Answers are grounded in this paper and cite their sources.
In examining the effects of teacher or school variables on student outcomes, studies often include a pretest measurement of the outcome variable as a covariate in the model. In multilevel modeling, the covariate can be incorporated either solely at the within level (L1), or at both the within and between level (L2). In practice, this is handled differently across applied studies, and methodological papers provide incoherent recommendations. We provide an overview of central methodological literature identifying various factors that might bias the estimated effect of teaching quality when the pretest is (not) included at L2. These include omitted confounders, contextual effects, and L2-endogeneity. We discuss these factors in light of educational research, showing in a targeted simulation study how the presence or absence of these factors determines whether pretest measures should be included at L2. We further present guidelines for researchers on when to include pretest measures at L2 in their models, exemplifying their application and discussing boundary conditions for our conclusions.
No takes yet. Share an insight, caveat, or question.
Köhler et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: