Experimental and Quasi-Experimental Analysis of Discrimination

National Bureau Of Economic Research Inc, Cambridge MA

Investigators

Abstract

The researchers propose projects that use experimental or quasi-experimental methods to study discrimination. In prior work, they sent out resumes in response to help-wanted ads and experimentally varied name (distinctively African-American or White) and skills to measure discrimination. The first project expands on this study and uses a much larger set of African-American, Hispanic and White names that are matched by parental education. It tests whether discrimination persists even when holding constant the social background implied by the name. The second project refines this methodology by exploiting the fact that resumes in India tend to include standardized test scores. This allows a more careful examination of how caste or religion affects employers' interpretation of skills. Caste is also often explicitly stated on the resume so the order of caste revelation can be experimentally manipulated (early or late in the resume), allowing a test of whether order affects the extent of discrimination and the interpretation of skills. The other three projects investigate how well various remedies for discrimination work in practice. One studies the affirmative action system in Indian colleges, where caste-specific test score thresholds determine admission. Because students barely above and below the thresholds are effectively randomly assigned college admission, a regression discontinuity design can be used to estimate the returns to schooling for all castes and measure the program's private benefits and costs. Outcomes for younger siblings of threshold students provide a way to also measure role-model effects. Another project examines whether contact reduces discrimination. It focuses on an MIT program where volunteers tutor poorer high school students over several months. Importantly, volunteers are randomly assigned to students, meaning that some tutors have intense contact with African Americans while others do not. The goal is to survey past tutors and measure their explicit race attitudes and implicit discrimination level. The final project draws on the de-biasing literature in psychology to understand whether different decision making rules can affect the extent of discrimination. It builds on a relationship with a bank in South Africa that wants to experimentally manipulate how its branch managers make loan decisions. The goal is to measure changes in who gets loans and in the profitability of these loans.