Evaluative Conditioning

Paradigms demonstrating evaluative conditioning (EC) effects (whereby a conditioned stimulus (CS) acquires the positive or negative valence of the unconditioned stimulus (US) it is paired with) remove the autonomy people have in everyday life to sample a stimulus or not. In this line of research, we test the consequences of sampling decisions in EC by allowing participants to decide which CSs to view. For instance, in one study (n=189), participants selected one of several CSs to view with a paired US on each trial. Participants in a yoked condition simply viewed the same information as a participant from the sampling condition. Results show that the EC effect became stronger with more samples in the yoked condition, but in the sampling condition, sampling a CS more frequently led people to like it more, regardless of US valence. As this difference could be due to a lack of autonomy and to a lack of involvement, a second yoked condition held constant the procedure with the sampling condition while removing choice: participants had to actively select the CSs they viewed, but were told which CS to sample on each trial. Results from this condition mirrored the other yoked condition such that the EC effect strengthened with more samples, providing stronger evidence that the difference in the sampling condition is due to making sampling decisions. Thus, this study shows how incorporating sampling into traditional paradigms can offer new insight and highlights the role of autonomy in evaluative learning.

Affective stimuli may either capture visual attention or lead to withdrawal of attention and avoidance behaviour. In the present two experiments (n = 40 and n = 44, respectively), we tested whether learned affective associations lead to attentional biases in terms of a prioritization or inhibition at the level of early visual processing. Therefore, neutral stimuli were paired with either positive, neutral, or negative images using an evaluative conditioning procedure. These conditioned stimuli were then used as targets in an iconic memory task, in which an array of eight stimuli was briefly presented (136 ms), and participants were asked to recall a target after a variable delay. In both experiments, the proportion of correct answers for the successfully conditioned stimuli was lower for negative targets than for neutral and positive targets. It appears that learned negative information can lead to a withdrawal of visual attention, as it was previously reported for unconditioned negative stimuli. It should be further investigated why, in some cases negative targets are prioritized or inhibited, compared to neutral or positive targets.

Evaluative learning can occur via simple stimulus pairings. The contiguity between a conditioned stimulus (CS) and an unconditioned stimulus (US) leads to a transfer of valence onto the CS. As valence constitutes an inherent feature of emotions, we investigate whether expressions of emotions can serve as USs. Further, we ask whether social identification with the USs moderates this effect. In Study 1 (N=94, preregistered), a mono-cultural sample rated CSs before and after the conditioning phase, during which they were paired with neutral, sad, and happy USs representing ingroup and outgroup members. CSs that were paired with happy USs were evaluated more positively than CSs that were paired with sad USs. In Study 2 (N=133, preregistered), a cross-cultural sample rated CSs before and after conditioning. CSs were paired with angry and happy USs representing ingroup and outgroup members. CSs that were paired with happy (angry) USs became more positive (negative) after conditioning. We further assessed memory for the displayed emotion and participants’ identification with the US groups. Memory was better for happy as opposed to angry USs. Although emotion memory was substantial, evaluative changes did not depend on it. Participants more strongly identified with ingroup versus outgroup USs. However, the obtained effects were not moderated by social category of the USs. We discuss theoretical implications for research into emotions and social groups as well as methodological caveats and possible refinements for follow-up studies.

Potential evidence for unconscious evaluative conditioning was reported by Greenwald and De Houwer (2017) from a speeded response-window learning procedure in which pattern-masked meaningless letter strings (CSs, 75 ms) were presented as primes before either pleasant or unpleasant target words (USs), whose valence had to be categorized. During the learning phase, some CSs always predicted positive targets (CSpos), and some always predicted negative targets (CSneg). In the test phase, the CSpos and CSneg primes preceded both pleasant and unpleasant USs (50 % contingency). USs were more often categorized correctly if preceded by a CS with which they had been paired in the learning phase (conditioning / priming effect). Greenwald and De Houwer regressed the conditioning effect onto objective visibility and found that the conditioning effect was observed in the absence of objective visibility (positive intercept) and that it was independent of objective visibility (zero slope). This pattern was interpreted as evidence for the independence of learning from awareness. The present study investigated whether the learning effect is evaluative, by assessing whether it generalizes to typical evaluative measures. First, we largely replicated the overall conditioning effect and the regression results. However, the conditioning effect did not reflect evaluative learning: Null effects were obtained on explicit evaluative ratings and an implicit measure (Affective Misattribution Procedure). This finding is consistent with recent research suggesting that evaluative learning depends on awareness.