Module 6 - 2: Comparing Two Populations Means using Criterion: When should we make the Equal Variance Assumption,

Independent Samples that is, when should we assume equal variances/standard deviations?

Given two (quantitative) populations, we often want to compare the pop-

ulation means by considering their difference. In this section, we will make ❼ If the larger of the two sample standard deviations is less than twice
inferences about the value of the parameter µ1 − µ2 based upon samples the smaller of the two sample standard deviations, that is,
which are independent of each other. larger s
<2
smaller s
Population Population Sample Sample Sample we will assume equal variances and use the pooled t-methods for mak-
Mean Standard Size Mean Standard ing inferences about µ1 − µ2. If
Deviation Deviation
larger s
Population 1 µ1 σ1 n1 ȳ1 s1 ≥2
smaller s
Population 2 µ2 σ2 n2 ȳ2 s2
we will not assume equal variances and use the nonpooled t-methods
for making inferences about µ1 − µ2.

Typically, both σ1 and σ2 are unknown (and will have to be estimated), ❼ Similar Spreads Condition: make side-by-side boxplots of the
however, in some situations, we may believe that they are equal. In this data for each sample and check that the IQRs are similar.
case, we can use pooling to estimate the common variance (and common
standard deviation).

Other Assumptions/Conditions:
We will consider two t-methods for making inferences about µ1 − µ2 when
❼ We have a random (representative) sample of independent observations
we have two independent samples:
from each population.
❼ Nonpooled t-methods: σ12 and σ22 (or σ1 and σ2 ) are not assumed ❼ The two samples are independent of each other.
to be equal.
❼ We require either
❼ Pooled t-methods: assume σ12 = σ22 (or σ1 = σ2 ).
– both samples are large enough (both n1 ≥ 30 and n2 ≥ 30), or
– both populations are normally distributed.

Module 6-2 Page 1 of 15 Module 6-2 Page 2 of 15

 Two-Sample t-Interval for the Difference between Means Hypothesis Test for the Difference between Means:
Two-Sample t-Test
Recall: a confidence interval has the form
point estimate ± margin of error
Two-Sample Nonpooled t-Test:


= point estimate ± critical value × standard error of the estimate
A hypothesis test for µ1 − µ2 using the t-method without pooling has five
steps:

When the relevant assumptions are made and conditions are met: 1. Assumptions/Conditions:
(a) Nonpooled t-interval: if we do not assume equal variances, then ❼ σ1 and σ2 are both unknown and are not assumed to be equal.
a 100(1 − α)% confidence interval for µ1 − µ2 is: larger s
– We have ≥ 2.
r smaller s
s21 s22 ❼ We have a random sample of independent observations from each
(ȳ1 − ȳ2) ± t∗df +
n1 n2 population.
❼ The two samples are independent of each other.
where t∗df is the critical value corresponding to the 100(1 − α)% con-
fidence level based on df = Min(n1 − 1, n2 − 1) (or the complicated ❼ We require either
formula in the Module 6-2 slides). – both n1 ≥ 30 and n2 ≥ 30, or
– if one or both of n1, n2 is less than 30, then both population
(b) Pooled t-interval: if we assume equal variances and so also σ1 = σ2, distributions are normal.
then a 100(1 − α)% confidence interval for µ1 − µ2 is:
r
1 1
(ȳ1 − ȳ2) ± t∗df sp + 2. Hypotheses:
n1 n2

where sp is the pooled sample standard deviation H0 : µ1 − µ2 = ∆0

r
(n1 − 1)s21 + (n2 − 1)s22 µ1 − µ 2 =
6 ∆0 (two-tailed test)
sp =
n1 + n2 − 2 H A : µ1 − µ 2 < ∆0 (lower-tailed test)
∗
and tdf is the critical value corresponding to the 100(1−α)% confidence µ1 − µ 2 > ∆0 (upper-tailed test)
level based on df = n1 + n2 − 2.

Module 6-2 Page 3 of 15 Module 6-2 Page 4 of 15

 3. Test Statistic: Two-Sample Pooled t-Test:

Estimate − H0 value ȳ1 − ȳ2 − ∆0 A hypothesis test for µ1 −µ2 using the t-method with pooling has five steps:
t0 = = r
SE(Estimate) s21 s22
+
n1 n2
1. Assumptions/Conditions:
❼ σ1 and σ2 are both unknown, but we assume σ1 = σ2 .
where t0 follows a t-distribution with df = Min(n1 − 1, n2 − 1) (or the
larger s
complicated formula on page 8 of the Module 6-2 slides) under the null – We have <2
smaller s
hypothesis.
❼ We have a random sample of independent observations from each
population.
4. P -value: Compute one of the following using software or the t-table, ❼ The two samples are independent of each other.
where df = Min(n1 − 1, n2 − 1) (or the complicated formula in the 6-2
❼ We require either
slides).
– both n1 ≥ 30 and n2 ≥ 30, or
Test P -value – if one or both of n1, n2 is less than 30, then both population
Two-tailed Test 2P (tdf > |t0|) distributions are normal.
Lower-tailed Test P (tdf < t0)
Upper-tailed Test P (tdf > t0)

If using a t-table, we will likely only be able to find an interval in which 2. Hypotheses:
the P-value lies.
H0 : µ1 − µ2 = ∆0

5. Conclusion: Report and interpret the P -value in context. Given a

significance level α, µ1 − µ 2 =
6 ∆0 (two-tailed test)
H A : µ1 − µ 2 < ∆0 (lower-tailed test)
❼ if P - value ≤ α, we reject H0 at level α
µ1 − µ 2 > ∆0 (upper-tailed test)
❼ if P - value > α, we do not reject H0 at level α

Module 6-2 Page 5 of 15 Module 6-2 Page 6 of 15

 3. Test Statistic: Example: Surgeons wanted to determine if a new surgical method re-
duced the operative time relative to the standard method for a common
Estimate − H0 value ȳ1 − ȳ2 − ∆0 surgical procedure. To conduct a randomized comparative experiment for
t0 = = r
SE(Estimate) 1 1 the two surgical methods, they randomly selected 63 patients requiring this
sp +
n1 n2 surgery and then randomly assigned 31 patients to the new method and 32
patients to the standard method. The operative times were recorded (in
minutes). The summary statistics are given below:
where r
(n1 − 1)s21 + (n2 − 1)s22 New Method Standard Method
sp = (pop 1) (pop 2)
n1 + n2 − 2
Sample Size 31 32
Sample Mean 394.64 468.33
is the pooled sample standard deviation and where t0 follows a Sample Standard Deviation 84.75 38.17
t-distribution with df = n1 + n2 − 2 under the null hypothesis.
(a) At the 5% significance level, do the data provide sufficient evidence to
conclude that the mean operative time is less with the new method
4. P -value: Compute one of the following using software or the t-table,
than with the standard method?
where df = n1 + n2 − 2.
1. Assumptions/Conditions:
Test P -value
Two-tailed Test 2P (tdf > |t0|)
Lower-tailed Test P (tdf < t0)
Upper-tailed Test P (tdf > t0)

If using a t-table, we will likely only be able to find an interval in which

the P-value lies.

5. Conclusion: Report and interpret the P -value in context. Given a

significance level α,
❼ if P - value ≤ α, we reject H0 at level α
❼ if P - value > α, we do not reject H0 at level α

Module 6-2 Page 7 of 15 Module 6-2 Page 8 of 15

 2. Hypotheses: 4. P -value:

3. Test Statistic:

5. Conclusion:

Since , we at
the 0.05 significance level, that is, there
statistical evidence to conclude that the mean operative time is less
with the new method than with the standard method.

Module 6-2 Page 9 of 15 Module 6-2 Page 10 of 15

(b) Construct a 90% confidence interval for the difference µ1 − µ2 in the Example: Researchers wanted to compare the sense of direction of adult
mean operative times between the two surgical methods. males and adult females, in particular, their frame of reference. They
obtained a random sample of 30 adult males and a random sample of 30
adult females, and then took them to an unfamiliar heavily wooded park.
One at a time, each person was separated from the group, given a pointer
attached to a large 360◦ protractor, and was asked to point south. The
number degrees that each person was off from south was recorded. The
summary statistics are given below:
Males Females
(pop 1) (pop 2)
Sample Size 30 30
Sample Mean 37.6 55.8
Sample Standard Deviation 38.5 48.3
(a) At the 1% significance level, do the data provide enough statistical
evidence to conclude that, on average, males have a better sense of
direction and, in particular, a better frame of reference than females?

1. Assumptions/Conditions:

Module 6-2 Page 11 of 15 Module 6-2 Page 12 of 15

 2. Hypotheses: 4. P -value:

3. Test Statistic:

5. Conclusion:

Since , we at
the 0.01 significance level, that is, there
statistical evidence to conclude that, on average, males have a better
sense of direction and, in particular, a better frame of reference than
females.

Module 6-2 Page 13 of 15 Module 6-2 Page 14 of 15

(b) Construct a 98% confidence interval for the difference µ1 − µ2 in the
mean number of degrees off from south between adult males and adult
females.

Module 6-2 Page 15 of 15