Explicit vs. statistical targeting in affirmative action: Theory and evidence from Chicago's exam schools

Prohibitions on using race in affirmative action have spurred a number of admissions systems to adopt race-neutral alternatives that encourage diversity without appearing to explicitly advantage any particular group. The new affirmative action system for Chicago's exam schools reserves seats for students based on their neighborhood and leaves the rest to be assigned via merit. Neighborhoods are divided into four tiers based on an index of socioeconomic disadvantage. At each school, an equal fraction of seats are reserved for each tier. We show that the order in which seats are processed at schools provides an additional lever to explicitly target disadvantaged applicants. We then characterize tier-blind processing rules that do not explicitly discriminate between tiers. Even under these rules, it is possible to favor certain applicants by exploiting the score distribution across tiers, a phenomenon we call statistical targeting. When disadvantaged applicants systematically have lower scores than other applicants, the optimal tier-blind processing order first assigns merit seats and then the tier seats. Our analysis shows that Chicago has been providing an additional boost to applicants from disadvantaged tiers beyond their reserved slots, a benefit comparable to what they received from the 2012 increase in reserve size.