STA 110B Fall 1997

Final Exam
December 15, 1997

Name: Section:

I understand and agree to abide by the Duke honor code,

Signed:

Instructions

This is a closed-book exam, however, one 8.5 by 11 inch “crib sheet” is permitted. You may use a calculator
if you find it useful. Show your work in the space provided, but be concise. Correct but unsubstantiated
answers will receive no credit.

Point assignments for each of the problems are given in parentheses in the table below. You have 3 hours
total; plan accordingly. You must hand the exam at or before Noon, no extra time will be given. Good luck!

Page 1 Page 2 Page 3

1. (50)

2. (50)

Total (100)

1
1) A researcher, interested in the relationship between a variable Y and a covariable X, collects a random
sample (n=122) from the population of interest. The researcher’s hypothesis is that Y and X are linearly
unrelated. To test this hypothesis he fits the regression model

Yi = α + βXi + i

and obtains the estimates b = 0.0200 of β with standard error s.e.(b) = 0.0100, and a = 1.50 of α with
standard error s.e.(a) = 1.25.

The following multiple choice questions relate to this problem, mark correct answer(s) with an X:

a) (4 points) The researcher’s hypothesis can be written

i) Ho : b = 0, ii) HA : β 6= 0, iii) Ho : β = 0,

iv) Ho : α = 0, v) HA : α 6= 0, vi) Ho : a = 0.

b) (4 points) The t-statistic for a test of the researcher’s hypothesis is

i) t = 0.02, ii) t = 0.20, iii) t = 2.00,

which has

i) df = 120, ii) df = 121, iii) df = 122.

degrees of freedom associated with it.

The researcher calculates the p-value of his test to be 0.048.

c) (5 points) The p-value is defined to be the probability

i) that the null hypothesis is true,

ii) of observing as (or more) extreme a t-statistic assuming the null hypothesis is true.

2
1) Continued. A researcher, interested in the relationship between a variable Y and a covariable X, collects
a random sample (n=122) from the population of interest. The researcher’s hypothesis is that Y and X are
linearly unrelated. To test this hypothesis he fits the regression model

Yi = α + βXi + i

and obtains the estimates b = 0.0200 of β with standard error s.e.(b) = 0.0100, and a = 1.50 of α with
standard error s.e.(a) = 1.25.

The following multiple choice questions relate to this problem, mark correct answer(s) with an X:

d) (4 points) The researcher can conclude that

i) a linear relationship between X and Y is discernible at the 10% level,

ii) a linear relationship between X and Y is discernible at the 5% level,

iii) a linear relationship between X and Y is discernible at the 2% level,

iv) a linear relationship between X and Y is discernible at the 1% level.

e) (4 points) The “level” referred to in part e), is the probability of

i) type I error, α, ii) type II error, β,

and is set by the researcher

i) in advance of the experiment. ii) after the sample is taken.

f ) (5 points) Another way for the researcher to conduct such a classical hypothesis test, without the use of
p-values, is to calculate the t-statistic and reject the null hypothesis if it is

i) more extreme than its critical value ii) less extreme than its critical value

g) (4 points) By drawing a larger sample, the researcher would have

i) reduced ii) increased

the probability of committing a type II error. A type II error is defined as

i) rejecting Ho when it is true. ii) failing to reject Ho when it is false.

3
1) Continued. A researcher, interested in the relationship between a variable Y and a covariable X, collects
a random sample (n=122) from the population of interest. The researcher’s hypothesis is that Y and X are
linearly unrelated. To test this hypothesis he fits the regression model

Yi = α + βXi + i

and obtains the estimates b = 0.0200 of β with standard error s.e.(b) = 0.0100, and a = 1.50 of α with
standard error s.e.(a) = 1.25.

h) (5 points) What value does the fitted regression predict for Y when X = 3?

i) (5 points) Calculate a 95% confidence interval for the population slope β

j) (6 points) A test of the researcher’s hypothesis at the 5% level could also be conducted by

i) rejecting the hypothesis ii) failing to reject the hypothesis

if the hypothesized
i) estimate b ii) parameter β

falls outside the 95% confidence interval for

i) the estimate b. ii) the parameter β.

k) (4 points) Using least squares, the researcher’s estimate of the regression slope will be

i) unbiased, ii) biased

if the regression model is correct.

4
2) Twenty-four subjects with good hearing were each given 4 different hearing exams in the presence of
background noise. The purpose of this exercise was to determine if the 4 exams were calibrated equally on
normal-hearing subjects before they were used to diagnose hearing ability. The variable of interest, score,
is the score obtained on the exam. A two-way analysis of variance is used to model the variability in exam
score as a function of two categorical variables: subject and exam. By including the variable subject
in the model we are controlling for variability between subjects. Hence, any effect ascribed to exam should
be due to differences in calibration between the four exams. The ANOVA table obtained from fitting the
two-way analysis of variance follows.

Hearing Test ANOVA Table

Source Df SS MS F P
Exam ?? ?????? ????? ???? 0.000042
Subject 23 3291.8 143.1 3.96 0.0000049
Error 69 2495.9 ?????
Total 95 6705.0

a) (5 points) Calculate the sum of squares explained by exam.

b) (5 points) Calculate the mean square for the variable exam.

c) (5 points) Calculate the F-ratio for the test for a difference in calibration of hearing exams.

d) (5 points) How many degrees of freedom are associated with this F-ratio?

e) (5 points) Can the hypothesis that there is no difference in hearing exams be rejected at the α = 5%
level? Why/why not?

5
2) Hearing experiment, continued. The 2-way analysis of variance model for the hearing test data
specifies that the expected hearing score for subject i on exam j is

µ + αi + βj ,

where µ is the baseline score, αi is the coefficient of the dummy variable Subjcti and βj is the coefficient of the
dummy variable Examj . Estimates of these coefficients, found using multiple linear regression with dummy
variables to identify subjects (Subjct1, ..., Subjct24) and exams (Exam1, ..., Exam4), follows. Use it
to answer the remaining questions. Note that Subject 9’s performance on Exam 4 serves as the “reference
group.”

6
7
2) Hearing experiment, continued.

f ) (4 points) What is the predicted hearing score for Subject 1 on Exam 2?

g) (4 points) On average, how much higher/lower did Subject 1 score than Subject 9?

h) (4 points) On average, how much higher/lower do subjects score on Exam 2, than on Exam 4?

i) (4 points) Are hearing scores on Exam 2 discernibly different at the 5% level from those on Exam 4?
Why/why not?

j) (5 points) Are hearing scores on Exam 2 discernibly different at the 1% level from those on Exam 4?
Why/why not?

8
2) Hearing experiment, continued.

k) (4 points) Calculate a 95% confidence interval for the mean difference in scores between Exam 1 and
Exam 4.