When and why do implicit measures predict behaviour? Empirical evidence for the moderating role of opportunity, motivation, and process reliance

Abstract

The ability of implicit measures to predict behaviour varies greatly across studies, emphasising the need for accounts of this variability. In order to organise and review the literature on moderators that influence individuals' information processing, we suggest a classification system of moderators with two dimensions. One dimension distinguishes moderators according to their influence on the opportunity to control, the motivation to control, or the reliance on either automatic or controlled processes without changes in opportunity or motivation. The second dimension classifies moderators according to whether they pertain to a disposition of the acting person, the situation in which the behaviour occurs, or the behaviour itself. Increased predictive validity of implicit measures is associated with conditions that foster the impact of automatic processes on behaviour determination. In the discussion we derive several additional moderators from the classification system, delineate emerging research questions, and discuss implications of the reviewed studies for research on self-regulation.

Keywords:

• Attitude–behaviour consistency
• Behaviour
• Implicit measures
• Moderator
• Predictive validity

Acknowledgments

We thank Mareike de Boer, Jan De Houwer, Jane Thompson, Russ Fazio, Arnd Florack, Jochim Hansen, Atilla Höfling, Simon Ineichen, Anita Todd, and Michaela Wänke for helpful discussions and/or valuable comments on an earlier version of this manuscript.

Notes

¹Although this class of measurement techniques enjoys great popularity there is no widespread agreement on what, exactly, the term “implicit” should indicate. Sometimes it is meant to imply that respondents are not aware of what is measured, other times that respondents cannot strategically control the outcome of the measure, and/or that these measures work without intention and work efficiently. These functional properties reflect the criteria of automaticity in the sense of Bargh (1994). Although it is likely that implicit measures fulfil some of these criteria, it is unlikely that there are measures that meet all of these expectations. Consequently, De Houwer (2006) suggested defining “an implicit measure as a measurement outcome that reflects a certain attitude or cognition in an automatic manner, where ‘automatic’ needs to be specified in terms of the presence of one or more functional features” (p. 14; for reviews, see De Houwer, 2006; De Houwer & Moors, 2007; De Houwer, Teige-Mocigemba, Spruyt, & Moors, in press).

²In this chapter we draw on dual-process models to organise the literature on moderators and to derive predictions for the predictive validity of implicit measures. Many of the primary research articles based their arguments on similar theoretical grounds. However, we would like to stress that this review is not intended to be a comprehensive test of the validity of dual-process models. Other models propose a single process (e.g., Kruglanski, Erb, Pierro, Mannetti, & Chun, 2006) or more than two different processes (e.g., Conrey et al., 2005; Sherman et al., 2008) to operate. We do not claim that only dual-process models can account for the empirical results reviewed here, nor is it a goal of this work to test this possibility. However, we do think that dual-process models provide a valuable framework for integration that appears to strike a good balance between explanatory power and parsimony.