Identifying predictors of treatment response

Abstract

This article provides a rationale for considering predictors of growth in a treatment group as inadequate to identifying predictors of treatment response. When we interpret predictors of growth in a treatment group as synonymous with predictors of treatment response, we implicitly attribute all of the treated children's growth to the treatment, an untenable assumption under most conditions. We also contend that the use of standard scores in predictors of growth studies does not allow us to differentiate growth from treatment, from growth from other factors. We present two research methodologies that are appropriate methods of identifying predictors of treatment response: (a) single-subject experimental logic utilized to identify the specific participants in which treatment responses (not just growth) were found, combined with follow-up group comparison logic to identify the characteristics on which responders and nonresponders differ, and (b) statistical interactions among child/family/context characteristics and randomly assigned group membership. Principles for selecting potential predictors of treatment response are provided.