Chapter 7: Comparison of Two Groups 49

Chapter 7: Comparison of Two Groups

7.1. These two are independent samples since the subjects in the two samples are different, with no matching
between one sample with the other sample.
7.2. (a) (i) These results would be based on independent samples if the respondents in the two samples were
different and no matching was done between the two samples.
(ii) These results would be based on dependent samples if the samples included either the same people
or matching was done between the two samples.
(b) The explanatory variable is year, and the response variable is the percent of adults from the developing
nations who expressed a positive attitude about adequacy of measures taken at the global level to protect
environment. The response variable will be categorical.
7.3. (a) pˆ 2017 - pˆ 2012 = 0.56 - 0.29 = 0.27, which is an estimated 27.0% decrease.

(b) The standard error of the difference is se = ( 0.040) 2 + ( 0.040) 2 = 0.057.

7.4. (a) pˆ2 - pˆ1 = 0.52 - 0.14 = 0.38

(b) se = ( 0.0157) 2 + ( 0.0110) 2 = 0.01917

7.5. (a) y2 - y1 = 65 - 52 = 13; se = ( 1.2) 2 + ( 1.2) 2 = 1.697; y2 - y1 = 10 does not vary much from m2 - m1.
(b) 65/52 = 1.25; 25% increase.
(c) For male: y2 - y1 = 81 - 71 = 10; There was an estimated 10 kg increase in the mean weight of adult
Indian men between 1968 and 2012. 81/71 = 1.14. There was an estimated 14% increase in the mean weight
of adult Indian men between 1968 and 2012.
7.6. (a) The response variable is household consumption expenditure in domestic currency in 2015. The
explanatory variable is household (rural, urban).
(b) (i) 2, 65, 000 - 51, 000 = 2,14, 000 in domestic currency unit. There is an estimated difference of
2,14,000 in domestic currency unit in median household consumption expenditure between urban
and rural households in 2015.
(ii) 265000/51000 . The median household consumption expenditure for urban households in
2015 was estimated to be 5.20 times the median consumption expenditure in rural households in
2015.
7.7. (a) 425/375 1.13. The proportion of dropouts among the girl students was 1.13 times the proportion for
male students.
(b) 0.0425-0.0375 0.005. The difference in proportions is extremely small.
(c) The ratio is more informative, especially since the proportions are very low.
7.8. (a) 0.00237 0.00069 0.00168. This estimated difference of 0.002 seems to be very small.
(b) 0.00237/0.00069 = 3.43. The probability that a male student fails to go abroad for higher studies is 3.43
times the probability that a female student fails to do that. The effect seems to be large.
(c) The ratio better summarizes results, especially when both proportions are very small.
7.9. (a) We can be 95% confident that the population proportion who sometimes donate is between 0.21 and 0.28
higher for residents of Country B than for Country A.
(b) Extremely strong evidence that the population proportion who sometimes donate is different in Country B
than in Country A.
7.10. (a) Since 0 is lying within the confidence interval, there may have been no change in support.
(b) If there was a decrease in the proportion of approval, the decrease may have been as much as 0.09 (9%).
If there was an increase in the proportion of approval, the increase was small, only as much as 0.03 (3%).

Copyright © 2018 Pearson Education Ltd.
50 Statistical Methods for the Social Sciences

pˆ1 ( 1 - pˆ1 ) pˆ2 ( 1 - pˆ2 ) 0.352 ( 1 - 0.352) 0.595 ( 1 - 0.595)

7.11. (a) se = + = + = 0.0098
n1 nn 4690 5120

(b) CI = ( pˆ1 - pˆ2 ) ± 1.96 ( 0.0098) = ( 0.22, 0.26) . We are 95% confident that interval 0.22 to 0.26 contains
the difference in the proportion of male and female school students who borrow books to build up images.
7.12. (a) We have two independent random samples with at least 10 observations in each category for each group.
Let p 1 = the proportion of students in 2010 who said they performed better in interschool competitions
because of reading science fictions and p 2 = the proportion of students in the latest survey who said they
performed better in interschool competitions because of reading science fictions. The null hypothesis is
H 0 : p 2 - p 1 = 0, and the alternative hypothesis is H a : p 2 - p 1 ¹ 0.
(b) If the true difference in proportions were really 0, the chance of seeing results as extreme as or more
extreme than we did would happen about 0.05% of the time. In other words, these results are unlikely to have
occurred by chance alone.
(c) We are 95% confident that the true difference in proportions between the latest survey and 2010 falls in
the interval 0.009 to 0.02. The lower bound of this interval is very close to 0, and the upper bound is not
strikingly different from 0 in a practical sense. Thus, while our hypothesis test results are statistically
significant, they do not appear to be practically important.

pˆ1 ( 1 - pˆ1 ) pˆ 2 ( 1 - pˆ 2 ) ( 0.55) ( 0.45) ( 0.74 ) ( 0.26)

7.13. se = + = + = 0.022; ( pˆ1 - pˆ 2 ) ± z ( se )
n1 nn 1219 733
= ( 0.74 - 0.55) ± 1.96 ( 0.022 ) = ( 0.147,0.233) ; We are 95% confident that the interval 0.147 to 0.233
contains the true difference in proportions that working women and men spend time cooking and washing up
during a typical day.
7.14. (a) pˆ1 = 989 1503 = 0.658; pˆ2 = 515 1655 = 0.311; pˆ2 - pˆ1 = 0.658 - 0.311 = 0.347;
pˆ1 ( 1 - pˆ1 ) pˆ2 ( 1 - pˆ 2 ) ( 0.658) ( 0.342) ( 0.311) ( 0.689 )
se = + = + = 0.017
n1 nn 1503 1655
(b) ( pˆ1 - pˆ2 ) ± z ( se ) = ( 0.658 - 0.311) ± 1.96 ( 0.017 ) = ( 0.31, 0.38) ; We are 95% confident that the interval
0.31 to 0.38 contains the true difference in proportions who agree that "it is much better if the man is the
achiever outside the home and the woman takes care of the home and family" between 1977 and 2014.
(c) The se would not change in the confidence interval. The sample proportions for each sample would be
the respective proportions in part (a) subtracted from 1. Thus, the 95% confidence interval for comparing the
proportions who did not agree in the two years would be ( 1 - 0.38, 1 - 0.31) = ( 0.62, 0.69 ) .

7.15. (a) Let p 1 = the proportion of females who agree with the statement and p 2 = the proportion of males who
agree with the statement. The null hypothesis is H 0 : p 2 - p 1 = 0, and the alternative hypothesis is
H a : p 2 - p 1 ¹ 0.
247 + 268 103
(b) pˆ2 - pˆ1 = 0.332 - 0.295 = 0.037; pˆ = = » 0.311;
745 + 910 331
pˆ ( 1 - pˆ ) pˆ ( 1 - pˆ ) ( 0.311) ( 0.689 ) ( 0.311) ( 0.689)
se0 = + = + = 0.023;
n1 nn 745 910
pˆ 2 - pˆ1 0.037
z= = = 1.61 ( 1.62 using unrounded values)
se0 0.023
(c) P = 0.105; If there were no difference between the proportion of men who agree with the statement and
the proportion of women who agree with the statement, we would see results as extreme as our sample results
with probability 0.105.

Copyright © 2018 Pearson Education Ltd.
 Chapter 7: Comparison of Two Groups 51

7.15. (continued)
(d) Let p 1 = the proportion of those having less education than a college degree who agree with the
statement and p 2 = the proportion of those having at least a college degree who agree with the statement.
The null hypothesis is H 0 : p 2 - p 1 = 0, and the alternative hypothesis is H a : p 2 - p 1 ¹ 0.
pˆ2 - pˆ1 = 0.371 - 0.212 = 0.159; pˆ » 0.311;
pˆ ( 1 - pˆ ) pˆ ( 1 - pˆ ) ( 0.311) ( 0.689 ) ( 0.311) ( 0.689) pˆ 2 - pˆ1 0.159
se0 = + = + = 0.023; z = = = 6.91;
n1 nn 1032 623 se0 0.023
P » 0; It appears that opinion differed more between the most and least educated than between men and
women.
7.16. (a) Let p 1 be the proportion of senior college teachers who use interactive teaching learning methods. And
let p 2 be the proportion of junior college teachers who use innovative teaching learning methods.
H 0 : p 2 - p 1 = 0, H a : p 2 - p 1 ¹ 0
(b) We are 95% confident that the proportion of junior college teachers who have used interactive teaching
learning method is between 2% and 10% higher than the proportion of senior college teachers who have used
innovative teaching learning method.
(c) If the difference between the true proportions of junior and senior college teachers who have ever used
interactive teaching learning methods were actually 0, we would see results like these with probability 0.002.
It appears that the proportion of junior college teachers who have ever used interactive teaching learning
methods is greater than the proportion of senior college teachers who have ever used interactive teaching
learning methods.
7.17. Let p 1 = the proportion of men judged to be compulsive buyers and p 2 = the proportion of women judged
44 80
to be compulsive buyers. H 0 : p 2 - p 1 = 0; H a : p 2 - p 1 ¹ 0; pˆ1 = = 0.055; pˆ2 = = 0.060;
800 1501
44 + 90 134 pˆ ( 1 - pˆ ) pˆ ( 1 - pˆ ) ( 0.058) ( 0.942 ) ( 0.058) ( 0.942)
pˆ = = = 0.058; se0 = + = + = 0.010;
800 + 1501 2301 n1 nn 800 1501
pˆ 2 - pˆ1 0.060 - 0.055
z= = = 0.48; P = 0.63; It appears that there is no difference between the proportion of
se0 0.010
men judged to be compulsive buyers and the proportion of women judged to be compulsive buyers. A 95%
confidence interval for this difference is ( - 0.015, 0.025) .

7.18. Let p 1 = the proportion of females who believe in an afterlife and p 2 = the proportion of males who believe
600 424 1024
in an afterlife. H 0 : p 2 - p 1 = 0; H a : p 2 - p 1 ¹ 0; pˆ1 = = 0.846; pˆ2 = = 0.714; pˆ = = 0.786;
709 594 1303
( 0.786) ( 0.214 ) ( 0.786) ( 0.214) pˆ 2 - pˆ1 0.714 - 0.786
se0 = + = 0.023; z = = = - 5.81; P » 0; It appears
709 594 se0 0.023
that there is a difference between the proportion of females who believe in an afterlife and the proportion of
males who believe in an afterlife. If an error has been made with this test, it would be a Type I error
(rejecting the null hypothesis when it is actually true).
7.19. (a) We are 95% confident that the mean number of close friends for males is between 1.5 close friends below
and 2.7 close friends above the mean number of close friends for females.
(b) It does not appear that the distribution of the number of close friends is normal for either sex, since the
standard deviations exceed their respective means (suggesting right-skewed distributions). Because the
sample sizes are large, the t procedures will be robust against the violation of normality.

Copyright © 2018 Pearson Education Ltd.
52 Statistical Methods for the Social Sciences

7.20. (a) Let m1 = the mean number of hours spent in laboratory per week for junior research fellows (JRF) and
m2 = the mean number of hours spent in laboratory per week for SRF. The estimated difference is
y2 - y1 = 7.5 - 5.5 = 2 hours.

s12 s22
(b) se0 = + = 0.58
n1 n2

(c) The 99% confidence interval is ( y2 - y1 ) ± t ( se) 2 ± 2.576(0.58) = ( 0.51, 3.49) . We are 99% confident
that the mean number of hours spent in laboratory per week is between 0.51 hours and 3.49 hours greater for
SRF than for JRF.
7.21. (a) We are 95% confident that the checklist means for observers is between 3.60 and 5.40 more than the
checklist means for ex-observers.
(b) The checklist sample data distribution for ex-observers appears to be right-skewed, since the standard
deviation is greater than the mean. If there are no outliers, the inference might not be affected too much,
since t procedures are robust against violations of normality and the sample size is large.
7.22. (a) Let m1 = the mean number of minutes spent on cooking and washing up per day for men and m2 = the
mean number of minutes spent on cooking and washing up per day for women. The estimated difference is

y2 - y1 = 37 - 23 = 14 minutes. The standard error is se = s12 s22 ( 32) 2 ( 16) 2

+ = + = 1.091
n1 n2 733 1219
y2 - y1 14
(b) H 0 : m2 - m1 = 0; H a : m2 - m1 ¹ 0; z = = = 12.8; P » 0.00; It appears that there is a
se 1.091
difference between the mean number of minutes spent per day cooking and washing up for males and
females.
7.23. We are 95% confident that the mean number of days in the past 15 days that executives have felt demotivated
is between 1.5 and 3.2 greater than the mean number of days in the past 15 days that non-executives have felt
demotivated. Both the confidence interval and the hypothesis test lead us to believe that there is a difference
in the mean number of days in the past 15 days that executives have felt demotivated and the mean number of
days in the past 15 days that nonexecutives have felt so. These results are statistically significant, but one
must decide if the confidence interval indicates practical significance.
7.24. (a) Let m1 = the mean number of hours a day that in-service people spend on gardening and m2 = the mean
number of hours a day that the retired persons spend on gardening.
s12 s22
y2 - y1 = 4.65 - 2.75 = 1.9. se0 = + = 0.09. H 0 : m2 - m1 = 0 and H a : m2 - m1 ¹ 0; z = 21.11;
n1 n2
P-value close to 0. It appears that there is a difference in the mean number of hours a day that retired persons
spend on gardening and the mean number of hours a day that in-service people spend on gardening.
(b) A 95% confidence interval comparing the means would not contain 0.
(c) The distribution appear to be normal, since the standard deviations are not as large as their respective
means. Moreover, the sample sizes are large and the t procedures are robust against violations of normality,
the inferences are probably fine.
7.25. (a) We are 95% confident that the mean number of hours that teenagers surf the Internet per day is between
1.08 and 1.62 hours more than the mean number of hours that retired people surf the Internet per day. Since
the entire confidence interval is above 0, it appears that the mean number of hours that teenagers surf the
Internet per day is higher than the mean number of hours that retired people surf the Internet per day.
(b) The P -value of 0.000 says that, if the means were equal, it would be almost impossible to observe results
as extreme as the results that we have.

Copyright © 2018 Pearson Education Ltd.
 Chapter 7: Comparison of Two Groups 53

7.25. (continued)
(c) The result of the significance test and the result of the confidence interval both lead us to believe that the
mean number of hours that teenagers surf internet per day is higher than the mean number of hours that
retired persons surf internet per day.
7.26. The desire for an additional type of drink for a subject at one length of time might influence the desire for an
additional drink at another length of time. The sample size is 25 + 1 = 26. If the length of time after drinking
a coffee with cream does not affect the desire for an additional drink, the chance of seeing results as extreme
as or more extreme than we did would happen about 0.11% of the time.
30.0 - 21.5
7.27. The test statistic is t = = 3.96; with df = 21 - 1 = 20, which has P < 0.01. It appears that the
9.84 21
tomoxeline treatment has a negative affect on mean ADHD ratings.
7.28. (a) We should treat these as dependent samples, since the same subjects are used in each sample.
(b) We are 95% confidence that the mean number of times students went to movies was between 7.6 below
and 15.6 above the mean number of times students went to sporting events.
(c) t = 4.0 5.112 = 0.78; P = 0.454; It appears that there is no significant difference between the mean
number of times students went to movies and the mean number of times students went to sporting events.
7.29. If there was no difference between the mean number of times students went to parties was between and the
mean number of times students went to sporting events, we would see results as extreme as ours with
probability 0.106. There does not appear to be a difference between mean attendance at parties and mean
attendance at sporting events. Since 0 is in the confidence interval, we would fail to reject Ho and come to
the same conclusion that we did in using the P -value.

( n1 - 1) s12 + ( n2 - 1) s22 ( 3 - 1) ( 10) 2 + ( 3 - 1) ( 8.7) 2 1 1 1 1

7.30. (a) s = = = 9.35; se = s + = 9.35 + = 7.64
n1 + n2 - 2 3+3- 2 n1 n2 3 3

(b) The 90% confidence interval is ( y2 - y1 ) ± t ( se ) = 20 ± 2.132 ( 7.64 ) = ( 3.7, 36.3) ; We are 90% confident
that the mean improvement scores for patients receiving therapy B is between 3.7 and 36.3 points higher than
the mean improvement scores for patients receiving therapy A.
y2 - y1 40 - 20
(c) The effect size is = = 2.1. The difference between the sample means is more than 2
s 9.35
standard deviations, a relatively large difference.
7.31. (a) (i) 40 - 20 = 20
(ii) 20; The difference between the means is equal to the mean difference.
(b) s = 5.0; se = 5 2 = 2.887

(c) The 95% confidence interval is yd ± t ( se ) = 20 ± 4.303 ( 2.887) = ( 7.6, 32.4 ) ; We are 95% confident that
the mean difference between patients receiving therapy B and patients receiving therapy A is between 7.6
and 32.4.
(d) t = 20 2.887 = 6.93; df = 3 - 1 = 2; 0.02 < P < 0.05 ( P = 0.0202 ) . There appears to be a difference
between the mean scores for patients receiving therapy B and those receiving therapy A.
7.32. (a) Let m1 = the mean weight loss for women in group treatment and m2 = the mean weight loss for women
in individual treatment. H 0 : m2 - m1 = 0; H a : m2 - m1 ¹ 0; The first test assumes the two samples have the
same level of variation among their values while the second test does not make this assumption. In both
cases, you would reject H 0 .
(b) We are 95% confident that the mean weight loss for group treatment is between 1.45 and 9.35 kg higher
than the mean weight loss for individual treatments.

Copyright © 2018 Pearson Education Ltd.
54 Statistical Methods for the Social Sciences

7.33. The sample standard deviations are quite different, so we might not trust the results based on assuming equal
variances. The approximate test without that assumption shows very strong evidence ( P = 0.0066 ) that the
means differ; the data suggest that vegetarians are more liberal, on average.
7.34. Let m1 = the mean number of hours per week that females spend on the WWW and m2 = the mean number
of hours per week that males spend on the WWW. y2 - y1 = 11.9 - 11.2 = 0.7;

( 778 - 1) ( 13.7) 2 + ( 620 - 1) ( 16.0) 2 1 1

= 0.795 H 0 : m2 - m1 = 0;
s= = 14.764; se = 14.764 +
778 + 620 - 2 778 620
0.7 - 0
H a : m2 - m1 ¹ 0; t = = 0.88; P = 0.379. It appears that there is no difference in the mean number of
0.795
hours a week that females spend on the WWW and the mean number of hours a week that males spend on the
WWW.
Minitab Output:
Two-Sample T-Test and CI
Sample N Mean StDev SE Mean
1 620 11.9 16.0 0.64
2 778 11.2 13.7 0.49
Difference = μ (1) - μ (2)
Estimate for difference: 0.700
95% CI for difference: (-0.859, 2.259)
T-Test of difference = 0 (vs ≠): T-Value = 0.88 P-Value = 0.379 DF =
1396
Both use Pooled StDev = 14.7641

7.35. (a) There is very strong evidence that the mean drop was higher for Course B. A 95% confidence interval for
the difference in means is: ( - 6.3, - 1.7 ) . We are 95% confident that the mean drop for Course B was between
1.7 and 6.3 points higher than the mean drop for Course A.
Minitab Output:
Two-Sample T-Test and CI: Drop Score, Course
Course N Mean StDev SE Mean
A 5 2.00 1.22 0.55
B 5 6.00 1.87 0.84
Difference = μ (A) - μ (B)
Estimate for difference: -4.00
95% CI for difference: (-6.31, -1.69)
T-Test of difference = 0 (vs ≠): T-Value = -4.00 P-Value = 0.004 DF =
8
Both use Pooled StDev = 1.5811
(b) P = 0.02; There is very strong evidence that the mean drop was higher for Course B.
Mann-Whitney Test and CI: Course A, Course B
Point estimate for η1 - η2 is -4.000
96.3 Percent CI for η1 - η2 is (-5.999,-1.001)
W = 16.0
Test of η1 = η2 vs η1 ≠ η2 is significant at 0.0216
The test is significant at 0.0192 (adjusted for ties)

(c) The effect size is ( 6.0 - 2.0 ) /1.581 = 2.5. The difference between the sample means is about 2.5 standard
deviations, a relatively large difference.
(d) The effect size P ( y B > y A ) = 0.92. The estimated probability that a student in Course B has a larger drop
in math phobia score than a student in Course A is 0.92. This is a relatively strong effect.

Copyright © 2018 Pearson Education Ltd.
 Chapter 7: Comparison of Two Groups 55

7.36. (a) Let p 1 = the proportion correct with GMDS and p 2 = the proportion correct with CDHMM.
pˆ1 = 1979 / 2000 = 0.99; pˆ2 = 1937 / 2000 = 0.97
1979 - 1937
(b) H 0 : p 2 - p 1 = 0, H a : p 2 - p 1 ¹ 0; McNemar z = = 0.67; P = 0.50; There is not enough
1979 + 1937
evidence to conclude that the proportion of correct word recognition differs between GMDS and CDHMM.
(c) pˆ1 - pˆ 2 = 0.02; se = ( 1979 + 1937) - ( 1979 - 1937) 2 2000 2000 = 0.0313; A 95% confidence interval
is 0.02 ± 1.96 ( 0.0313) = ( - 0.041,0.0813) ; We are 95% confident that the interval –0.041 to 0.081 contains
the true difference in the proportion of correct word recognition between GMDS and CDHMM.
7.37. (a) Let p 1 = the proportion who favor law enforcement spending and p 2 = the proportion who favor health
spending. pˆ1 = 306 340 = 0.900; pˆ2 = 317 340 = 0.932
25 - 14
(b) H 0 : p 2 - p 1 = 0, H a : p 2 - p 1 ¹ 0; McNemar z = = 1.76; P = 0.078; There is not enough
25 + 14
evidence to conclude that the proportion of people who favor law enforcement spending differs from the
proportion of people who favor health spending.
(c) pˆ1 - pˆ 2 = - 0.032; se = ( 25 + 14) - ( 25 - 14) 2 340 340 = 0.0183; A 95% confidence interval is
- 0.032 ± 1.96 ( 0.0183) = ( - 0.067,0.004 ) ; We are 95% confident that the population proportion for increased
spending is between 0.004 less and 0.067 higher for spending on health than for spending on law
enforcement.
7.38. Let p 1 = the proportion of subjects in the 1981–1984 survey that responded “never” and p 2 = the
proportion of subjects in the 2010–2014 survey that responded “never”. H 0 : p 2 - p 1 = 0; H a : p 2 - p 1 ¹ 0;
0.624 ( 2325) + 0.240 ( 2232 ) ( 0.436) ( 0.564) ( 0.436) ( 0.564 )
pˆ1 = 0.624; pˆ2 = 0.240; pˆ = = 0.436; se0 = +
2325 + 2232 2325 2232
0.240 - 0.624
= 0.014; z = = - 25.6; P » 0; It appears that there is a difference between the proportion of
0.015
subjects that responded “never” in the two surveys.
7.39. (a) Fisher's exact test is used because the groups have small sample sizes.
(b) The P -value is 0.0034, so there is strong evidence that the probability of prosocial behavior is higher in
the Harm condition.
7.40. Using software, Fisher's exact test has a one-sided P -value of 0.022. There is strong evidence that the
probability is higher for those raised with lesbian mothers.
7.41. (a) Let m1 = the mean political ideology for Democrats and m2 = the mean political ideology for
Republicans. H 0 : m2 - m1 = 0; H a : m2 - m1 ¹ 0; t = - 7.94; P = 0.00
Minitab Output
Two-Sample T-Test and CI: Ideology, Affiliation
Affiliation N Mean StDev SE Mean
Democrat 21 2.000 0.837 0.18
Republican 15 5.13 1.36 0.35

Difference = μ (Democrat) - μ (Republican)

Estimate for difference: -3.133
95% CI for difference: (-3.954, -2.312)
T-Test of difference = 0 (vs ≠): T-Value = -7.94 P-Value = 0.000 DF =
21

Copyright © 2018 Pearson Education Ltd.
56 Statistical Methods for the Social Sciences

7.41 (continued)

1 2 3 4 5 6 7
Democrat Republican
30

25

20

Frequency
15

10

0
1 2 3 4 5 6 7
Political Ideology

The hypothesis test tells us that there is enough evidence to conclude that the mean political ideology differs
for students identifying with the Democratic party and with the Republican party. The confidence interval
tells us that we are 95% confident that the mean political ideology for students identifying with the
Democratic party is between 2.3 and 4.0 below (more liberal) the mean political ideology for students
identifying with the Republican party.
(b) Let p 1 = the proportion of males that believe abortion should be available and p 2 = the proportion of
females that believe abortion should be available H 0 : p 2 - p 1 = 0; H a : p 2 - p 1 ¹ 0; z = 1.08; P = 0.28
Minitab Output
Test and CI for Two Proportions: Abortion, Gender
Gender X N Sample p
Female 26 31 0.838710
Male 21 29 0.724138

Difference = p (Female) - p (Male)

Estimate for difference: 0.114572
95% CI for difference: (-0.0933331, 0.322477)
Test for difference = 0 (vs ≠ 0): Z = 1.08 P-Value = 0.280
Fisher’s exact test: P-Value = 0.355
There is not enough evidence to conclude that the proportion of females who support the legalization of
abortion differs from the proportion of males who support the legalization of abortion. We are 95% confident
that the interval –0.09 to 0.32 contains the difference in the proportion of females who support the
legalization of abortion and the proportion of males who support the legalization of abortion.
(c) Let m1 = the mean weekly time in sports and other physical exercise and m2 = the mean weekly time
spent watching TV. H 0 : m2 - m1 = 0; H a : m2 - m1 ¹ 0; t = 1.78; P = 0.078
Minitab Output
Two-Sample T-Test and CI: TV, Sports
N Mean StDev SE Mean
TV 60 7.27 6.72 0.87
Sports 60 5.48 3.87 0.50

Difference = μ (TV) - μ (Sports)

Estimate for difference: 1.78
95% CI for difference: (-0.20, 3.77)
T-Test of difference = 0 (vs ≠): T-Value = 1.78 P-Value = 0.078 DF =
94

Copyright © 2018 Pearson Education Ltd.
 Chapter 7: Comparison of Two Groups 57

7.41 (continued)
There is not enough evidence to conclude that the mean weekly time spent watching TV differs from the
mean weekly time in sports and other physical exercise. We are 95% confident that the interval –0.2 to 3.8
contains the difference in the mean weekly time spent watching TV differs from the mean weekly time in
sports and other physical exercise.
7.42. Answers will vary.

7.43. (a) Answers will vary. One possible solution is ( 0,5,5,10 ) and ( 5,10,10,15) .
(b) Answers will vary. One possible solution is ( 4,5,5,6 ) and ( 9,10,10,11) , for which there is much less
variability within each group.
7.44. Answers will vary. 7.46. Answers will vary.
7.45. Answers will vary.
7.47. The report should include the following: m1 = the proportion of teens who watch TV less than 1 hour per day
and have committed aggressive acts and m2 = the proportion of teens who watch TV at least 1 hour per day
and have committed aggressive acts. pˆ1 = 5 88 = 0.06; pˆ2 = 154 619 = 0.25; The report should also include
a hypothesis test ( z = 4.04, P < 0.0001) and/or a confidence interval ( 95% CI: ( 0.25, 0.13) ) .

7.48. Answers will vary. One possible approach is to compare the number of annual gun homicides per year for
each possible pair of countries using either confidence intervals or significance tests.
7.49. (a) (i) The margin of error of the confidence interval is 0.2.
(ii) The P -value is 0.001.
(b) Obtaining the given results, if the population means were actually the same, would occur 0.01% of the
time. This is unlikely enough to make us reject the claim that the mean tolerance level for homosexuality was
the same for both surveys.
7.50. One possible approach compares attractiveness within each gender. For example, a 95% confidence interval

for More – Less for Men is: ( 9.7 - 9.9 ) ± 1.96 ( 10) 2 ( 12.6) 2
+ = ( - 5.5, 5.1) , for which it is plausible that
35 36
there is no difference. By contrast, a 95% confidence interval for More – Less for Women is:
( 14.2 ) 2 ( 16.6) 2
( 17.8 - 10.4 ) ± 1.96 + = ( - 0.52, 15.3) . Here, 0 is barely in the interval, and it is plausible that
33 27
the mean is very much larger for the more attractive group. One could alternatively compare means between
genders within each level of attractiveness.
7.51. The samples are dependent, so you need to know the sample proportions for the four possible sequences of
responses on the two questions (i.e., (yes, yes), (yes, no), (no, yes), (no, no) for confidence in (European
Union, the Netherlands))
7.52. We would not have enough information. We also need to know the number of male managers and number of
female managers.
7.53. For one to be confident that an actual difference exists between the population means, the effect size would
need to be at least 2 (there would have to be a difference of at least 2 standard deviations between the means).
For example looking at the difference between Canada and the U.S., we have an effect size of
( 286.9 - 277.9 ) 70.7 = 0.13, which is a relatively small difference.
1( 0 ) + 29 ( 5) + 0 ( 10 )
7.54. (a) The mean for hospital A is = 4.83 and the mean for hospital B is
1 + 29 + 0
8 ( 0 ) + 8 ( 5) + 14 ( 10)
= 6.00.
8 + 8 + 14

Copyright © 2018 Pearson Education Ltd.
58 Statistical Methods for the Social Sciences

7.54 (continued)
1( 0 ) + 29 ( 9 ) + 0 ( 10 )
(b) The mean for hospital A is = 8.70 and the mean for hospital B is
1 + 29 + 0
8 ( 0 ) + 8 ( 9 ) + 14 ( 10)
= 7.07.
8 + 8 + 14
The first researcher concludes that Hospital B is better, whereas the second researcher concludes that
Hospital A is better.
7.55. The se value for estimating a difference between two means will be larger than the se value for estimating a
single mean. Thus, it will be more difficult to reject the null hypothesis since the test statistic will be smaller.
7.56. (a) No; Since the two samples are the same, we cannot treat the results as if they came from independent
samples.
(b) In order to make an inferential comparison of proportions, one would need to know how many people
said yes to both questions, no to both questions, yes to the first question and no to the second question, and
no to the first question and yes to the second question.
7.57. (a)
Number of Males in the Sample Possible Samples of Size 3
0 ( F1, F2 , F3 )
1 ( M1, F1 , F2 ) , ( M1 , F1, F3 ) , ( M1 , F2 , F3 )
1 ( M 2 , F1, F2 ) , ( M 2 , F1, F3 ) , ( M 2 , F2 , F3 )
1 ( M 3 , F1, F2 ) , ( M 3 , F1, F3 ) , ( M 3 , F2 , F3 )
2 ( F1, M1 , M 2 ) , ( F1 , M1, M 3 ) , ( F1, M 2 , M 3 )
2 ( F2 , M1, M 2 ) , ( F2 , M1, M 3 ) , ( F2 , M 2 , M 3 )
2 ( F3 , M1, M 2 ) , ( F3 , M1 , M 3 ) , ( F3 , M 2 , M 3 )
3 ( M1, M 2 , M 3 )
(b) Each of the 20 samples is equally likely. The 10 samples with 2 or 3 males chosen have pˆ1 - pˆ 2 ³ 1 3.
(c) The one-sided P -value is P = 1/ 20 = 0.05.
7.58. Answers will vary.
7.59. (a) False; the confidence interval addresses the difference between the population proportions for Hispanic
and white youths, not the population proportion of white youths alone.
(b) False, the samples are independent, because a youth is either a female or a male.
7.60. False, we can conclude that m2 is greater than m2 but the confidence interval does not indicate plausible
values for the separate population means.
7.61. True, from the formula on page 181 you could calculate the standard error of the difference between the
sample means.
7.62. Response (b) is correct.
7.63. Responses (a), (c), and (d) are correct.
7.64. Responses (a) and (b) are correct.
7.65. (a) The sample proportion correct has approximately a normal sampling distribution with mean 0.5 and
standard error ( 0.5) ( 0.5) 100 = 0.05. A score of 70 has z = ( 0.7 - 0.5) 0.05 = 4.0. The probability of a
score of at least 70 is approximately the probability that a normal random variable falls at least 4 standard
deviations above its mean, which is less than 0.0001.

Copyright © 2018 Pearson Education Ltd.
 Chapter 7: Comparison of Two Groups 59

7.65 (continued)
(b) The sampling distribution of the difference between Jane's and Joe's proportions is approximately normal
with mean 0.60.5 = 0.1 and standard error ( 0.6) ( 0.4 ) 100 + ( 0.5) ( 0.5) 100 = 0.07. The probability that the
difference is negative is approximately equal to the probability that a z -score from a standard normal
distribution is less than ( 0 - 0.1) 0.07 = 1.43, which is 0.08.
(c) The standard errors decrease as the number of questions increases, and the probabilities decrease. Thus,
larger sample sizes are more likely to reveal the truth concerning Joe's achievement and concerning the fact
that Jane knows somewhat more than Joe.

7.66. (a) yd = å yd n = å ( yi 2 - yi1 ) n = å yi 2 n - å yi1 n = y2 - y1

(b) no

Copyright © 2018 Pearson Education Ltd.