Methodological Challenges in Assessing the Causal Effects of Teaching

In this paper, we focus on three particular challenges in specifying appropriate models that can be used to estimate the causal effects of teaching in nonrandomized designs. First, we clarify that from a causal perspective the ANCOVA and change score approaches address the same research question (i.e., estimating the causal effects of teaching) but rely on different assumptions to identify the causal effects. Second, we argue that the cumulative effects of teaching (over several years) are often underestimated with two-occasion data. Thereby, we also point out the great potential of marginal structural models for analyzing the effects of time-varying treatments. Finally, we briefly discuss the role of measurement error and compositional effects, which we believe deserve further attention in future methodological research.