Issues With Cross-Lagged Analyses in Developmental Research: The Case of Reciprocal Effects Between Self-Beliefs and Achievement at School

Résumé En

Developmental processes are a central feature of educational theories. These developmental processes can relate to evolutions across time for a given individual (within-person changes) or to evolutions across time in the differences between individuals (between-person changes). Although theories mostly focus on within-person processes of change, traditional modeling strategies (cross-lagged analyses, residual change analyses) are regarded as informing on between-person processes. This gap between theory and evidence may represent an “ecological fallacy” if between- and within-processes of change prove to be different (Curran & Bauer, 2011). The goal of this communication will be to provide insights on “ecological fallacies” in developmental research in education, and to discuss the meaning of these methods and findings relatively to educational theories. We will illustrate our point by testing the robustness of the Reciprocal Effects Model (REM; Marsh & Craven, 2006). The REM epitomizes a central tenet of educational theories, namely that student motivation develops as a function of positive, reciprocal effects between beliefs about the self (e.g., control, competence, esteem) and achievement at school. Although the REM paradigm focuses on processes of change within the individual, the REM evidence relies on statistical model that reveal processes of change between individuals. To disentangle within- and between-person processes of change in REM paradigm, we ran analyses on a sample of high school students (N > 1900, 6 waves). Using conventional cross-lagged analyses, we replicate REM findings. Using non-conventional cross-lagged analyses that reveal within-processes of change, REM findings are not entirely replicated, i.e., we find no evidence that prior self-perceptions increase subsequent achievement, and we find some evidence for the reciprocal effect. In sum, conventional cross-lagged analyses inform us on developmental processes. However, this information may not always match the theoretical assumptions under investigation. Implications concerning the empirical validity of educational theories will be discussed.