Leveraging Experimental and Observational Evidence to Assess the Generalizability of the Effects of Early Colleges in North Carolina

Abstract

Early college high schools (ECHSs) in North Carolina are small public schools of choice on college campuses that seek to promote attaining postsecondary credits in high school, college readiness, and postsecondary enrollment for underrepresented groups. Evidence from randomized control trials (RCTs) has shown positive effects of the ECHS model on important high school and postsecondary outcomes but appear to be underpowered to detect moderation effects. Furthermore, RCTs rarely address the key question of primary policy interest: Is the program effective on average across the population? This leaves us uncertain about (1) whether the early college intervention is a good strategy for helping to close enrollment and attainment gaps between under- and overrepresented groups, and (2) whether the expansion of the ECHS model will lead to the positive results that the RCT studies suggest. This study uses administrative data on all ECHSs in North Carolina including those that were part of a lottery study. This allows us to generate RCT estimates for the ECHSs in the lottery sample and quasi-experimental estimates for both the lottery and non-lottery ECHSs. We leverage this unique circumstance to generate estimates of the effect of ECHS on postsecondary outcomes that simultaneously maximize both internal and external validity. Specifically, because generalization depends on both moderation and sample selection, we (1) investigate sample selection, (2) conduct a moderation analysis to determine whether the effects of the intervention vary by key factors that also predict sample selection, and (3) produce a pooled estimate by extending a method called cross-design synthesis to incorporate both RCT evidence and quasi-experimental evidence. We find strong evidence that the positive results of the RCTs generalize to the full sample of ECHSs, which provides stronger evidence of effectiveness.