Essays in applied econometrics

This dissertation develops and applies new techniques for the measurement of causal effects in observational data. The first chapter studies the employment effects of the minimum wage using a novel empirical strategy that can identify treatment effects when more than one control group is available, even if all such control groups are imperfect. Expanding on previous research that analyzes labor market outcomes across neighboring regions that experience differential changes to the minimum wage, I compare employment outcomes in border counties where the minimum wage increases to a set of neighboring counties, a set of neighbor-of-neighboring counties, and so on. The key innovation is to model the ratio of the biases present in each of these comparisons. The model I choose uses the relative similarity of control groups to the treated group on observables as a guide to their relative similarity on unobservables. Crucially, models of this type have a testable implication when there are enough control groups. Using data from the United States, I find that recent minimum wage increases have produced modest or zero disemployment effects for teenagers.;The second chapter extends the method developed in the first chapter to the domain of regression discontinuity designs. In particular, I study the effects of party incumbency on margin of victory in United States House of Representatives elections. There are two key findings. First, the selection ratio method developed in the first chapter produces estimates consistent with regression discontinuity design estimates for districts where elections are narrowly decided. Since regression discontinuity designs are widely considered to be credible, I interpret this as evidence in favor of the selection ratio method in a naturalistic setting. Second, I show that the same method can also be used to recover treatment effects for populations away from the discontinuity in a regression discontinuity design. Incumbency effects appear to be approximately stable across the set of elections decided by fewer than 20 percentage points.;The third chapter is concerned with a common research design in economics: instrumental variables. Instrumental variables are widespread in empirical economics, but their use is plagued by concerns that proposed instruments do not satisfy the validity condition. This chapter develops a framework for testing the validity of instruments with covariates, using the motivating example of proximity to college as an instrument for the effect of college attendance on wages (Card 1995). I focus on the case of a binary instrument for a binary treatment, but the approach can be extended to continuous variables. Because this approach makes it possible to quantify the degree of invalidity, it can often allow for point and set identification of treatment effects even when the instrument is invalid. Applying the approach, I find that the proximity to college instrument is invalid and likely overestimates the returns to college.