Section 4.

Inferences About Variances (ATTENDANCE 7) 127

4.4 Inferences About Variances

σ2
We look at hypothesis tests and confidence intervals for σ 2 and σ12 . From a “big
2
picture” point of view, we look at the (large n) one–sample and two–sample variance
(standard deviation) problems.

mean variance proportion

µ σ2 π
one large n, 3.7, 3.8, 3.9, 3.10, 4.6 4.4 6.2
small n, 4.3, 4.6
sample two large n, 3.11 4.4 6.3
small n, 4.3
multiple chapters not 6.2, 6.3
7, 8, 9 done

Exercise 4.6 (Test and Confidence Interval for Variance (or Standard De-
viation), σ) We first look at examples of the hypothesis test and confidence interval
when S 2 is calculated from a random sample of size n from a normal population with
variance σ 2 . The test statistic is
(n − 1)S 2
χ2 =
σ02

with critical value χ2c where,

H0 H1 reject H0 at level α if
σ 2 ≤ σ02 σ 2 > σ02 χ2c > χ2 (n − 1, α)
σ 2 ≥ σ02 σ 2 < σ02 χ2c < χ2 (n − 1, 1 − α)
σ 2 = σ02 σ 2 6= σ02 χc > χ (n − 1, α/2) or χ2c < χ2 (n − 1, 1 − α/2)
2 2

The confidence intervals are:

(n−1)S 2 (n−1)S 2
³ ´
• (two–sided) confidence interval, CI: χ2 (n−1,α/2)
, χ2 (n−1,1−α/2)

(n−1)S 2
³ ´
• lower confidence interval, LCI: χ2 (n−1,α)
, ∞
(n−1)S 2
³ ´
• upper confidence interval, UCI: −∞, χ2 (n−1,1−α)

See Lab 7: Test For Variance.

1. Hypothesis Test: Carbon Content. In a random sample of 28 soil samples, the

variance in carbon content was found to be s2 = 0.7 parts per million (ppm).
Test if the variance is greater than 0.4 at α = 0.05.
128Chapter 4. Inferences About One Or Two Populations: Interval Data (ATTENDANCE 7)

(a) Test Statistic Versus Critical Value, Standardized.

i. Statement. The statement of the test is (circle one)
A. H0 : σ 2 = 0.4 versus H1 : σ 2 > 0.4
B. H0 : σ 2 = 0.4 versus H1 : σ 2 < 0.4
C. H0 : σ 2 = 0.4 versus H1 : σ 2 6= 0.4
ii. Test. The standardized test statistic of s = 0.7 is
(n − 1)s2 (28 − 1)(0.7)
χ2 test statistic = 2
= =
σ 0.4

(circle one) 30.2 / 47.3 / 82.7.

The standardized critical value at 0.05 is
χ2 (n − 1, α) = χ2 (28 − 1, 0.05) = (circle one) 39.2 / 40.1 / 43.2
(Use PRGM INVCHI2 ENTER 27 ENTER 0.95 ENTER)
iii. Conclusion. Since the test statistic, 47.3, is larger than the critical
value, 40.1, we (circle one) accept / reject the null hypothesis that
σ = 0.4.
(b) P–Value Versus Level of Significance, Standardized.
i. Statement. The statement of the test is (circle one)
A. H0 : σ 2 = 0.4 versus H1 : σ 2 > 0.4
B. H0 : σ 2 = 0.4 versus H1 : σ 2 < 0.4
C. H0 : σ 2 = 0.4 versus H1 : σ 2 6= 0.4
ii. Test. Since the standardized test statistic is χ2 = 47.3, the p–value,
at n − 1 = 28 − 1 = 27 df, is given by

p–value = P (χ2 ≥ 47.3)

which equals (circle one) 0.01 / 0.05 / 0.10.

(Use 2nd DISTR 2:χ2 cdf(47.3,E99,27).)
The level of significance is 0.05.
iii. Conclusion. Since the p–value, 0.01, is smaller than the level of signif-
icance, 0.05, we (circle one) accept / reject the null hypothesis that
σ = 0.4.
(c) Related Questions.
i. Since the χ2 test statistic is (circle one)
A. larger than the critical value, χ2 (27, 0.05), we will accept that the
variance in carbon content exceeds 0.4 mm.
B. larger than the level of significance, α = 0.05, we will accept that
the variance in carbon content exceeds 0.4 mm.
Section 4. Inferences About Variances (ATTENDANCE 7) 129

C. smaller than the critical value, χ2 (27, 0.05), we will accept that
the variance in carbon content exceeds 0.4 mm.
D. larger than the p–value we will accept that the variance in carbon
content exceeds 0.4 mm.
ii. Since the p–value is (circle one)
A. larger than the critical value, χ2 (27, 0.05), we will accept that the
variance in carbon content exceeds 0.4 mm.
B. smaller than the level of significance, α = 0.05, we will accept that
the variance in carbon content exceeds 0.4 mm.
C. smaller than the critical value, χ2 (27, 0.05), we will accept that
the variance in carbon content exceeds 0.4 mm.
D. larger than the test statistic we will reject that the variance in
carbon content is equal to 0.4 mm.
iii. Population, Parameter, Sample and Statistic. Match the statistical
items with the appropriate parts of this soil example.

terms carbon example

(i) population (i) variance in carbon content, of 28 samples, s2
(ii) sample (ii) variance in carbon content, of all samples, σ 2
(iii) statistic (iii) carbon content, of all samples
(iv) parameter (iv) carbon content, of 28 samples
terms (i) (ii) (iii) (iv)
carbon example
2. Hypothesis Test: Fish Lengths. In a random sample of 18 fish lengths, the
standard deviation in lengths was found to be s = 12 cm. Test if the standard
deviation is less than 13 cm at α = 0.05.
(a) Test Statistic Versus Critical Value, Standardized.
i. Statement. The statement of the test is (circle one)
A. H0 : σ = 13 versus H1 : σ > 13
B. H0 : σ = 13 versus H1 : σ < 13
C. H0 : σ = 13 versus H1 : σ 6= 13
ii. Test. The standardized test statistic of s = 12 is
(n − 1)s2 (18 − 1)(12)2
χ2 test statistic = = =
σ2 (13)2

(circle one) 14.5 / 60.1 / 82.7.

The standardized critical value at α = 0.05 is
χ2 (n − 1, 1 − α) = χ2 (18 − 1, 0.95) = (circle one) 8.7 / 40.1 / 43.2
(Use PRGM INVCHI2 ENTER 17 ENTER 0.05 ENTER)
130Chapter 4. Inferences About One Or Two Populations: Interval Data (ATTENDANCE 7)

iii. Conclusion. Since the test statistic, 14.5, is larger than the critical
value, 8.7, we (circle one) accept / reject the null hypothesis that
σ = 13.
(b) P–Value Versus Level of Significance, Standardized.
i. Statement. The statement of the test is (circle one)
A. H0 : σ = 13 versus H1 : σ > 13
B. H0 : σ = 13 versus H1 : σ < 13
C. H0 : σ = 13 versus H1 : σ 6= 13
ii. Test. Since the standardized test statistic is χ2 = 14.5, the p–value,
at n − 1 = 18 − 1 = 17 df, is given by

p–value = P (χ2 ≤ 14.5)

which equals (circle one) 0.00 / 0.05 / 0.37.

(Use 2nd DISTR 2:χ2 cdf(−E99,14.5,17).)
The level of significance is 0.05.
iii. Conclusion. Since the p–value, 0.37, is larger than the level of signifi-
cance, 0.05, we (circle one) accept / reject the null hypothesis that
σ = 13.
3. Confidence Interval: Car Door and Jamb. In a random sample of 28 cars, the
variance in the distance between door and jamb was found to be s2 = 0.7. Let
σ 2 = 0.4. Calculate a 95% CI using the formula,
(n − 1)S 2 (n − 1)S 2
à !
,
χ2 (n − 1, α/2) χ2 (n − 1, 1 − α/2)

(a) The critical value χ2 (n − 1, α/2) = χ2 (28 − 1, 0.05/2) =

(circle one) 8.7 / 40.1 / 43.2
(Use PRGM INVCHI2 ENTER 27 ENTER 0.975 ENTER)
(b) The critical value χ2 (n − 1, 1 − α/2) = χ2 (28 − 1, 1 − 0.05/2) = is
(circle one) 14.6 / 40.1 / 43.2
(Use PRGM INVCHI2 ENTER 27 ENTER 0.025 ENTER)
(c) So,
³ the CI is
(n−1)s2 (n−1)s2
´ ³ ´
(28−1)0.7 (28−1)0.7
2
χ (n−1,α/2)
, 2
χ (n−1,1−α/2)
= 43.2
, 14.6
=
(circle one) (0.6, 1.6) / (0.5, 1.2) / (0.44, 1.29).
(d) Since the 95% CI (0.44, 1.29) excludes 0.4, this indicates that the variance
in the distance between door and jamb (circle one) is / is not 0.4 mm.
4. Confidence Interval: Machine Parts. In a random sample of 18 machine parts,
the variance in lengths was found to be s2 = 144. Let σ = 92. Calculate a 90%
CI.
Section 4. Inferences About Variances (ATTENDANCE 7) 131

(a) The critical value χ2 (n − 1, α/2) = χ2 (28 − 1, 0.10/2) =

(circle one) 8.7 / 27.6 / 43.2
(Use PRGM INVCHI2 ENTER 17 ENTER 0.95 ENTER)
(b) The critical value χ2 (n − 1, 1 − α/2) = χ2 (28 − 1, 1 − 0.10/2) = is
(circle one) 8.7 / 40.1 / 43.2
(Use PRGM INVCHI2 ENTER 17 ENTER 0.05 ENTER)
(c) So,
³ the CI is
(n−1)s2 (n−1)s2 (18−1)122 (18−1)122
´ ³ ´
2
χ (n−1,α/2)
, 2
χ (n−1,1−α/2)
= 27.6
, 8.7
=
(circle one) (70, 100) / (84.6, 291.4) / (88.6, 281.4).
(d) Since the 95% CI (88.6, 281.4) includes 92, this indicates that the variance
in lengths (circle one) is / is not 92 mm.

Exercise 4.7 (Hypothesis Test and Confidence Interval of σσ21 ) We now look at
examples of the hypothesis test and confidence interval when S12 and S22 are calculated
from two independent random samples of size n1 and n2 from two normal populations
with variances σ12 and σ22 . The test statistic is

S12
F =
S22
with critical value Fc where,

H0 H1 reject H0 at level α if
σ12 σ12
σ22
≤ δ0 σ22
> δ0 Fc > δ0 F (n1 − 1, n2 − 1, α)
σ12 σ12
σ22
≥ δ0 σ22
< δ0 Fc < δ0 F (n1 − 1, n2 − 1, 1 − α)
σ12 σ12
σ22
= δ0 σ22
6= δ0 Fc > δ0 F (n1 − 1, n2 − 1, α/2) or Fc < δ0 F (n1 − 1, n2 − 1, 1 − α/2)

where δ0 is a nonnegative number (usually δ0 = 1). The confidence intervals are:

S12 S12
³ ´
1 1
• (two–sided) confidence interval, CI: S22 F (n1 −1,n2 −1,α/2)
, S22 F (n1 −1,n2 −1,1−α/2)

S12
³ ´
1
• lower confidence interval, LCI: S22 F (n1 −1,n2 −1,α)
, ∞

S12
³ ´
1
• upper confidence interval, UCI: −∞, S22 F (n1 −1,n2 −1,1−α)

σ1
See Lab 7: Tests and CIs For σ2
.

1. Hypothesis Test: Plasma Levels Again. The average and standard deviation of
the plasma levels for males and females are given below.
132Chapter 4. Inferences About One Or Two Populations: Interval Data (ATTENDANCE 7)

males (1) females (2)

ȳ 3.259 1.413
s2 0.16 0.13
n 9 6

Test if σ12 > σ22 at α = 0.05.

(a) Test Statistic Versus Critical Value.
i. Statement. The statement of the test is (circle one)
A. H0 : σ12 = σ22 versus H1 : σ12 > σ22
B. H0 : σ12 = σ22 versus H1 : σ12 < σ22
C. H0 : σ12 = σ22 versus H1 : σ12 6= σ22
where, notice, if σ12 = σ22 , then
σ12
σ22
= δ0 = (circle one) −1 / 0 / 1
and if σ12 > σ22 , then
σ12
σ2
> δ0 = (circle one) −1 / 0 / 1.
2
ii. Test. The test statistic is
s21 0.16
F test statistic = 2
= =
s2 0.13

(circle one) 1.03 / 1.23 / 2.27.

The critical value at α = 0.05, with n1 − 1 = 9 − 1 = 8 and n2 − 1 =
6 − 1 = 5 degrees of freedom, is
F (n1 − 1, n2 − 1, α) = F (8, 5, 0.05) = (circle one) 3.22 / 4.15 / 4.82
(Use PRGM INVF ENTER 8 ENTER 5 ENTER 0.95 ENTER)
iii. Conclusion. Since the test statistic, 1.23, is less than the critical value,
4.82, we (circle one) accept / reject the null hypothesis that σ12 = σ22 .
(b) P–Value Versus Level of Significance.
i. Statement. The statement of the test is (circle one)
A. H0 : σ12 = σ22 versus H1 : σ12 > σ22
B. H0 : σ12 = σ22 versus H1 : σ12 < σ22
C. H0 : σ12 = σ22 versus H1 : σ12 6= σ22
ii. Test. Since the test statistic is F = 1.23, with n1 − 1 = 9 − 1 = 8 and
n2 − 1 = 6 − 1 = 5 degrees of freedom, the p–value is given by

p–value = P (F ≥ 1.23)

which equals (circle one) 0.14 / 0.35 / 0.43.

(Use 2nd DISTR 9:F cdf(1.23,E99,8,5).)
The level of significance is 0.05.
Section 4. Inferences About Variances (ATTENDANCE 7) 133

iii. Conclusion. Since the p–value, 0.43, is larger than the level of signifi-
cance, 0.05, we (circle one) accept / reject the null hypothesis that
σ12 = σ22 .
(c) Related Questions. Match the statistical terms with the appropriate parts
of the plasma example.
terms plasma example
(i) population (i) nine male, six female plasma levels
s2
(ii) sample (ii) F = s21
2
(iii) statistic (iii) all 17–year–olds plasma levels
σ2
(iv) parameter (iv) F = σ12
2

terms (i) (ii) (iii) (iv)

plasma example
2. Hypothesis Test: Plasma Levels Again. The average and standard deviation of
the plasma levels for males and females is given below.
males (1) females (2)
ȳ 2.159 1.713
s2 0.16 0.15
n 9 13

Test if σ12 6= σ22 at α = 0.05.

(a) Test Statistic Versus Critical Value.
i. Statement. The statement of the test is (circle one)
A. H0 : σ12 = σ22 versus H1 : σ12 > σ22
B. H0 : σ12 = σ22 versus H1 : σ12 < σ22
C. H0 : σ12 = σ22 versus H1 : σ12 6= σ22
ii. Test. The test statistic is
s2 0.16
F test statistic = 12 = =
s2 0.15

(circle one) 1.07 / 1.23 / 2.27.

Since this is two–sided, the critical value at α = 0.5
2
= 0.025, with
n1 − 1 = 9 − 1 = 8 and n2 − 1 = 13 − 1 = 12 degrees of freedom, is
F (n1 − 1, n2 − 1, α/2) = F (8, 12, 0.025) =
(circle one) 3.51 / 4.15 / 4.82
(Use PRGM INVF ENTER 8 ENTER 12 ENTER 0.975 ENTER)
and F (n1 − 1, n2 − 1, 1 − α/2) = F (8, 12, 0.975) =
(circle one) 0.148 / 0.238 / 2.825
(Use PRGM INVF ENTER 8 ENTER 12 ENTER 0.025 ENTER)
134Chapter 4. Inferences About One Or Two Populations: Interval Data (ATTENDANCE 7)

iii. Conclusion. Since the test statistic, 1.07, is in between the two crit-
ical values, (0.238, 3.51), we (circle one) accept / reject the null
hypothesis that σ12 = σ22 .
(b) P–Value Versus Level of Significance.
i. Statement. The statement of the test is (circle one)
A. H0 : σ12 = σ22 versus H1 : σ12 > σ22
B. H0 : σ12 = σ22 versus H1 : σ12 < σ22
C. H0 : σ12 = σ22 versus H1 : σ12 6= σ22
ii. Test. Since the test statistic is F = 1.07, the p–value is given by
p–value = 2 × P r(F ≥ 1.07)
which equals (circle one) 0.14 / 0.44 / 0.88.
(Use 2nd DISTR 9:F cdf(1.07,E99,8,12).)
The level of significance is 0.05.
iii. Conclusion. Since the p–value, 0.88, is larger than the level of signifi-
cance, 0.05, we (circle one) accept / reject the null hypothesis that
σ12 = σ22 .

4.5 Statistical Models

Not covered.

4.6 Predicting Future Observations

Exercise 4.8 (Prediction Interval For The Mean of m Future Observations,

Normal) Three prediction intervals for the average of m future unknown observations
are:
(two–sided) prediction
• µ r interval, PI: r³ ¶
³ ´ ´
1 1 1 1
Y − t(n − 1, α/2) m
+ n
σ2, Y + t(n − 1, α/2) m
+ n
σ2
µ r³ ´ ¶
1 1
• lower prediction bound, LPB: Y − t(n − 1, α) m
+ n
σ2, ∞
µ r³ ´ ¶
1 1
• upper prediction bound, UPB: −∞, Y + t(n − 1, α) m
+ n
σ2

If n is large (n > 30, say), and the prediction is about one individual, the t critical
values can be replaced by z critical values and the resulting prediction intervals are
said to be normal limits. From a “big picture” point of view, we look at more of the
details of the large n and small n one–sample mean problems.
Section 6. Predicting Future Observations (ATTENDANCE 7) 135

mean variance proportion

µ σ2 π
one large n, 3.7, 3.8, 3.9, 3.10, 4.6 4.4 6.2
small n, 4.3, 4.6
sample two large n, 3.11 4.4 6.3
small n, 4.3
multiple chapters not 6.2, 6.3
7, 8, 9 done
1. Average Touch–Sensitivity. In a random sample of 32 totally blind people taken
from a normally distributed population, the average touch–sensitivity is found
to be ȳ = 0.013 with a standard deviation of σ = 0.003.
(a) The 95% lower prediction bound, LPB, for the mean of m = 10 future
observations of touch–sensitivity
µ r³ ¶is given
µ by r³ ¶
´ ´
1 1 2 1 1 2
ȳ − t(n − 1, α) m + n σ , ∞ ≈ 0.013 − z(0.05) 10 + 32 0.003 , ∞ =
(circle one) (0.009, ∞) / (0.010, ∞) / (0.011, ∞).
(Notice that t(n − 1, α) is replaced by z(0.05) since n = 32 > 30 and, also,
for z(0.05), type 2nd DISTR 3:invNorm(0.95) ENTER.)
(b) The 93% UPB, for the mean of m = 15 future observations of touch–
sensitivity
µ is given
r by ¶ µ r ¶
³ ´ ³ ´
1 1 1 1
−∞, ȳ + z(α) m
+ n
σ 2 = −∞, 0.013 + z(0.07) 15
+ 32
0.0032 =
(circle one) (−∞, 0.012) / (−∞, 0.013) / (−∞, 0.014).
(2nd ENTRY will retrieve the last typed calculator command.)
(c) The 98% PI, for the mean of m = 12 future observation of touch–sensitivity
is given by r r
³ ´ ³ ´
1 1 1 1
ȳ ± z(α/2) m
+ n
σ 2 = 0.013 ± z(0.01) 12
+ 32
0.0032 =
(circle one) (0.009, 0.013) / (0.010, 0.014) / (0.011, 0.015).
(For z(0.02/2) = z(0.01), type 2nd DISTR 3:invNorm(0.99) ENTER.)
(d) (Review.) The 98% confidence interval (not a prediction interval and for
no future observations)
µ r of touch–sensitivity
r ¶
is given by
³ ´ ³ ´
1 1
ȳ + z(α/2) n
σ 2 , ȳ − z(α/2) n
σ2 =
µ r³ ´ r³ ´ ¶
1 1
0.013 + z(0.01) 32
0.0032 , 0.013 + z(0.01) 32
0.0032 =
(circle one) (0.009, 0.012) / (0.010, 0.013) / (0.012, 0.014).
(Type STAT TESTS ZInterval... Stats 0.013 0.003 32 0.98 Calculate.)
(e) True / False Prediction Intervals are always wider (larger, broader) than
confidence intervals because they estimate both the population average as
well as the uncertainty of future unknown observations, whereas confidence
intervals estimate just the population average.
136Chapter 4. Inferences About One Or Two Populations: Interval Data (ATTENDANCE 7)

2. Average Ph Levels in Soil. In a random sample of 28 soil samples taken from a

normally distributed population, the average Ph level is found to be ȳ = 10.55
with a standard deviation of σ = 3.01. The 95% prediction interval, PI, for
m
µ = 5 future observations
r of the Ph levels is givenrby ¶
³ ´ ³ ´
1 1 1 1
ȳ − t(n − 1, α/2) m
+ n
σ 2 , ȳ + t(n − 1, α/2) m
+ n
σ2 =
µ r³ ´ r³ ´ ¶
1 1 1 1
= 10.55 − t(27, 0.05/2) 5
+ 28
3.012 , 10.55 + t(27, 0.05/2) 5
+ 28
3.012 =
(circle one) (7.55, 13.54) / (8.146, 13.954) / (8.146, 14.954).
(For t(27, 0.025), type PRGM INVT 27 ENTER 0.975 ENTER.)
3. Log–Normal Distribution: Soil–Water Fluxes. Y is log–normal if ln Y is normal.
Consider the following sample of ten soil–water fluxes.
0.306, 0.363, 0.437, 0.787, 0.899, 1.272, 1.424, 1.634, 1.682, 5.128
Assume the soil–water flux has a log–normal distribution; in other words, the
natural logarithm (ln) of the soil–water flux has a normal distribution. Calculate
a 95% PI for this data.
(a) True / False Since the soil–water flux is nonnormal (in fact, it is log–
normal) we are not able to calculate a 95% PI for the data as given, since
one of the assumptions of calculating a PI is that the data follows from a
normal distribution.
(b) Use your calculator and transform the data by taking the ln of these ten
measurements (and so transform the data into that which follows from a
normal distribution). This gives (circle one)
i. 0.306, 0.363, 0.437, 0.787, 0.899, 1.272, 1.424, 1.634, 1.682, 5.128
ii. -1.185, -1.103, -0.829, -0.239, -0.106, 0.241, 0.354, 0.491, 0.520, 1.635
(Type STAT EDIT, then type ten fluxes into L1 and define L2 as ln(L1 ).)
(c) The average (mean) of the ten ln–fluxes is ȳ = −0.013 and the standard
deviation is
s = (circle one) 0.853 / 0.925 / 1.253.
(d) The 95% prediction interval, PI, for the mean of m = 5 future observations
of
µ the ln–fluxes is r given by r ¶
³ ´ ³ ´
1 1 1 1
ȳ − t(n − 1, α/2) m
+ n
σ 2 , ȳ + t(n − 1, α/2) m
+ n
σ2 =
µ r³ ´ r³ ´ ¶
1 1 1 1
= −0.013 − t(9, 0.05/2) 5
+ 10
0.8532 , −0.013 + t(9, 0.05/2) 5
+ 10
0.8532 =
(circle one) (−1.06, 1.04) / (−0.681, 0.855) / (−0.581, 0.955).
(e) The 95% prediction interval, PI, for m = 5 future observations of the
original soil–water fluxes (not the ln–fluxes!) is obtained by introducing e ·
in the following way (e−1.06 , e1.04 ) =
(circle one) (0.343, 2.840) / (0.458, 2.128) / (0.558, 2.228).
Section 7. The Assumptions for t, χ2 and F Test Procedures (ATTENDANCE 7)137

4.7 The Assumptions for t, χ2 and F Test Proce-

dures

Exercise 4.9 (Violating the Assumptions of the Statistical Model.) All sta-
tistical models thus far involve one or more of the following three assumptions (some-
times called the analysis of variance (ANOVA) assumptions):

1. The populations have normal distributions.

2. The population variances are equal but unknown.

3. The data are independent random samples from the populations.

1. Normality. A distribution is nonnormal if it is skewed. A distribution is also

nonnormal if it either heavy–tailed or light–tailed.

normal

heavy-tailed
light-tailed

Figure 4.1 (Light–Tailed and Heavy–Tailed Distributions)

Use your calculator to draw a box–plot of the 28 Ph levels of soil data:

4.3 5 5.9 6.5 7.6 7.7 7.7 8.2 8.3 9.5

10.4 10.4 10.5 10.8 11.5 12 12 12.3 12.6 12.6
13 13.1 13.2 13.5 13.6 14.1 14.1 15.1

The box–plot indicates the data is (circle one) skewed right / symmetric /
skewed left and (circle one) heavy–tailed / normal / light–tailed (where
heavy–tailed distributions are indicated by extreme outliers).

2. Normality Again. Log–normal data is an example where a nonnormal distri-

bution can be transformed to a normal distribution (and so allow statistical
inference which requires normality to proceed). It is (circle one) always /
sometimes possible to do this to nonnormal data.

3. Equal Variance. A violation of the assumption of equal variance most detri-

mentally effects calculation of tests and confidence intervals for (circle one)
138Chapter 4. Inferences About One Or Two Populations: Interval Data (ATTENDANCE 7)

(a) θ = µ, where σ̂θ̂ = √sn

and we assume normality and a small random sample
q
(b) θ = µ1 − µ2 , where σ̂θ̂ = Sp n11 + n12
and we assume normality, small independent random samples
and unknown σ12 = σ22
r
S2 S2
(c) θ = µ1 − µ2 , where σ̂θ̂ = n11 + n22
and we assume normality, small independent random samples
and unknown σ12 6= σ22
SD
(d) θ = µD = µ1 − µ2 , where σ̂θ̂ = √ n
and we assume normality and small dependent (paired) random samples

4. Independence. True / False Independence is violated if the sample has not been
collected properly. A systematic dependence of error occurs in time series.