Research Design Meets Market Design: Using Centralized Assignment for Impact Evaluation

New Findings, School Assignment, School Reform, November 2016

A grow­ing num­ber of school dis­tricts use cen­tral­ized assign­ment mech­a­nisms to allo­cate school seats in a man­ner that reflects stu­dent pref­er­ences and school pri­or­i­ties. Many of these assign­ment schemes use lot­ter­ies to ration seats when schools are over­sub­scribed. The result­ing ran­dom assign­ment opens the door to cred­i­ble qua­si-exper­i­men­tal research designs for the eval­u­a­tion of school effec­tive­ness. Yet the ques­tion of how best to sep­a­rate the lot­tery-gen­er­at­ed vari­a­tion inte­gral to such designs from non-ran­dom pref­er­ences and pri­or­i­ties remains open. This paper devel­ops eas­i­ly-imple­ment­ed empir­i­cal strate­gies that ful­ly exploit the ran­dom assign­ment embed­ded in the wide­ly-used deferred accep­tance mech­a­nism and its vari­ants. We use these meth­ods to eval­u­ate char­ter schools in Denver, one of a grow­ing num­ber of dis­tricts that inte­grate char­ter and tra­di­tion­al pub­lic schools in a uni­fied assign­ment sys­tem. The result­ing esti­mates show large achieve­ment gains from char­ter school atten­dance. Our approach expands the scope for impact eval­u­a­tion by max­i­miz­ing the num­ber of stu­dents and schools that can be stud­ied using ran­dom assign­ment. We also show how to use DA to iden­ti­fy causal effects in mod­els with mul­ti­ple school sectors.