You are on page 1of 5

1. A researcher is interested in the effect of a Science & Engineering study (SE) on earnings (Y).

She obtains data from a sample of graduates from university education. X is a vector of control
variables including gender, age, ethnicity, SE=1 for graduates from Science & Engineering
studies, SE=0 for graduates from other studies.

a. The researcher estimates the parameters of the equation


Yi   0  1 SEi   2 X i   i
using OLS. What is the main assumption for getting a consistent estimate of the treatment
effect? Do you think that this assumption holds? Explain why you think that this assumption
holds or does not hold. [2 points]
b. The researcher discovers that S&E studies use a rule for accepting applicants based on the
average exam score in secondary education ( pi ). Those students that scored above a
threshold are allowed to enroll in S&E studies. The variable denoting eligibility for treatment
( Ei ) is defined as follows:

Ei  1 if pi  8

Ei  0 if pi  8

Let Ti denote actual treatment and k ( pi ) denote a polynomial function of pi .


How would you estimate the causal effect of S&E studies on earnings in a situation with perfect
compliance with the acceptance rule? Specify the equation that you would like to estimate. What
are the main assumptions? [5 points]

c. The researcher wants to check the validity of her approach. Which issues should be checked in
applying this approach? [2 points]

Next, the researcher discovers that there is no perfect compliance with the acceptance rule. Some
students that were not eligible for S&E studies managed to enroll and some students that were
eligible did not enroll.
d. How would you proceed to get a consistent estimator of the treatment effect? Specify the
equation(s) that you would like to estimate. Explain why this approach yields a consistent
estimator. What are the main assumptions with this approach and why do you think these
assumptions will hold? Specify the treatment on the treated effect ( E (Y1i  Y0i Ti  1) ) in terms of

the observed outcomes E (Yi Ei  1) and E (Yi Ei  0) and in terms of the observed treatments

E (Ti Ei  1) and E (Ti Ei  0) [6 points]


2. A researcher is interested in the long term effects of military conscription (14 months of
military training). She aims to estimate the causal effect of military conscription on hourly wages
in 2000 using data from a sample of men born between 1955 and 1965. At the age of 18 all
young men were tested on various aspects such as physical fitness, IQ and emotional stability.
The test results are used for the recruitment of young men for military conscription. In addition,
some young men are not recruited because of older brothers that have served in the army. The
researcher only has information on hourly wages, military conscription, date of birth and some
control variables.

a. The researcher estimates the parameters of the equation

ln Wi   0  1M i   2 X i   i

with Wi is hourly wage, M i is dummy for military conscription or not, and X i is a


vector of control variables. Does this yield a consistent estimate of the effect of military
conscription on wages? Explain why you think this estimate is consistent or not.
[2 points]

b. Next, the researcher discovers that the army used a lottery for the recruitment of young
men. She obtains the outcome of the lottery for each individual (coded as a dummy
variable L) and tabulates the results by the lottery results (see Table below). Each cell
contains the mean of the row variable. Hence, 75 % of the lottery winners (L=1)
participated in the army and the mean of their log earnings is 0.6.

L=0 L=1
Participation in army (%) 0.25 0.75
Log earnings 0.40 0.60

Calculate the intention to treat effect (ITT) and the treatment on the treated effect (TOT).
What are the main assumptions for determining the TOT-effect and why do you think
these assumptions will hold. [6 points]
c. In your calculation of the ITT effect and the TOT effect you didn’t use information about
the covariates X. Do you think that this omission could yield a substantial bias for the
estimates? Explain your answer. [2 points]

d. With the first approach the researcher finds a strong negative effect; military conscription
reduces hourly wages with 10 percent. However, the second approach suggest a small
negative effect of 2 percent. Which factors might explain this large difference between
the estimates? [5 points]
Answers Exam Applied Micro-econometrics July 2019
1a cov( SEi ,  i )  0
Not plausible because of selection into type of education (2 points)

1b Yi   0  1 Ei  k ( pi )   2 X i   i (3 points)
Assumptions: all other observed and unobserved factors behave smoothly around cutoff,
density should not change at cutoff (2 points)

1c Smoothness: are covariates similar across the cutoff (1 point)


Density: no spike at cutoff (1 point)

1d Use Ei as instrument for Ti (1 point)


Ti   0  1 Ei  k ( pi )   2 X i   i
Yi   0   1Tˆi  k ( pi )   2 X i   i (2 points)
Assumptions: first stage, independence, exclusion (with explanation) (1 point)
E (Y1i  Y0i T  1)  [ E (Yi E  1)  E (Yi E  0)] /( E (Ti  1 E  1)  ( E (Ti  1 E  0)
(2 points)

2a No, M is endogenous. Selection on unobserved factors might biased the estimates.

2b ITT = 0.6-0.4
TOT = (0.6-0.4) / (0.75-0.25)

Standard IV-assumptions: independence, exclusion, first stage and monotonicity


Likely to hold because of random assignment. Moreover, exclusion restriction likely to
hold, not obvious that losing or winning this lottery might influence future outcomes.

2c. Omitting X will not yield a large bias due to the randomization of L.

2d Two arguments; First estimate probably biased by unobserved factors. Second estimate is
LATE, hence holds for sample of compliers.

You might also like