression line), and r has
the same sign as the sample slope β̂1 .
15.13.3 Applications
40. (a) Overall the simple linear regression model looks to be reasonable here. A
straight line ft looks very reasonable (there is no indication of curvature).
There does appear to be some increasing variance. (As head width increases,
the variance of the residuals increases. We often see this with count data,
as the variance often increases with the mean.) We do have more advanced
statistical methods that deal with this type of data, but overall the simple
linear regression model is not bad.
(b) No, the assumptions of the simple linear regression model are not perfectly true
in this case. The model assumes that for a given X, Y is normally distributed
about the true line. That cannot possibly be true here, as Y represents a
discrete count and thus cannot be perfectly normally distributed. But we can
still use the model. It may not be perfect, but it may very well provide a very
reasonable approximation.
(c) Ŷ = −29.400 + 17.004X, where Ŷ represents the predicted number of larvae
and X represents head width.
(d) A 1 mm increase in head width (X) results in an increase in the estimated
mean number of larvae (Y ) of 17.004.
(e) It is estimated that the larger blowfies carry 0.2 × 17.004 = 3.4 more larvae
on average.
(f) Yes, there is extremely strong evidence of a relationship between female head
width and the number of live larvae that they carry. The p-value of the test
of H0 : β1 = 0 is 9.01 × 10−14 . This is a minuscule p-value, and there is almost
surely a relationship between these variables.
(g) Ŷ = −29.400 + 17.004 × 4.0 = 38.616. (If we were using the model to predict
the count for a single female with a head width of 4.0 mm, our best guess
would be 39 larvae.)
(h) 0.6969 (the value of R2 given in the output).
41. (a) H0 : β1 = 0 (there is no linear relationship between hand grip strength and
facial attractiveness).
Ha : β1 =
6 0 (there is a linear relationship between hand grip strength and facial
attractiveness).
t = β̂1 −0 = 0.01309
0.01845 = 0.709 (note that this t statistic is given in the output).
SE(β̂1 )
n − 2 = 32 − 2 = 30 degrees of freedom. The p-value of 0.4836 is given in the
output. This is a large p-value, giving no evidence against the null hypothesis.
This sample gives no evidence of a relationship between hand grip strength
and facial attractiveness.
(Side note: The authors of the original paper found that after using more
advanced statistical methods to control for other important variables (e.g.
weight), there was evidence of a relationship.)

(b) 1.65%. (The value of R2 is given in the output as 0.0165.)
42. (a) Although the sample size is small and there is nothing defnitive, it appears as
though the relationship between GE and cadence may be more of a curvilinear
relationship. The residuals corresponding to both the small and large values of
cadence are large and positive, whereas the residuals corresponding to values of
cadence near the mean are negative. (The model may be underestimating GE
for large and small cadence values, and overestimating GE for values of cadence
near the mean.) The simple linear regression model may not be adequate.
One option for us would be to ft a curve through the points using multiple
regression techniques. (We crazy statisticians can do that!) For this example,
if we ft a curve by including an X 2 term in our model, we would get the
curvilinear relationship shown in Figure 15.3a. The residuals for this model
are shown in Figure15.3b. This model seems to ft the data much better.
26
1.0
24
0.5
Residuals
GE (%)
22
20 0.0
18
-0.5
16
80 85 90 95 100 80 85 90 95 100
Cadence (rpm) Cadence (rpm)
(a) Scatterplot with a ftted curve. (b) Residuals for the ftted curve.
Figure 15.3: Scatterplot of GE vs cadence and plot of the residuals for the ftted curve.
(b) Yes, there is very strong evidence of a relationship between cadence and GE.
The p-value of the test of H0 : β1 = 0 is given in the R output as 0.00130.
There is strong evidence that as cadence increases, GE tends to decrease.
43. (a) H0 : β0 = 0, Ha : β0 6= 0. This test does not have much practical meaning.
This tests the hypothesis that the mean memory score is 0 when the right
hippocampus has a volume of 0. This doesn’t have any practical meaning.
(b) H0 : β1 = 0, Ha : β1 =
6 0. Yes, this is meaningful. It is a test of the null hypoth-
esis that there is no linear relationship between right hippocampus volume and
score on the memory test.
(c) There is very strong evidence (two-sided p-value = 0.0017) of an increasing
relationship between right hippocampus volume and score on the Wechsler
memory scale.
(d) 0.08835 ± 2.086 × 0.02440, which works out to 0.08835 ± 0.0509 or (0.037,
0.139).
(e) e1 = Y1 − Ŷ1 . Y1 = 28. Ŷ1 = −41.72277 + 0.08835 × 955 = 42.65148.
e1 = 28 − 42.65148 = −14.65.
(f) 0.396 (The value of R2 √ given in the output.)
√
(g) The magnitude of r is r2 = 0.396 = 0.629. And since r has the same sign
as the slope, r = 0.629.
44. (a) Ŷ = 0.13215 + 0.362 × 44 = 16.06015.
(b) That is a test on the slope: H0 : β1 = 0, Ha : β1 =

6 0. A p-value of 7.59 × 10−11
gives extremely strong evidence against the null hypothesis (we would see a
relationship this strong less than 1 time in 100 billion if there is truly no
relationship). There is extremely strong evidence that as the insulation level
rises, the temperature rise above ambient tends to increase.
(c) H0 : β1 = 0.30, Ha : β1 =
6 0.30.
t = 0.3620−0.30
0.03211 = 1.931. The p-value is double the area to the right of 1.931
under a t distribution with 25 − 2 = 23 degrees of freedom. By computer:
p-value = 0.066. By the table: 0.05 < p-value < 0.10. There is not signifcant
evidence at α = 0.05 that the true slope di˙ers from 0.30.
SSXY 1100
45. (a) β̂1 = = ≈ 1.1996.
SSXXq 917 q
2
(b) SE(β̂1 ) = SSsXX = 24.935
917 ≈ 0.1649.
(c) 1.1996±2.306×0.1649, which works out to 1.1996±0.3803, or (0.8193, 1.5799).
(d) Yes. The entire interval lies well to the right of 0.
(e) H0 : β1 = 1, H0 : β1 =
6 1. t = 1.1996−1
.1649 = 1.2104. For a two-sided alternative,
0.20 < p-value < 0.30. We have no evidence that β1 di˙ers from 1.
(f) Ŷ = 161.6358 + 1.1996(92) = 271.999.
(g) s2 = 24.935.
2
(h) Under our usual assumptions, β̂1 ∼ N (β1 , SSσXX )
(i) Ŷ = 161.6358 + 1.1996(−2) = 159.2366. This is a nonsensical estimate, as it
estimates the height of a tree that is −2 months old.

JB Ies 109 Exercises Answers

Uploaded by

Copyright:

Available Formats

JB Ies 109 Exercises Answers

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

JB Ies 109 Exercises Answers

Uploaded by

Copyright:

Available Formats

Introductory Statistics Explained (1.

©2015 Jeremy Balka

4.3 Rules of Probability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

5 Discrete Random Variables and Discrete Probability Distributions 51

6 Continuous Random Variables and Continuous Probability Distribu-

6.5 Normal Quantile-Quantile Plots . . . . . . . . . . . . . . . . . . . . . . . . 82

8 Confdence Intervals 105

9 Hypothesis Testing 123

9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123

10 Inference for Two Means 153

10.7.4 Extra Practice Questions . . . . . . . . . . . . . . . . . . . . . . . 168

11 Inference for Proportions 175

12 Inference for Variances 193

13 χ2 Tests for Count Data 195

14 One-Way ANOVA 209

14.8 Basic Calculations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215

15 Introduction to Simple Linear Regression 225

2.2 Populations and Samples, Parameters and Statistics

1. A parameter is a numerical characteristic of a population, whereas a statistic is a

2. (a) All three-year-old walleye in this lake.

2.3 Types of Sampling

2.3.1 Simple Random Sampling

3. (a) The 200 pacemakers in the shipment.

(b) The 8 pacemakers.

2.3.2 Other Types of Random Sampling

2.4 Experiments and Observational Studies

2.5 Chapter Exercises

9. (a) All students at this university.

12. (a) The weight of the tumour is the response variable.

The sample in a) has many sources of potential bias, including:

The sample in b) has many sources of potential bias, including:

• Only students in line for student loans are in the sample.

15. (a) All full-time female graduate students at the university.

16. (a) All black bear gall bladders in Ontario.

19. (a) All third graders in the school.

24. (a) Task completion time.

28. (a) Reduction in blood pressure.

30. (a) The bonding agents.

31. (a) An observational study (a study of coroner records, for example).

3.2 Plots for Qualitative and Quantitative Variables

3.2.1 Plots for Qualitative Variables

Figure 3.1: Relationship to the victim for male murderers in Finland.

3.2.2 Plots for Quantitative Variables

3.3 Numerical Measures

3.3.1 Summation Notation

2. (a) 22 (the third number in the list).

3.3.2 Measures of Central Tendency

3.3.2.1 Mean, Median, and Mode

3.3.2.2 Other Measures of Central Tendency

3.3.3 Measures of Variability

7. (a) Range = P Maximum − Minimum = 3620 − 2620 = 1000.

(e) grams2 (the square of the units of the variable).

8. (a) The 4 values must be: 0, 0, 500, 500.

3.3.3.1 Interpreting the standard deviation

3.3.3.2 Why divide by n − 1 in the sample variance formula?

14. Dividing by n − 1 results in an estimator that, on average, equals the population

3.3.4 Measures of Relative Standing

16. (a) z1 = x1s−x̄ = 2870−3057.5

17. (a) True.