|
Abstract:
|
Widespread access to high-quality educational data sets with rich variable pools has made it easier for educational researchers to search for answers to causal research questions with non-randomized data through conditioning strategies. Despite the overwhelming focus on the overall average treatment effect, in many cases the efficacy of educational programs and interventions may vary based on student background characteristics. In this study we use evidence from recent simulation studies relevant to detection and estimation of effect heterogeneity with large-scale education data (Keller, 2021) to inform a case study with the Beginning Postsecondary Students Longitudinal Study (BPS), an NCES-funded data set. In particular, we examine the effect of participation in remedial coursework on community college graduation rate. Two methods are used for detection of effect modifiers: (a) CART-based regression with estimated ITEs (Keller & Chen, 2019) and (b) lasso regression on residualized outcomes with selective inference (Zhao, Small, & Ertefaie, 2018). Conditional average treamtent effects are estimated via BART (Hill, 2011) and causal forests (Wager & Athey, 2018).
|
