Formats
Format
BibTeX
MARC
MARCXML
DublinCore
EndNote
NLM
RefWorks
RIS

### Abstract

The first chapter of this dissertation provides the most rigorous quasi-experimental evidence to date of the impact of deferred acceptance algorithms on match quality. I worked with Teach For America (TFA) to match high school teachers to schools in Chicago using the DAA at a series of "interview days", while keeping the mechanism for matching elementary school teachers unchanged. I show that TFA's original interview day mechanism promotes strategic early hiring in theory and in practice. I estimate the effect of adopting the DAA using a difference-in-difference strategy - comparing changes in teacher retention rates over time for TFA high school teachers to the change among TFA elementary school teachers in Chicago to the same changes in four other TFA regions. Adopting a variant of the DAA reduces attrition through the start and end of teachers' first school year by 7 and 8 percentage points, respectively. These effects are arguably a lower bound for other markets because substantial heterogeneity in schools' preferences over teachers reduced inefficiency prior to the intervention. Achieving similar retention increases via higher salaries would cost nearly \$3.2 million. The second chapter, which is co-authored with Dr. Sara Heller, uses medium-term results from two randomized controlled trials in Chicago to better understand why summer jobs reduce violence and for whom such programs work best. We see almost identical decreases in arrests for violent crime- equal to 5 arrests per 100 youth- in both the initial and follow up study, but we find no effect on other types of crime. We find limited effects on schooling, including a marginally significant reduction in graduation two years after the program. We find an increase in post-program formal employment with the program providers, but not among other employers. One hypothesis for the program's success is that it targets many youth prior to school exit, acting as unemployment prevention rather than remediation. We use recent developments in supervised machine learning to explore which subgroups' violent crime rates are most responsive to the intervention. We find that criminally involved males who are still enrolled in school but have poor attendance and minimally criminally involved females early in their high school career benefit most. We find that more disconnected males who are older, more criminally involved and do not have recent formal work experience are most adversely affected by the program. As this group is often the target of similar social programs, this has potentially important implications about optimally targeting similar light touch'' social programs. The third chapter, which is co-authored with Dr. B. Pablo Montagnes, develops theory to demonstrate matching problems within an organization are distinct from traditional applications in public markets. Organizational assignment problems are constrained by individual rationality so an organization may select any assignment that is acceptable to its members. We show that there are no guarantees that assignment mechanisms that respect preferences will perform well from an organization's perspective. In some cases, an organization can better achieve its objectives by ignoring preferences and randomly choosing assignments, even when market participants have preferences aligned with the organization's, have outside options, and have private information about match qualities.