ANCOVA versus Change Score for the Analysis of Two-Wave Data

There is a longstanding debate on whether the analysis of covariance (ANCOVA) or change score approach more appropriate when analyzing non-experimental longitudinal data. In this article, we use structural modeling perspective to clarify that ANCOVA based assumption all relevant covariates are measured (i.e., sufficient remove confounding) and provides biased estimates treatment effect in presence unmeasured confounding variables. By contrast, offers option controlling for unobserved confounders but relies strong assumptions about effects these does not allow dynamic causal relationships pretest affects variable). Furthermore, discuss four issues received less attention previous research: (1) inclusion additional approaches, (2) correction measurement error predictor variables, (3) multilevel designs cluster-level variable, (4) estimation heterogeneity nonlinearity. Implications data educational psychological research discussed.