Lecture delivered by Jennifer Flashman (University of Oxford). Adolescents experience different levels of exposure to individuals of other races. Their exposure may shape their racial preferences for friends in important ways, with serious implications for school integration, bussing, and tracking policies. A small body of work studies the impact of school racial composition on racial preferences for friends using discrete choice models. This work uniformly shows that preferences for friends of a particular racial group decline as the size of that group increases within a school. However, the validity of these estimates rests on the assumption that the odds of choosing one possible friend over another remain constant regardless of the other friend alternatives included in or excluded from the set of possible choices. This assumption is known as the IIA assumption (independence of irrelevant alternatives). Violations of IIA can dramatically affect estimations of individuals? preferences. Given that adolescents have a racial preference for friends, if racially identical friend alternatives are included in the choice set, the preference an individual has for friends of that race are distributed across those identical alternatives. If IIA is violated, choice models will provide an underestimate of preferences for black friends when there are many black students within a school and an overestimate of preferences for black friends when there are few black students within a school. Consequently, results from past research suggesting that blacks have stronger preferences for black friends when there are few blacks in a school may be an artifact of violations of the assumption inherent in the modeling strategy. Through a careful analysis of both simulated and actual data, this presentation provides a corrective to past research on friendship choice by showing 1) that key model assumptions are violated when discrete choice analysis is used to model friendship choice, 2) that results are extremely sensitive to violations of model assumptions, and 3) that after correcting models, estimations show that increased contact between racial groups leads to stronger preferences for cross-race friends.