Fitting regression models to research questions for analyzing change in nonexperimental research

This article discusses the problems associated with analyzing change in nonexperimental data. Three ordinary least squares regression models are presented and empirically tested to illustrate similarities and differences. Model one, based on raw difference change scores, applies best to studying change processes. Model two, based on outcome scores, applies best to assessing the consequences of change. Model three, based on residualized scores, applies best to developing predictive models or interpreting the amount of variance accounted for in the dependent variable. The discussion justifies these different applications and suggests the kinds of social work research questions that best fit with the different models of analysis.