Download as docx, pdf, or txt
Download as docx, pdf, or txt
You are on page 1of 11

NAME: _________________________________________________________________

Practice Final Exam

PLEASE DO THIS NOW:

 Write your NAME on this page of the exam (above).

General Instructions:

1. There are 145 points possible on the exam.

2. This exam is closed book. A calculator is permitted. You may not share resources with
anyone.

3. We provide space on the exam for your answers. Don’t feel that you have to use up all
the space provided.

4. For each question, we specify the maximum number of sentences you may write. You do
not have to reach that maximum. Quality matters more than quantity.

5. Read each question carefully and be sure to answer each question fully.

Good luck!
Question Possible points Actual points
Multiple choice 40 ____ * 4 = ____
1 10
2 10
3 10
4 5
5 10
6 10
7 10
8 10
9 10
10 10
11 10
TOTAL 145

PART I – MULTIPLE CHOICE (4 points each)


Directions: Choose the letter of the best answer and fill it in the grid below.

Question Answer
1
2
3
4
5
6
7
8
9
10
1. You have wage data for individuals in North America, as well as information on whether
the individual lives in the U.S., Canada, or Mexico. You regress wages on a Canadian
dummy, a Mexican dummy, and a constant. What does the constant represent?

A) The average wages for Canadians.


B) The average wages for Americans.
C) The average difference in wages between Americans and Canadians.
D) The average difference in wages between Americans and Mexicans.

2. With data on individual workers, you regress daily earnings on years of education, years of
education squared, and a constant. You obtain the following sample regression function:
daily earnings=36+8∗education+2∗education2
^
What is the estimated increase in earnings associated with a one year increase in education?

A) $8 per day
B) $10 per day
C) $12 per day
D) It is impossible to tell without more information.

3. What can you learn from the estimated coefficients from a probit regression?

A) Statistical significance only


B) Magnitude of the effect only
C) Statistical significance and sign only
D) Statistical significance and magnitude only

4. Mozambique's government asks you to evaluate the impact of a cash transfer program on
poverty outcomes. Households with less than 5 acres of land receive cash transfers, while
those with more than 5 acres do not. The government only has data on poverty outcomes one
year after the intervention. Which method would you use to evaluate this program?

A) Regression discontinuity
B) Difference-in-difference
C) Instrumental variables
D) Fixed effects
5. You regress the log of house prices on pollution and obtain the following results:
^
log ⁡(house prices)=5+.2 pollution
How would you interpret the coefficient on pollution?
A) A one unit increase in pollution is associated with a 20 percent increase in house prices.
B) A one unit increase in pollution is associated with a 0.20 percent increase in house prices.
C) A one unit increase in pollution is associated with a 20 percentage point increase in house
prices.
D) A one unit increase in pollution is associated with a 0.20 percentage point increase in
house prices.

6. Which of the following is true of an instrumental variables model?

A) You cannot test for exogeneity of the instrument.


B) You cannot test for relevance of the instrument.
C) An instrument is still valid even if the first stage is insignificant.
D) If you have exogeneity, relevance is unimportant.

7. You regress an individual’s annual earnings on a female dummy, years of education and
the interaction of those two variables. What test should you conduct to determine whether the
association between years of education and earnings varies by gender?

A) F-test on the female dummy, years of education, and the interaction of female and
education
B) F-test on the female dummy
C) T-test on the female dummy
D) T-test on the interaction between female and years of education
8. Massachusetts runs advertisements on state-wide television and radio stations to promote
safe driving and you are asked to evaluate this campaign. Legislators then provide you with
county-level data on accident rates for the year before and the year after the ads ran. The
legislators ask you to run a county fixed effects model to determine the impact of the
program. What should you do?

A) Regress accidents rates on county fixed effects and the interaction of whether the county
received the program and an "after" dummy.
B) Regress accidents rates on county fixed effects, year fixed effects, and the interaction of
whether the county received the program and an "after" dummy.
C) Explain that you cannot run a fixed effects model since you would need more than two
years of data.
D) Explain that you cannot run a fixed effects model since all counties were equally affected
by the program.

9. Why do some people think linear probability models are worse than probits?

A) The coefficients are hard to interpret.


B) You can get predictions greater than 1.
C) You can get predictions greater than 0.
D) They only give you associations, not causal evidence.

10. You have a data set with two time periods (before and after a policy change) and two
groups (one affected by a policy change and one unaffected by the change). How do you test
whether a difference-in-differences estimate of the policy's impact is statistically significant?

A) Compare the affected group's mean outcome in the before period to its mean outcome in
the after period. Use a t-test on that mean difference.
B) Compare the affected group's mean outcome in the after period to the unaffected group's
outcome mean outcome in the after period. Use a t-test on that mean difference.
C) Regress the outcome on an "affected" dummy, an "after" dummy, and the interaction of
those two dummies. Use a t-test on the interaction coefficient.
D) Regress the outcome on an "affected" dummy, an "after" dummy, and the interaction of
those two dummies. Use an F-test on all three coefficients.
PART II – Short Answer

1. (10 points): Whether an individual has a college degree is clearly not random. To
identify the causal impact of a college degree on wages, researchers decide to use
father’s college degree status as an instrument for an individual's college degree
status. Assuming there is a strong correlation between fathers' and children's college
degree statuses, would this instrumental variable approach successfully estimate the
causal effect of college education on wages? Explain. [4 sentences]

2. (10 points): You have data from U.S. cities on public expenditures, whether the
mayor was a Democrat or a Republican, and what fraction of voters in the mayoral
election voted for the Democrat. Assume there are only two candidates, so that the
Democrat wins if the Democratic vote share is higher than 50%. You estimate the
following regression:

Public Expenditures = 0 + 1Democrat + 2DV + u

o Democrat = 1 if the elected mayor is a Democrat, 0 if a Republican.


o DV = vote share for the Democratic candidate.

Which coefficient would estimate the impact of Democratic mayors on public


expenditures? Would that estimate be internally valid? Explain. [4 sentences]
3. (10 points): Recall the difference-in-difference example from class in which
researchers tried to estimate the impact of a higher minimum wage on employment.
New Jersey had increased its minimum wage and Pennsylvania had not. They ran the
following regression

Employment = 0 + 1NJ + 2After + 3 NJ*After + u

and found that the interaction coefficient was positive, suggesting that the minimum
wage increase had actually increased employment in fast food restaurants.

Critic 1 of this study argues that the estimate is biased due to the fact that New
Jersey’s workforce is less educated (and thus more likely to work in low paid jobs)
than Pennsylvania’s workforce. Critic 2 argues that the estimate is biased due to the
fact that Pennsylvania experienced a worse recession than New Jersey around the
time the minimum wage law was passed.

Does either of these critics have a valid argument? Explain. [5 sentences]


PART III - Do Textbooks Affect Child Learning?

In many developing countries, there are not enough textbooks for students. In Busia, Kenya,
for example, about 80% of students were in classrooms with less than one textbook per
student. Researchers tried to understand whether providing additional textbooks would raise
test scores. They first regressed test scores (in z-scores) on a dummy variable for whether a
student owned a textbook. Here are the results:

Table 1. Cross-sectional Retrospective Estimates of Impact of Textbooks on Test Scores

4. (5 points): Interpret the coefficient in the English column [1 sentence].

5. (10 points): Suppose richer children were more likely to own textbooks. Would
these estimates be overestimates or underestimates of the true effect of owning a
textbook? Explain. [4 sentences]
The researchers found that, in the schools studied, there was only one textbook per six
students. They ran a randomized experiment to test the effect of giving many more textbooks
to schools. From the 333 schools in the Busia district, they randomly selected 100 to be
treated. They regressed overall test scores (in z-scores) on whether the school was randomly
given more textbooks. The results are below:

Table 2.

6. (10 points): Interpret the coefficient. Is it statistically significant? What does this
suggest about the impact of textbooks on student learning in Kenya? [4 sentences]

7. (10 points) These findings caused much debate. The Minister of Education argues
that these estimates are not causal because this was not a random sample of schools
in Kenya. How would you respond? [4 sentences]

8. (10 points): A local education advocate argues giving textbooks to the lowest
performing students stops them from dropping out of school (and thus the study).
Would this attrition cause the researchers to overestimate, underestimate or have no
effect on the estimate of the program's true impact? [4 sentences]

The textbooks may have been too hard for the typical student. The researchers divided the
sample into two groups, students with low baseline test scores and students with high
baseline test scores. For each of those two groups, they regressed test scores after the
intervention on a treatment dummy.
Table 3.
Low Performing Students High Performing Students
(1) (2)
Textbook School -.049 .218
(.064) (.096)

Note: Standard errors in parentheses.

9. (10 points): What can you conclude from the regressions in Table 3? What is a
better way to test whether the effect of textbooks differs between the two types of
students? Show the population regression function you would use and explain why
it is better than the table above. [4 sentences]
10. (10 points): You find out afterwards that because Kenya has a diversity of different
languages, the national textbook is in English. As a result, only 15% of students can
read the textbook. Does this affect the internal validity of the results? [4 sentences]

11. (10 points): The Tanzanian government is planning a program to increase textbooks
in classrooms. Based on your previous answers, describe two things you would
want to know in order to evaluate whether this would be a good idea. [4 sentences]

You might also like