Longitudinal biases against corrective actions.

SCIENTIFIC Corrective actions are inherently confounded with the problems they are intended to correct. Adverse outcomes of a corrective action are therefore due to an unknown combination of the poor prognosis of the original problem and the effect of the corrective action in modifying that prognosis for better or for worse. Early evaluations of Head Start and job training programs highlighted this problem, but lessons from those examples have not influenced research in psychology as broadly as in economics. An increasing number of statistically controlled longitudinal studies have found adverse effects of corrective actions by parents, including corrective disciplinary actions, homework assistance, and talking to youth against deviant actions and peers. Similar longitudinal analyses of empirically supported corrective actions by professionals often show similar apparently adverse effects. This can be partially explained by Lord’s (1967) paradox, which showed that ANCOVA-type analyses of residualized change scores are biased against corrective actions compared to analyses of simple change scores, as in repeated-measures ANOVA or linear growth models. After summarizing some previous explanations of Lord’s paradox, we show that these biases relative to each other are due to different standards of comparison implicit in the null hypotheses in the two types of change-score analyses. Research is needed to clarify how to approximate valid causal inferences more closely in longitudinal analyses. In the meantime, robust results across both kinds of change scores have been recommended by several experts. (PsycINFO Database Record (c) 2018 APA, all rights reserved)