Empirical Exercise # 4
Problem 1:
The following problem is related to the theory of discrimination, and uses your knowledge of
the Oaxaca Decomposition.
1. Suppose years of schooling, s, is the only variable that affects earnings. After running the
wage regressions separately for men and women we obtain the following:
^
wm 500 100s m
^
w f 300 75s f
We also know that the average level of school among men in the sample is 14 and the average
level among women is 12.
a.) What is the male-female wage differential in the labor market?
b.) The Oaxaca decomposition is a technique that decomposes the raw wage differential into a
portion related to differences in skills and a portion related to labor market discrimination.
The former is measured by looking at differences in the observed characteristics of workers,
and the latter is measured by looking at the differences in the returns (prices) to these
characteristics across workers. Use the Oaxaca decomposition to calculate how much of this
wage differential is due to discrimination.
Problem 2:
This exercise will make use of data on computer use in the workplace. The impact of computer
use was analyzed by Krueger (1993), and DiNardo and Pischke (1997). The data you will use for
this exercise comes from the DiNardo & Pischke paper using German data.
Data: GermanData7985_1.dta
a.) Table I of DiNardo & Pischke documents that the percentage of computer users is increasing
over time. Table II in DiNardo and Pischke suggests that the measured return to computer
use is also increasing over time. Describe a model consistent with these two facts. Describe
a scenario that would predict the opposite of what they find-that is, a decreasing return to
computer use as the number of computer users increases.
b.) Krueger (1993) calculates that a large fraction of the increase in returns to education in the
1980s can be explained by the rise in computer use. Are the returns to computer use higher
for individuals with higher levels of schooling in the German dataset? Reconcile your
findings with Krueger’s calculations using U.S. data.
c.) Run regressions to estimate the return to using computers and other tools on the job, such
as calculators, telephones, pencils, and hammers. Why do these estimates cast doubt on
the interpretation that the computer use wage differential reflects a true return?
d.) One concern with the simple cross-section regressions is that unobserved ability may be
correlated with computer use. Krueger (1993) tries to address this issue by looking at
growth in wages and computer usage at the occupation level. Using 485 occupation
categories, he finds that the difference in average log wages in occupation j is strongly
positively related to the difference in the average computer use in occupation j. Run a
similar regression with German data by using the observations from 1979 and 1985 datasets
which can be matched by occupation number. How do your results compare? What
assumptions must hold for these types of regressions to uncover the causal effect of using a
computer on wages? Note: Since you are using German data, the number of occupations
will differ from Krueger’s.
e.) Investigate whether it is appropriate to treat the composition of occupation groups, except
for computer use, as fixed from 1979 to 1985.
f.) Many of the occupation categories are only observed in one of the years. How many
occupations are in 1979? In 1985? Does the fact that you can only match a selected sub-
sample of occupations cause you to distrust your estimates?
g.) Suppose you are still worried that unobserved ability is correlated with computer use. What
happens when you include grades in German and math classes in a cross sectional wage
regression using the 1979 data? Does this solve the problem?
h.) Would it be a good idea to instrument computer usage with grades? What about
instrumenting computer usage with father’s occupation group? Try running each of these IV
regressions on the 1979 data and comment on the results.
Problem 3:
The following problem is related to the theory of discrimination.
Suppose a firm’s production function is given by q EW E B , where and are the
number of white and black workers employed by the firm respectively. Suppose the market
wage for black workers is $10, the market wage for white workers is $20, and price of each unit
of output is $100. The firm takes both the market wage and product price as given, and incurs
no costs of production other than labor costs.
How many workers (of each race) does the firm hire if it does not discriminate? How much
profit does it earn? How many workers (of each race) does the firm hire, and how much profit
does it make if it discriminates against black workers, such that the discrimination coefficient,
d=0.5? What if d=1.25? Discuss whether the above answers are consistent with Becker’s theory
of taste based discrimination.
