What you see is what will change: Evaluative conditioning effects depend on a focus on valence

Abstract

This study investigated whether evaluative conditioning (EC) effects depend on an evaluative focus during the learning phase. An EC effect is a valence change of an originally neutral stimulus (conditioned stimulus or CS) that is due to the former pairing with a positive or negative stimulus (unconditioned stimulus or US). In three experiments, the task focus during the conditioning phase was manipulated. Participants judged CS–US pairings either with respect to their valence or with respect to another stimulus dimension. EC effects on explicit and implicit measures were found when valence was task relevant but not when the non-valent stimulus dimension was task relevant. Two accounts for the valence focus effect are proposed: (1) An additional direct learning of the relation of CS and evaluative responses in the valence focus condition, or (2) a stronger activation of US valence in the valence focus condition compared to the non-valent focus condition.

Keywords:

• Evaluative conditioning
• Affective learning
• Valence focus
• Attitudes
• Goal dependence
• Attention

Acknowledgements

An earlier version of this paper is part of the doctoral dissertation of the first author.

We thank Catharina Casper, Agnes Moors, Wilhelm Hofmann, Michael Olson, and one anonymous reviewer for helpful discussions and comments on an earlier version of the manuscript.

Notes

¹Ferguson and Bargh (2004) showed that goal-relevant stimuli are evaluated more positively than goal-irrelevant stimuli. In this article, however, we are concerned with global differences regarding the readiness with which all kinds of stimuli are evaluated, rather than with differences regarding the valence of specific stimuli that are related to a particular goal.

²It is unclear how it can be explained that a secondary task sometimes leads to increased and sometimes to decreased EC effects. Similar to the diverging results on contingency awareness, these discrepant findings suggest that EC effects might be due to different processes that respond differently to cognitive load.

³Both individual and non-individual stimulus selection procedures are common in evaluative conditioning research. As an individual preselection phase (that is usually accomplished right before the conditioning phase) might bring participants into a valence focus that remains during the conditioning phase, the valence focus hypothesis predicts stronger EC effects after individual stimulus selection. Effects in fact are stronger when USs are selected on an individual basis. This is not the case for individually preselected CSs (Hofmann, De Houwer, Perugini, Baeyens, & Crombez, in press). Effects of preselecting CSs, however, could be influenced by several variables (on the one hand, individual selection of CSs could lead to reduction of error variance but, on the other hand, to a decrease of the effect due to participants” aim to remain consistent) and are therefore difficult to interpret anyway. These explanations do not apply in the same manner to preselecting the USs. The valence focus mechanism, however, can explain the effect of US preselection.

⁴Inclusion of those participants of whom data were available did not lead to different results.

⁵Contingency awareness can also be understood as awareness of the valence of the US the CS was paired with (Stahl & Unkelbach, 2009; Stahl, Unkelbach, & Corneille, 2009). Counting the trials in which the participant selects a US of the correct valence (not necessarily the correct US) gives an estimate of this valence awareness. In this experiment, however, only eight participants ever selected a wrong US that was of the right valence. Thus stimulus awareness and valence awareness are highly correlated. All contingency awareness analyses lead to very similar results. Importantly, valence awareness was also unrelated to the EC effect, β=.01, t(48) = 0.04, ns. Accordingly, there was also no mediation of the valence focus effect by valence awareness.

⁶It has been shown that the influence of contingency awareness on EC effects is best analysed on a within-participant basis by comparing EC effects for remembered pairs with EC effects for non-remembered pairs (Pleyers, Corneille, Luminet, & Yzerbyt, 2007). However, in Experiment 1, there were only four CS–US pairs for each participant and most of them were remembered correctly. Thus only eight participants actually had both positive and negative non-remembered pairs. Only three participants had both remembered and non-remembered positive and negative CS–US pairs. Thus, the within-participant awareness factor does not vary for the majority of the participants. Analysing only remembered pairs shows an EC effect, F(1, 39) = 8.61, p<.01, η² _partial=.18, and an interaction of valence focus and EC, F(1, 39) = 14.18, p<.001, η² _partial=.27. Analysing only non-remembered pairs shows no EC effect, F(1, 6) = 1.64, p=.25, η² _partial=.22, and no interaction of valence focus and EC, F(1, 6) < 1, p=.86, η² _partial=.005. Please note the small power for non-remembered pairs.

⁷The fact that no influence of contingency awareness on EC effects was found in our study should not be interpreted as showing that EC is not influenced by contingency awareness. The general level of contingency awareness was high in our study so that a ceiling of what is necessary for EC might have been reached.

⁸Although the age dimension is not completely valent free, it is clearly less valent than the valence dimension.

⁹Reaction times above 896 ms were more than three interquartile ranges above the third quartile of the response time distribution (“far out values”; Tukey, 1977). Analysing only those responses that were given before the target turned red (≤ 750 ms) led to the same results.

¹⁰This difference is possibly relevant because it could bias the group differences in the EC effect. It does not, however, significantly correlate with the EC effect, either across all participants, or in one of the groups (all r<.26, ns). Therefore, entering US ratings as covariates seemed not warranted. Nevertheless, we performed additional analyses with US-ratings as covariates. These led to the same pattern of results.

¹¹These mediation analyses could unfortunately not be performed on the stimulus-pair level because repeated measures regression analyses require either the calculation of regression coefficients for each participant separately or the entering of N − 1 subject-dummy variables for every predictor and their interactions (Lorch & Myers, 1990). This leads to two problems: (1) Between-subjects effects cannot be calculated because they do not vary within the subjects; and (2) The complete analysis involves a very large number of predictors that almost reaches the number of data points.

¹²The fact that in our studies the EC effect only obtained when participants had the goal to evaluate does in a strict sense not prove that EC effects never appear without the goal to evaluate. It only shows that within the constraints and conditions of our experiment we found such a dependency (see Moors & De Houwer, 2006, for a discussion of this issue).

¹³An additional point is that independently of the process mediating the valence focus effect, it is possible that this mechanism just adds up to a standard EC effect, which by itself might be based on a different process. If, for example, the valence focus effect is based on a CS–ER link then this does not mean that all of the EC effect is based on a CS–ER link; it might be partially based on a CS–US link. Of course, the nature of the process underlying the valence focus effect might be informative about the process underlying the EC effect.