0% found this document useful (0 votes)
6 views2 pages

MIT 14.32 Problem Set V: Wage Structure Replication

Uploaded by

Houda Boubaker
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
6 views2 pages

MIT 14.32 Problem Set V: Wage Structure Replication

Uploaded by

Houda Boubaker
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd

Problem Set V

MIT (14.32)
Spring 2007

A. From Wooldridge: 8.1, 8.2, 8.5, C8.5, 10.1, C10.5, 12.1, C12.10

B. Additional problems:

This problem set asks you to replicate Krueger’s 1993 QJE paper, “How Computers
Have Changed the Wage Structure.” The paper can be downloaded from JSTOR. CPS
extracts similar to those Krueger used are posted on the course website in SAS format.
Many variables are constructed already, but you should note that Krueger uses sample
restrictions (e.g. age, work status, maximum/minimum wages allowed in sample, etc.)
that you will need to replicate (see Appendix A of his paper for details, but note that the
$999 vs. $1,923 top-coding issue is already taken care of).

You should attempt to reproduce the results as closely as possible. Note, however, that
you should not expect to get exactly the same results/sample size as Krueger – there are
simply too many choices to make on an empirical research project! But all coefficients
should be of the same sign, significance, and close in magnitude to Krueger’s. For
example, a coefficient of 0.090 with a standard error of (0.02) would be considered
close to a coefficient of 0.098. A coefficient of 0.050 with a standard error of (0.02)
would not be considered close to a coefficient of 0.098.

Include SAS logs and listings when submitting your problem set. Also, organize your
results into three tables with the same layout as the three Krueger tables, and list
Krueger’s results next to your own (i.e. one column of your results, and then the
corresponding Krueger column next to it, and then the next column of your results, and
then next Krueger column, and so on).

Cite as: Joshua Angrist, James Berry, and Emre Kocatulum, course materials for 14.32 Econometrics, Spring 2007. MIT
OpenCourseWare ([Link] Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].
Specific tasks

Replication

1. Reproduce all Table I results, with the exception of the occupation means (the
definition of these variables is not clear). Also note that Krueger’s part-time
variable does not match ours (but everything else should match).
2. Reproduce all Table II results. Everything should match closely except for the
“other race” variable.
3. Reproduce all Table III results. Everything should match reasonably closely.

Explore variations

4. Run the regression in Table II, Column (6) without region dummies. Do the results
change much? What do the numbers in Table I suggest the consequences of this are
likely to be?
5. Is serial correlation an issue in these models? Why or why not? What about
heteroscedasticity? For the results in Table II, columns 1 and 4, report regular and
heteroscedasticity-consistent standard errors.
6. Building upon Krueger’s specification in Column 6, Table II, let the computer-use
effect vary by union status. Test whether it does. Let both the computer-use and
schooling effects vary by union status. Test these interactions for statistical
significance and briefly discuss the results.
7. We have seen in class that blacks may have a higher return to education than
whites. Construct and run a regression model that allows both the computer use and
schooling effects to vary in the two nonwhite race groups (i.e., black vs. white,
other vs. white). Construct F-tests for these differences, one for each race group
and jointly. How might you have expected the computer-use effect to vary by race?

Cite as: Joshua Angrist, James Berry, and Emre Kocatulum, course materials for 14.32 Econometrics, Spring 2007. MIT
OpenCourseWare ([Link] Massachusetts Institute of Technology. Downloaded on [DD Month YYYY].

Common questions

Powered by AI

The inclusion or exclusion of region dummies can affect the consistency and unbiasedness of the regression coefficients. In Krueger’s study, running regression in Table II, Column (6) without region dummies may lead to omitted variable bias if there are regional differences affecting wages that are not captured by other variables .

Differing expectations of the computer-use effect by race in Krueger's study are addressed by allowing computer use and schooling effects to vary between racial groups and testing for significant differences using F-tests. It explores hypotheses that nonwhite groups might experience different effects than whites, based on contextual socioeconomic factors .

Serial correlation can lead to inefficient estimates and biased standard errors, while heteroscedasticity can also lead to inefficient estimates and biased inference. In Krueger's study replication, heteroscedasticity-consistent standard errors were reported for certain results in Table II to address this potential issue .

Testing interactions between computer use and schooling effects by union status presents analytical challenges such as multicollinearity and model complexity. These challenges are addressed by including interaction terms in the regression and using robust standard errors or F-tests to assess significance, ensuring that the models are correctly specified and results are interpretable .

Researchers manage differences between samples and outcomes by carefully reconstructing datasets with similar restrictions as the original, accounting for measurement differences like top-coding, and closely comparing coefficients' sign, significance, and magnitude with the original study, accepting minor deviations due to sample variance .

In Krueger's study, the potential impact of union status on the computer-use effect is examined by interacting computer use with union status in regression models. This interaction is tested for statistical significance using interaction terms in regression analysis, allowing observation of how union status influences the impact of computer use on wages .

Failing to match all variables and conditions exactly in a replication like Krueger's can lead to systematic differences in results, such as discrepancies in the magnitude or significance of coefficients. This can obscure true effects, challenge the replication's credibility, and limit the comparability of findings, emphasizing the importance of precise methodological adherence .

Krueger's study represents broader methodological challenges in empirical research such as sample selection, definition of variables, issues with top-coding, and replication difficulties due to multiple empirical choices. These challenges necessitate meticulous documentation, adjustment for sample differences, and thorough validation through comparison against known results .

Replicating Krueger's 1993 study involves reconstructing the original dataset with necessary sample restrictions such as age and wage limits, and reproducing results across multiple tables. Critical steps include accounting for variables similarly defined in the study, executing the regressions as specified, and comparing replications to the original outcomes in terms of sign, significance, and magnitude of coefficients .

F-tests in Krueger's replication study are used to examine racial differences by testing the hypothesis that coefficients for racial group interactions are significantly different from zero. This helps determine if nonwhite groups experience different returns to computer use than whites, providing a statistically rigorous way to assess group disparities .

You might also like