Examining change using regression analysis: Three approaches compared

Abstract

Although social scientists are frequently faced with the task of assessing the correlates of change, the methods for doing so are often misunderstood. This analysis compares three approaches for dealing with this task using regression procedures: (a) change or difference score (Time2 ‐ Time1 [T2 ‐ T1] as the dependent variable with T1 score included as a covariate; (b) T2 score as the dependent variable and T1, score as a covariate; and (c) residualized T2 score as the dependent variable. All are shown to result in identical interpretations of the relationship between substantive variables and change. We argue, however, that the use of difference scores as the dependent variable is generally preferable (when T1, score is controlled) because the resulting analysis deals directly with adjusted change.