To accurately estimate implicit influences on health behavior, accurately estimate explicit influences.

Abstract

This project considered how inattention to left-out variable error and measurement correspondence in the assessment of explicit measures can result in upwardly biased estimates of the predictive utility of implicit measures designed to predict health behaviors. A pilot study (n = 96) used a cross-sectional design to predict beer consumption and a main study (n = 132) used a longitudinal design to predict binge drinking. In each study, a battery of 4 implicit inventories (implicit association test, personalized implicit association test, evaluative priming, and attribution misattribution paradigm) and a battery of correspondent explicit measures (based on the Reasoned Action Model and the Prototype Willingness Model) were administered to college youth. The Implicit Association Test and evaluative priming measures were not predictive of alcohol consumption in either study, but the personalized implicit association test (PIAT) and affective misattribution paradigm (AMP) accounted for between 5% and 12% in behavioral criteria, when analyzed in isolation or after explicit measures were statistically controlled following measurement conventions in this research domain. When implicit measures were folded into a structural equation model derived from the Reasoned Action and Prototype Willingness Models, The PIAT was no longer a significant predictor of behavior and the AMP resulted in a 1%-2% incremental increase in accounted for variance. Left-out variable error and measurement correspondence are core principles that need to be considered when modeling the relative contributions of implicit and explicit constructs in the prediction of health behaviors. (PsycINFO Database Record