The impact of instruction- and experience-based evaluative learning on IAT performance: a Quad model perspective

ABSTRACT

Learning procedures such as mere exposure, evaluative conditioning, and approach/avoidance training have been used to establish evaluative responses as measured by the Implicit Association Test (IAT). In this paper, we used the Quad model to disentangle the processes driving IAT responses instantiated by these evaluative learning procedures. Half of the participants experienced one of these three procedures whereas the other half only received instructions about how the procedure would work. Across three experiments (total n = 4231), we examined the extent to which instruction-based versus experience-based evaluative learning impacted Quad estimates of the Activation of evaluative information in IAT responses. Relative to a control condition, both instruction- and experience-based evaluative learning procedures influenced Activation. Moreover, and contrary to what prevailing models of implicit evaluations would predict, in no instance did experience-based procedures influence (positive or negative) Activation more strongly than instruction-based procedures. This was true for analyses which combined procedures and also when testing all three procedures individually. Implications for the processes that mediate evaluative learning effects and the conditions under which those processes operate are discussed.

KEYWORDS:

• Attitudes
• IAT
• Quad modelling
• learning
• implicit evaluations

Disclosure statement

No potential conflict of interest was reported by the authors.

ORCID

Pieter Van Dessel http://orcid.org/0000-0002-3401-780X

Jan De Houwer http://orcid.org/0000-0003-0488-5224

Notes

1 In previous research, the Activation parameter (commonly abbreviated as AC) has been almost exclusively referred to as reflecting the “activation of associations” (e.g. Conrey et al., 2005). However, the Quad model does not speak to the representational nature of what is activated, only that something has been activated. Moreover, past work has never conclusively demonstrated that associations rather than other cognitive representations are captured by this parameter. Consequently, in the present manuscript we describe the Activation parameter using language that does not rely on a priori assumptions about the underlying representational structure. This more conservative approach corresponds more closely to how the Activation parameter is described by Sherman et al. (2008): “the activation of an impulsive response tendency” (p. 316).

2 The Activation parameter does not necessarily have to reflect evaluative information. Instead, it can reflect relationships between concepts and attributes such as stereotypes (e.g. Calanchini, Sherman, Klauer, & Lai, 2014).

3 The data from Experiment 1 are a subset of data from a larger study (Hughes, Van Dessel, Smith, & De Houwer, 2019) which included additional explicit measures and experimental conditions not relevant for the current purposes. The full design of that study is available on the Open Science Framework (see Supplement 1: osf.io/v7y4s). Participants included in the current analyses are those who were in conditions for which the sole manipulation was the evaluative learning procedures described here (e.g. participants were not included who also received counter-attitudinal information before completing the IAT).

4 In all instruction-based conditions, an attention check was used to ensure that participants could accurately recall the instructions, and progression to the next part of the experiment was contingent on a correct response. Subsequent data were included from all participants; those who initially answered this item incorrectly were asked to respond again until a correct response was made.

5 Though IAT research often focuses on D-scores (Greenwald, Nosek, & Banaji, 2003), the current manuscript is focused on analysis of Quad model parameters. As such, for the sake of space and clarity, we do not include D-scores in the main text, but report them for all three experiments in Appendix B.

6 In all three experiments, participants answered five questions (e.g. explicit attitude, confidence) which are not relevant for the current purposes and which we did not analyze. The full text of these items is available at the OSF project page (see Supplement 3: osf.io/v7y4s).

7 Guessing is operationalized in the Quad model as the tendency to select “pleasant” versus “unpleasant” responses. However, Guessing should not be interpreted as a specific cognitive process but, rather, as reflecting any processes that influence responses in addition to Activation, Detection, and Overcoming Bias.

8 The Quad model is structured such that the Overcoming Bias parameter influences responses to target (i.e. Empeya, Vetke) but not attribute (i.e. pleasant, unpleasant) stimuli, and only in the incompatible blocks of the IAT. In contrast, Activation, Detection, and Guessing influence responses to both type of stimuli in both blocks of the IAT. Consequently, the Overcoming Bias parameter is estimated from fewer trials and, thus, less reliably than the other three parameters. As such, the present research should not be interpreted as strong evidence that evaluative learning has absolutely no effect on Overcoming Bias but, instead, that any effects are too small to be reliably detected given the present samples.

9 This is not true of the Guessing parameter, which is anchored at .5 rather than 0. Guessing estimates >.5 reflect a tendency to respond with the “good” key, estimates <.5 reflect a tendency to respond with the “bad” key, and estimates =.5 reflect no evaluative response bias. Consequently, Guessing parameters cannot be interpreted in the same way as the other Quad parameters.

10 The Activation, Detection, and Guessing parameters are specified in the Quad model to influence responses to both target and attribute stimuli in both blocks of the IAT. In contrast, the Overcoming Bias parameter is specified to only influence responses to target stimuli and only in the incompatible blocks of the IAT.

Additional information

Funding

This research was conducted with the support of a postdoctoral research fellowship from the Alexander von Humboldt Foundation to JC, a post-doctoral research fellowship from the Fellowship of the Research Foundation – Flanders (FWO) to PVD, and Bijzonder Onderzoeksfonds from Ghent University to JDH [Grant Number BOF16/MET_V/002].