Does having legal access to oral contraceptives cause women to get married later? This is a big task, so the work will span two assignments. There are multiple goals for this task:
Understand and use difference-in-differences strategies
Get experience interpreting methods in research papers
Get experience writing out methods and data descriptions
Understand that precision is crucial in scientific writing
Get experience cleaning data
Your will be replicating some of the work in the following report:Goldin, Claudia, and Lawrence Katz (2002) “The Power of the Pill: Oral Contraceptives and Women's Career and Marriage Decisions”, Journal of Political Economy, Vol. 110.Thus you should read that report. You do not need to read all of it because this is not a generic economics course. Instead you should read everything that is necessary to understand the justification for the “identification strategy” and estimation process. In other words, you may ignore the “Frameworks to Understand the Effect of the Pill on Marriage and Career” section. You can also ignore the sections only on professional choices or that use cohort-level regressions (“Career and Marital Status Outcomes: Aggregate Cohort Analysis”). Remember that statistical work is never separate from understanding the qualities of the data and research design.
An “identification strategy” is a way of finding some effect. Specifically, the phrase is often used to refer to ways of getting causal effects from observational data. For instance, in the example with returns to private colleges in the reading, the identification strategywas to match people on the qualities that we believe would impact their future earnings (namely ambition, intelligence, and academic performance). In the Card and Krueger minimum wage paper, the identification strategy was to compare only fast food places near the border in NJ and Pennsylvania under the assumption that they would share markets.
Problem Set 6
Suppose you just regressed age at first marriage on pill use at age 18. What are two potential confounds in this strategy? What is the identification strategy you will be using to get around these confounds (this is like asking how they argue that the legal changes are as good as random)? Why does it work?
What are the methods G&K use to evaluate their hypotheses? What evidence other than a regression table do they provide?
What was the result that G&K found? Are you convinced by it?
Write out the econometric model you will estimate. Explain the reason why each variable is included and what it means. Read their description, internalize it, understand it, and then write out a description. What is the level of observation in this model?
Write out a detailed explanation for exactly what sample you will use. Think of this as a recipe. Someone should be able to follow your recipe to reproduce your work exactly. There should be no ambiguity (for instance, if you just said, "I will use young people", no one would be able to reproduce that sample).
Evaluate this claim: difference-in-differences models solve the problem of omitted variable bias/selection bias.