Sociology 592 - Research Statistics I

Exam 1 ANSWER KEY

September 27, 2002

Where appropriate, show your work - partial credit may be given. (On the other hand, don't waste a lot of time on
excess verbiage.) Do not spend too much time on any one problem. It is legitimate (and probably essential) to refer
to results that have previously been proven in class or homework, without re-proving them - for example, you
wouldn't need to prove that P(-1.96 Z 1.96) = .95, since we have already shown that in class. Likewise, you are
free to refer to anything that was demonstrated in the homework or handouts.

1. (5 points each, 20 points total). Indicate whether the following statements are true or false. If you think the
statement is false, indicate how the statement could be corrected. For false statements, do not just say that you could
substitute not equals for equals. For example, the statement P(Z 0) = .7 is false. To make it correct, don't just say
P(Z 0) <> .7, instead say P(Z 0) = .5 or P(Z .525) = .7.

A. V(5X) = 5 * V(X)

False. We know V(aX) = a2V(X), so V(5X) = 25 * V(X)

B. P (Z .84) = .79954586.

False. P (Z .84) = .79954586 or else P(Z .84) = 1 - .7995486 = .2004514.

C. If E(X2) = 30 and E(X) = -5, then V(X) = 55.

False. E(X2) E(X)2 = 30 (-5)2 = 30 25 = 5

D. If a fair coin is tossed 5 times, there is a 50% of getting exactly 3 heads.

False. There is a 50% chance of getting 3 heads or more. Or, P(3 heads) =

N r N r 5 3 2 5!
p q = .5 .5 = .03125 = 10 * .03125 = .3125
r 3 3!2!

You can also look at Table 2 in Appx E, N = 5, p = .5, r = 3.

2. (10 points each, 30 points total) Answer three of the following. The answers to most of these are fairly
straightforward, so do not spend a great deal of time on any one problem. NOTE: I will give up to 5 points extra
credit for each additional problem you do correctly.

A. X = 7, N = 14. Determine the 95% confidence interval when

a. = 20

For s = 20, look at Appx E, Table 3, v = 13, 2Q = .05:

x ( t /2,v * s/ N ), i.e.
7 - ( 2.16 * 20 / 14 ) 7 + ( 2.16 * 20 / 14 ), i.e.
4.546 18.546

Sociology 592, Exam 1 Answer Key, September 27, 2002 Page 1

Stata double-check:
. cii 14 7 20

Variable | Obs Mean Std. Err. [95% Conf. Interval]

-------------+---------------------------------------------------------------
| 14 7 5.345225 -4.547656 18.54766

b. = 15

For = 15:

x ( z/2 * ), i.e.,
N
7 - (1.96 * 15 / 14 ) 7 + (1.96 * 15 / 14 ), i.e.,
.857 14.857

Stata double-check:
. ztesti 7 15 0 14, level(95)

Number of obs = 14

------------------------------------------------------------------------------
Variable | Mean Std. Err. z P>|z| [95% Conf. Interval]
---------+--------------------------------------------------------------------
x | 7 4.008919 1.74611 0.0808 -.8573361 14.85734
------------------------------------------------------------------------------

B. Here are the results from a previous cohorts first exam in statistics. Compute the mean and
variance of the scores. There were 10 Sociology Students in the class.

Score
74
83
84
92
93
95
97
98
104
106

Sociology 592, Exam 1 Answer Key, September 27, 2002 Page 2

Lets expand the table as follows:

(Xi - )2 Xi2
Score
74 345.96 5476
83 92.16 6889
84 73.96 7056
92 0.36 8464
93 0.16 8649
95 5.76 9025
97 19.36 9409
98 29.16 9604
104 129.96 10816
106 179.56 11236

926 876.4 86624

So, = 926/10 = 92.6. 2 = 876.4/10 = 87.64, = 9.36.

Or, if you prefer, 2 = E(X2) E(X)2 = 86624/10 92.62 = 87.64, = 9.36.

C. A company has been deluged with dozens of job applications. Only those candidates scoring in
the top third on an aptitude test will be hired. If Scores ~ N(80, 102), how high does your score have to be for you to
get hired?

Z = .43, since F(.43) = .6664. So, the corresponding X score is X = Z + = .43*10 +

80 = 84.3.

Confirming with Stata,

. display invnorm(2/3) * 10 + 80

84.307273

D. It is October 26, 2002. The undefeated Notre Dame football team has continued its amazing
success story, winning each of its last three games with 50 yard field goals in the closing seconds. Today, however,
it faces its toughest challenge of the season: Florida State University.
The Irish coaches estimate that, if they can defeat Florida State, there is a 90% chance that Notre Dame will
get to play in the Bowl Championship Series (BCS). But, if they lose, NDs chances of being in the BCS drop to
50%. They further estimate that there is a 15% chance that they will lose to Florida State and also play in the BCS.
What is the probability that Notre Dame will beat Florida State? What is the probability that Notre Dame
will play in the BCS?

We are told
P(BCS | Beat FSU) = .90
P(BCS | Lose to FSU) = .50
P(BCS Lose to FSU) = .15.

Sociology 592, Exam 1 Answer Key, September 27, 2002 Page 3

This implies P(Lose to FSU) = .30, since
P(BCS Lose to FSU) = P(Lose to FSU) * P(BCS | Lose to FSU) =
P(Lose to FSU) *.50 = .15.

So, P(Beat FSU) = .70 (Complements rule).

P(BCS) = P(Losing) * P(BCS|Lose to FSU) + P(Winning) * P(BCS|Beat FSU) =

(.30 * .50) + (.70 * .90) = .78

E. A polling firm reports that President Bushs foreign policy is supported by 70% of the American
public. The firm further reports that the (approximate) 99% confidence interval for Bushs approval is .582 p
.818. What was the sample size used in the study?

We know the formula for the lower and upper bounds of the approximate c.i. We also
know that the upper c.i. was .818, the sample p was .7, and the critical value for Z is
2.58. So, all we have to do is algebraically solve for N.

p q .21 Formula for upper bound of c.i.

.818 = .7 + 2.58 * = .7 + 2.58
N N

.21 Subtract .7 from both sides

.118 = 2.58
N

.21 Divide both sides by 2.58

.045736 =
N

N Take reciprocals
21.8644 =
.21

N Square both sides

478.05 =
.21

100 = N Multiply both sides by .21

3. (25 points) A recent study done in Finland suggests that stressful events have a bigger impact on mens
health than they do on womens. According to the synopsis of the report printed on the WebMD web pages
(http://content.health.msn.com/content/article/1685.53485)

The study found men who suffered a stressful life event were more likely than women to miss work due to illness in the following
months Interpersonal problems, financial difficulties, and violence among men were linked to psychological problems, such as
anxiety, mental distress, and lack of coherence. Financial difficulties and violence were also associated with heightened use of
cigarettes and alcohol, which was thought to lead to sick days. For women, none of these events increased the likelihood of a sick
leave.

A researcher in the United States has decided to replicate this study. Data are collected from 1000 men and 1000
women who had experienced major stressful life events in the past six months. Each respondent is coded as either

Sociology 592, Exam 1 Answer Key, September 27, 2002 Page 4

having missed work because of illness or not having missed work. In addition, the support networks (number of
people the respondent talks to about their problems) is measured. Support networks are classified as large or
small. She finds that 60% of men, but only 40% of women, became ill and missed work after experiencing
stressful events. She also finds that 70% of men have small support networks, while 60% of all women have large
support networks. Finally, she found that 200 men and 200 women had small support networks and did not miss
work.

a. (10 pts) Complete the following table, Remember, there were 1000 men and 1000 women in this
study.

Male Female

Income/GPA Small Large Small Large

support support support support
network network network network

Became ill and missed

work

Did not miss work

1000 1000

Items in italics can be inferred from the information given, and the rest of the table can
easily be deduced from there:

Male Female

Income/GPA Small Large Small Large

support support support support
network network network network

Became ill and missed 500 100 600 200 200 400
work

Did not miss work 200 200 400 200 400 600

700 300 1000 400 600 1000

b. (5 pts) What percentage of those who missed work had small support networks?
1,000 missed work (600 men and 400 women), and 700 of those (500 men, 200
women, a total of 70%) had small support networks.

c. (10 pts) As these figures show, men tend to have smaller support networks than do women, and
those with smaller support networks are more likely to get ill after stressful events and miss work. Suppose that men
had the same support network distribution as women did, i.e. 60% of men had large support networks. Suppose
further that it continued to be the case that men maintained their network-specific illness rates. What percentage of
men would then become ill and miss work after stressful events? Based on these results, do you think differences in
networks explain much of the difference in stress-related illness between men and women, or does it explain
relatively little?

Sociology 592, Exam 1 Answer Key, September 27, 2002 Page 5

For women, P(Small) = .4, P(Large) = .6.
For men, P(ILL|Small) = 5/7, P(ILL|Large) = 1/3.

So, with female composition and male illness rates,

P(ILL) = P(Small)W * P(ILL|Small)H + P(Large)W * P(ILL|Large)M =

(.4 * 5/7) + (.6 * 1/3) = .486

About 486 men would still miss work, compared to the original 600. Support networks
account for more than half (57%) of the male/female differences, but differences still
remain.

4. (25 points) A stock broker company claims that, despite recent rough economic conditions, only 30% of its
customers lost money last year. Angry investors suspect otherwise. A random sample of 70 investors is taken, and
it is discovered that 30 lost money last year. Test the companys claim at the .01 level of significance. Be sure to
indicate:

(a) The null and alternative hypotheses - and whether a one-tailed or two-tailed test is called for.
H0: p = .30 or else E(X) = 21
HA: p > .30 or else E(X) > 21

(b) The appropriate test statistic

The appropriate test statistic is

# of losers CC - Np 0 x CC - (70 * .3) x CC - 21

z= = =
Np0 q0 70 * .3 * .7 3.834

For the correction for continuity, we will subtract .5 if there are more than 21 losers, we
will add .5 if there are less than 21 losers.

(c) The critical region

For the critical region, we will reject H0 if Zc > 2.33

(d) The computed value of the test statistic

# of losers CC - Np0 x CC - (70 * .3) x CC - 21 30 .5 21 8.5
z= = = = = = 2.217
Np 0 q0 70 * .3 * .7 3.834 3.834 3.834

(e) Your decision - should the null hypothesis be rejected or not be rejected? Why?

Do not reject the null. The computed test statistic falls just barely within the acceptance
region.

Sociology 592, Exam 1 Answer Key, September 27, 2002 Page 6

Confirming with the Stata bintesti command,
. bintesti 70 29.5 .30 , normal

Variable | Obs Proportion Std. Error

---------+----------------------------------
x | 70 .4214286 .0590189

Ho: p = .3
z = 2.22

. display 1-norm(2.22)
.01320938

So, using the normal approximation to the binomial with correction for continuity, we see
that there is a 1.3% chance of getting this many investors losing money if the company
is telling the truth. Do not reject if using the .01 level of significance.

Using the more precise bitesti command,

. bitesti 70 30 .30

N Observed k Expected k Assumed p Observed p

------------------------------------------------------------
70 30 21 0.30000 0.42857

Pr(k >= 30) = 0.015236 (one-sided test)

Pr(k <= 30) = 0.992020 (one-sided test)
Pr(k <= 12 or k >= 30) = 0.025739 (two-sided test)

So, there is about a 1.5% chance that you would get this many investors losing money if
the company is telling the truth. Do not reject if using the .01 level of significance.

Sociology 592, Exam 1 Answer Key, September 27, 2002 Page 7