Some Ideas for Honors Thesis Projects¶
This page contains a collection of ideas that may be useful starting points for honors theses.
Disclaimer: I haven't thought about these ideas carefully, so they may not be good ideas. Or they may have been done (note the date next to each idea). Or they may not be doable. Or boring, or fundamentally mistaken, or any of the other things that tend to derail research projects.
Don't take the ideas as given. Use them to get started thinking about a set of questions.
A good thesis topic explores cause and effect (e.g., minimum wages affect unemployment) with some clever identification strategy (e.g., instrumental variables). Most of the topics I am listing here don't fit that pattern. They are really just data descriptions. The reason is that I work with structural models. For the questions that I study, getting answers with data only is usually not possible.
Colleges without earnings gains¶
A New York Times article discusses why about half of all colleges appear to produce no earnings gains relative to typical high school graduates.
It would be interesting to look more deeply into the (publicly available) data to figure out:
- how many students do these colleges actually enroll?
- how many are colleges that specialize in professions with low earnings but high amenities (e.g., the arts)?
- who attends these colleges? Are there earnings gains relative to what similar students would have earned without college?
College qualities over time¶
Are college qualities highly persistent over time? Do colleges move around in the quality distribution?
Which colleges move up and why? Is there a mechanism that "rewards" better colleges and forces "worse" colleges to exit?
Drawbacks: this is descriptive.
Hoxby (2009) shows that colleges became more stratified in the 1960s. Her data end in 2006. They are also not publicly available.
Moreover, Hoxby's data show that initially highly selective colleges became more selective over time and vice versa. A different, but related, question is: did colleges become more homogeneous? How did the CDF of college "qualities" change over time? What happened more recently?
Possible data sources: IPEDS (since about 1985) and HERI freshmen surveys.
Drawback: this is descriptive.
Cross-country Income Differences¶
Occupational downgrading of immigrants (2020)¶
Idea: If immigrants from poor countries have less human capital (given schooling), they should be employed in jobs that require less human capital. Those are jobs held by natives with lower schooling.
Construct average native schooling by [occupation, industry].
For each source/host pair: compute the average gap between immigrant and native schooling in [occ, ind] cells. This is a measure of occupational downgrading.
To what extent is the wage gap between immigrants and similar natives explained by this?
Jones (2014) has strong claims about downgrading. How do those hold up?
The task content of immigrant jobs (2020)¶
Todd Schoellman may have done this.
Where do Immigrants do well?¶
... and how has this changed over time? And how about their children?
Related to Abramitzky & Boustan's "Streets of Gold".
Hsieh/Klenow for Immigrants¶
Hsieh et al 2019 show that women and black men were underrepresented in certain occupations in the 1960s. Over time, the gaps diminished, suggesting that the allocation of talent improved.
Is there evidence for a similar convergence among immigrants?
Drawback: this is really just a replication of Hsieh et al for a different population group.
Sources of earnings "shocks" (2021)¶
Administrative data show that earnings "shocks" are asymmetric (frequent small positive and rare large negative shocks).
What observable events are associated with earnings shocks?
What fraction of the "shocks" are due to
- employer changes including layoffs
- occupation changes
- big changes in hours worked
- family events (e.g., having children)
The goal is to inform how one could model earnings shocks (in structural models). There is a recent paper (for which I cannot find a reference) that does something related using administrative data from a Nordic country.
- may be hard to do with publicly available data
How predictable are lifetime earnings? (2020)¶
It is fairly easy to get a lower bound on predictability. Take a panel dataset. Use half to fit a statistical model. Use the other half to perform out of sample prediction.
Specification search is a problem.
Uncertainty about aggregate shocks (basically uncertainty about the relationship between individual characteristics and earnings) are not measured. But the same is true in structural models.
Skill premium variation across U.S. states / cities¶
Dispersion supposedly has decreased. Could one explore empirically possible explanations?
This is very open ended without a clear hypothesis or method.
Giannone, Elisa. n.d. “Skill-Biased Technical Change and Regional Convergence”