Similarity-based and rule-based generalisation in the acquisition of attitudes via evaluative conditioning

ABSTRACT

Generalisation in learning means that learning with one particular stimulus influences responding to other novel stimuli. Such generalisation effects have largely been overlooked within research on attitude acquisition via Evaluative Conditioning (i.e. EC effects). In five experiments, we investigated whether and when generalisation of EC effects is based on similarity or on abstract rules. Experiments 1, 2a, 2b and 3 showed that participants who abstracted a rule during the learning phase used that rule for category judgments of novel stimuli. However, evaluative ratings of the same stimuli were unaffected by the learned rule but followed the similarity to learned stimuli. Experiment 4 showed that this similarity-based pattern of generalisation is not specific to evaluative ratings. Rather, resemblance between judgment task and learning task seems to determine whether acquired rules are taken into account. We discuss how dual-process and single-process models of EC may account for the obtained generalisation results.

KEYWORDS:

• Evaluative conditioning
• attitudes
• generalisation

Disclosure statement

No potential conflict of interest was reported by the authors.

ORCID

Christoph Stahl http://orcid.org/0000-0002-9033-894X

Notes

1 As we were unsure about what correlation among repeated measures to assume, we tested values between r = 0.1 and r = 0.5 in steps of 0.1. These analyses showed that we could potentially detect effects as small as f = 0.17 (Experiment 1, 2a, 2b, and 4) and f = 0.14 (Experiment 3). In the text, we report only the most conservative estimates. Note also, that the correlation among repeated measures and the resulting sensitivity will most likely be higher for the categorisation DV than for the evaluation DV because concerning the former, we collected five responses per stimulus per participant and only one per stimulus per participant for the latter.

2 Experiment 2b was conducted last in the present series of experiments, as a result of the revision process of this paper.

3 Due to a programming error, the conditions were not equal in size and they were not fully counterbalanced: We manipulated rule instruction and counterbalanced the order of DVs and assignment of response keys, resulting in eight counterbalancing conditions. We instructed the rule to five instead of four of those conditions. Thus, 74 participants received the rule, 44 did not. Also, among participants who did not receive the rule, we lacked a condition in which participants first underwent the evaluation block and then the categorisation block of the generalisation phase and responded with the key “a” for the reptile and “l” for the mammal category. That exact cell was twice as big among participants who received the rule. We consider the lack of full counterbalancing of minor importance as analyses from Experiment 1, 2a and 2b have not shown any relevant effects of the two variables we counterbalanced. Further, accidentally increasing the rule instruction condition lead to a higher number of rule learners in the sample. We consider this rather unproblematic for two reasons: First, rule learners are the subgroup of interest and more central to our arguments than non-rule learners. Second, non-rule learners constituted the vast majority of the samples in Experiments 1, 2a and 2b which gave ample opportunity to draw inferences about them.