We use data from the 1993 and 1998 National Surveys of Small Business Finances to examine the existence of racial discrimination in the small-business credit market. We conduct an econometric analysis of loan outcomes by race and find that black-owned small businesses are about twice as likely to be denied credit even after controlling for differences in creditworthiness and other factors. A series of specification checks indicates that this gap is unlikely to be explained by omitted variable bias. These results indicate that the racial disparity in credit availability is likely caused by discrimination.

I use data from Williams College to implement a quasi-experimental empirical strategy aimed at measuring peer effects in academic outcomes. In particular, I use data on individual students' grades, their SAT scores, and the SAT scores of their roommates. I argue that first-year roommates are assigned randomly with respect to academic ability. This allows me to measure differences in grades of high-, medium-, or low-SAT students living with high-, medium-, or low-SAT roommates. With random assignment these estimates would provide compelling estimates of the effect of roommates' academic characteristics on an individual's grades. I also consider the effect of peers at somewhat more aggregated levels. In particular, I consider the effects associated with different academic environments in clusters of rooms that define distinct social units. The results suggest that peer effects are almost always linked more strongly with verbal SAT scores than with math SAT scores. Students in the middle of the SAT distribution may have somewhat worse grades if they share a room with a student who is in the bottom 15% of the verbal SAT distribution. The effects are not large, but are statistically significant in many models.

Data on brothers and on fathers and sons from the National Longitudinal Survey are used to consider the impact of omitted variables and measurement errors on the economic returns to schooling. The analysis suggests that the upward bias in estimated returns due to omitted variables is likely offset by an equal downward bias resulting from measurement errors in reported schooling. Controlling for both of these potential sources of bias yields results comparable to conventional regression estimates of the economic return to schooling.