The influence of instructions on generalised valence – conditional stimulus instructions after evaluative conditioning update the explicit and implicit evaluations of generalisation stimuli

ABSTRACT

Generalisation in evaluative conditioning occurs when the valence acquired by a conditional stimulus (CS), after repeated pairing with an unconditional stimulus (US), spreads to stimuli that are similar to the CS (generalisation stimuli, GS). CS evaluations can be updated via CS instructions that conflict with prior conditioning (negative conditioning + positive instruction). We examined whether CS instructions can update GS evaluations after conditioning. We used alien stimuli where one alien (CSp) from a fictional group was paired with pleasant US images and another alien (CSu) from a different group was paired with unpleasant US images. The other members from the two groups were used as GSs. After conditioning, participants received negative CSp instructions and positive CSu instructions. In Experiment 1, explicit and implicit GS evaluations were measured before and after the instructions. In Experiment 2, we used a between-participants design where one group received positive/negative CS instructions while a control group received neutral instructions. In both experiments, the positive/negative CS instructions caused a reversal of explicit GS evaluations and an elimination of implicit GS evaluations. The findings suggest that generalised evaluations can change after CS instructions which may have implications for interventions aimed at reducing negative group attitudes.

KEYWORDS:

• Evaluative conditioning
• generalisation
• instruction
• implicit evaluation
• explicit evaluation

Evaluations of people, objects, and situations can be formed through the pairings between a neutral stimulus (conditional stimulus or CS) and a pleasant or unpleasant stimulus (unconditional stimulus or US; De Houwer et al., 2020). For example, when a person (CS) is paired with a negative event (US), the person can acquire the negative valence of that US. This phenomenon, known as evaluative conditioning, is not constrained to the specific CS that was paired with the US. Rather, positive and negative valence acquired via evaluative conditioning can generalise from a single CS to stimuli that share some similarity with the CS but were never paired with the US. For example, negative evaluations of an individual acquired via evaluative conditioning can generalise to other people based on shared features such as race, age, gender, or facial similarities (Kocsor & Bereczkei, 2017; Olson & Fazio, 2006; Spruyt et al., 2014) and is evident in both explicit (self-reported) and implicit (indirectly measured) evaluations (Gawronski & Quinn, 2013; Ranganath & Nosek, 2008). Generalisation of evaluative conditioning can explain how negative evaluations of one individual can lead to widespread prejudice towards an entire group.

Generalisation can occur based on the physical similarities between group members. For instance, Kocsor and Bereczkei (2017) paired pictures of neutral faces (CS) with pleasant or unpleasant pictures (US), such that those paired with pleasant pictures (CSp) acquired positive valence, whereas those paired with unpleasant pictures (CSu) acquired negative valence. These evaluations generalised to novel composite faces (generalisation stimuli or GS) that shared physical similarities with either the CSp or CSu (e.g. identical eyes or nose). Glaser and Kuchenbrandt (2017) demonstrated that generalisation occurs across alien groups, where members within each group shared physical characteristics (e.g. head shape). One alien was paired with pleasant pictures while another was paired with unpleasant pictures. The positive and negative valence acquired via evaluative conditioning generalised to other aliens from the same group, but only when participants were aware of the groups prior to evaluative conditioning.

Evaluations can also generalise after instructions, as demonstrated by Luck et al. (2021). The researchers examined generalisation across categorically similar stimuli (e.g. vegetables and furniture) after providing contingency instructions only. Instead of viewing CS-US pairings, participants were presented with the CSp (e.g. a picture of a carrot) along with the instructions “This image will be paired with pleasant images”, while the CSu (e.g. a picture of a chair) was paired with the instructions “This image will be paired with unpleasant images”. Evaluations generalised from the CSp and CSu to other stimuli from the same category, which was evident in both an explicit (self-reported pleasantness ratings) and implicit (affective priming task) valence measure. This study only examined generalisation via instructions before conditioning and therefore, we do not know whether and how evaluations would generalise if CS instructions were presented after conditioning that conflicts with the instructions (i.e. negative instructions after positive conditioning).

The evidence for the capacity of instructions to change CS evaluations after evaluative conditioning is mixed. For instance, instructed counterconditioning is aimed at changing the CS-US contingency after evaluative conditioning (e.g. instructing participants that the CSp will now be paired with unpleasant pictures and the CSu will now be paired with pleasant pictures). These instructions cause a reversal of explicit evaluations but are less effective in shifting implicit evaluations (Gast & De Houwer, 2013; Hu et al., 2017). On the other hand, behavioural instructions about a social target that are highly diagnostic (that is, highly reflective of a person’s character) cause a reversal of explicit evaluations and have proven reliable in eliminating negative implicit CS evaluations. For instance, Mann et al. (2020) presented a male face (CS) paired with an aversive scream (US), which caused explicit and implicit evaluations of the CS to become negative. The CS was then paired with positive behavioural information that was highly diagnostic (e.g. “This man rushed into a burning house to save a baby … ”) which reversed explicit CS evaluations and eliminated negative implicit evaluations. This finding was replicated across four experiments, where in each experiment negative implicit evaluations were eliminated by the instructions (i.e. no significant effect was observed in the implicit measure after the instructions). However, when the data from all four experiments were pooled together in one highly powered analysis (n = 1503), the positive instructions caused a reversal of implicit evaluations.

It is promising that evaluations of an individual acquired via evaluative conditioning can be updated by presenting conflicting information about that person. However, it has not yet been examined whether generalised evaluations would update in the same way. For instance, we know that when an individual is paired with an aversive US, the resulting negative evaluations will generalise to other members of the same group. But, if positive information about that individual was presented, would the negative evaluations of the other group members update in line with this new information, or would evaluations of other group members remain negative? The Lateral Attitude Change (LAC) model (Glaser et al., 2015) proposes that generalisation effects occur when evaluative changes towards a focal stimulus (in this case, the CS) transfer to other related objects. According to the LAC model, implicit evaluations should generalise from the CS to related stimuli, but explicit evaluations may resist generalisation if the individual has reason not to generalise (i.e. if it seems illogical). It is unclear whether evaluations that have already generalised after evaluative conditioning (i.e. the individual has already evaluated the members belonging to the CS group as positive or negative), would then change in response to new information about the specific CS that is inconsistent with prior conditioning. In this case, we would expect that evaluations of other group members would also update in line with the new CS information as generalisation should be based on the valence of the CS.

There are competing theories that explain the mechanisms underlying how evaluations are formed in evaluative conditioning (e.g. propositional processes vs. associative processes; for a review see Hofmann et al., 2010), but these theories cannot predict how stimulus evaluations will generalise and whether generalised evaluations can be changed. For instance, these theories cannot tell us whether changes in GS evaluations will always mirror changes in CS evaluations, or whether GS evaluations are independent of changes in CS evaluations after initial generalisation has occurred. According to Glaser et al. (2015), generalisation may involve both associative processes (in which changes in CS evaluations automatically activate the same evaluation of similar stimuli) and propositional processes (in which the individual considers whether the generalisation is valid or not), but there is currently no evidence that allows us to draw any conclusions about what processes may underlie generalisation. In the current study, we do not aim to disentangle whether propositional or associative processes underly generalisation. Rather we aim to investigate how generalisation responds to changes in CS valence. It is important to understand more about how generalisation responds to different manipulations in order to generate a theoretical understanding of generalisation in evaluative conditioning so we can predict how evaluations will transfer across different contexts and situations.

In the current study, we investigated whether presenting diagnostic behavioural instructions about a fictitious alien (CS), that conflicted with prior evaluative conditioning would influence evaluations of other aliens (GSs) from the same group. We adopted a similar design to Mann et al. (2020) where behavioural information about the CSs was presented after evaluative conditioning. We conducted two studies using groups of fictitious aliens as the CSs and GSs. In both experiments, a single alien (CSp) from one group was paired with pleasant US images while a single alien from a second group was paired with unpleasant US images. The other members from the CSp and CSu groups were used as GSs to assess generalisation. To counteract the evaluative conditioning, the CSp was paired with negative information (e.g. “this alien was arrested for animal cruelty”), while the CSu was paired with positive information (e.g. “this alien rescued a baby from a fire”). To maximise the chance of updating implicit GS evaluations, we used information that was extreme and highly diagnostic of the alien’s character. GS evaluations were assessed explicitly, using self-report likability ratings, and implicitly, using an affective priming task (Fazio & Olson, 2003). In Experiment 1, we used a pre–post design where implicit GS evaluations were assessed twice – once after acquisition and again after the CS instructions were presented. In Experiment 2, we used a between-participants design where one group received positive instructions about the CSu and negative instructions about the CSp, while a second group (control) received neutral instructions about the CSp and CSu. Implicit GS valence was assessed only once after the CS instructions to reduce any effects of repetition on affective priming.

Experiment 1

To investigate whether CS instructions can update implicit GS evaluations after evaluative conditioning, a single alien from one of four groups was used as the CSp and paired with pleasant USs during acquisition, while another alien from a different group was used as the CSu and paired with unpleasant USs. We then provided participants with information about the CSs that was incongruent with the prior conditioning (the CSp was paired with negative information, and the CSu was paired with positive information). We measured generalisation of evaluations to the other group members at post-acquisition and post-instruction using both explicit (likability ratings) and implicit (affective priming) measures of GS evaluations. Previous research examining generalisation across fictional alien groups found that evaluations only generalised when participants were aware of the groups prior to conditioning (Glaser & Kuchenbrandt, 2017). Therefore, to ensure acquisition learning generalised, participants were informed about the two groups before conditioning.

We hypothesised that acquisition training with single aliens would generalise to other group members. Based on the findings from Mann et al. (2020), we hypothesised that the CS instructions would be successful in updating CS evaluations. We also predicted that updating the valence of the CS would also change GS evaluations. More specifically, we predicted that the instructions would eliminate (but not reverse) implicit GS evaluations, as indicated by a significant priming effect after acquisition followed by no priming effect after the CS instructions. We also predicted that the CS instructions would cause a reversal of explicit GS evaluations, such that the GSp would become unpleasant and the GSu would become pleasant.

Method

Participants. 252 participants (139 male; 101 female; 12 missing data) from the United States aged between 21 and 71 (M = 39.00, SD = 10.74) were recruited via Amazon’s Mechanical Turk (Buhrmester et al., 2011) and were compensated $4.50 (USD). The research protocol was approved by the Curtin University Human Research Ethics Committee. An a-priori power analysis conducted using G*Power (Faul et al., 2007) revealed that 199 participants would be required to be 80% confident of detecting a small effect (f = .2) for the critical GS × Target Valence × Phase interaction at an alpha of .05. The power analysis was conducted based on an ANOVA F test (within-subjects interaction; 2 measurement points [post-acquisition, post-instruction]; Effect size specification: as in Cohen (1988); nonsphericity correction: 1). We aimed to recruit 250 participants to account for error and attrition in the two priming tasks. Seventeen participants left the experiment before the second priming task, and 31 participants were excluded due to making more than 30% of error during priming. This resulted in 204 participants in the implicit evaluations analyses and 252 participants in the explicit evaluations analyses.

Apparatus/Stimuli. The experiment was programmed using Inquisit 4 software.

Explicit Measures. Explicit stimulus evaluations were measured using a 9-point Likert scale. CSs and GSs were evaluated on likability (1 = very unlikable, 9 = very likable) and USs were evaluated on pleasantness (1 = very unpleasant, 9 = very pleasant).

Implicit Measures. Implicit GS evaluations were measured using an affective priming task (Fazio & Olson, 2003). In this task, the three GSp and three GSu were presented as primes at a size of 40% of the participant’s screen followed by a positive or negative target word. Target words included 10 positive words (pleasant, good, outstanding, beautiful, magnificent, marvellous, excellent, appealing, delightful, and nice) and 10 negative words (unpleasant, bad, horrible, miserable, hideous, dreadful, painful, repulsive, awful, and ugly). Target words were presented in white font at a size of 8% of the participant’s screen. All primes and target words were presented on a black screen with a 500 ms inter-trial interval.

Conditioning Stimuli. The CSs and GSs were images of fictitious aliens created using Adobe Photoshop 2017. We created four alien groups (four aliens per group, 16 aliens in total; see Figure 1 for alien stimuli). We used fictitious groups to prevent any pre-existing evaluations of real-life groups from influencing the conditioning procedure. Each alien group was distinguishable by head shape (rectangle, pentagon, oval, or triangle) and skin colour (red, blue, yellow, or green). Body colour and head shape were identical within each group, however, the head colour varied slightly in shade across group members. The members of each group also had varied hair colour/style, eye shape, and body shape so that they were recognisable as separate individuals. A single alien from one group served as the CSp (paired with pleasant pictures during acquisition) while a single alien from a second group served as the CSu (paired with unpleasant pictures during acquisition). The remaining three aliens from each of the CSp and CSu groups were used as the GSs to measure generalisation in the explicit and implicit valence assessments. Which two groups the participants viewed, which group was used as the CSp and CSu group, and which aliens from the groups were used as the CSs and GSs was counterbalanced across participants. To ensure there were no differences in a-priori stimulus valence, we conducted a separate pilot study where participants (Pilot 1; n = 223) evaluated all sixteen aliens. The mean evaluations of the four alien groups were close to neutral, with minor differences in between groups (See Table S1 in the supplement for mean evaluations of all groups and individual aliens). The full methods and results of the pilot study are reported in the supplement. In the current experiment, all CSs and GSs were presented at a size of 50% of the participant’s screen against a black background. The USs included six pleasant pictures (mean valence rating = 8.14; mean arousal rating = 5.05; subject matter: happy family, animals, nature scenery; codes: 1460, 1710, 2154, 2340, 5825, 5833) and six unpleasant pictures (mean valence rating = 2.11; mean arousal rating = 5.17; subject matter: rubbish, starving children, cemetery scenes, sick animals; codes: 9560, 9340, 9295, 9220, 2800, 2703) taken from the International Affective Picture System (IAPS; Lang et al., 2008) and presented at a size of 60% of the participant’s screen.

Figure 1. Alien Stimuli.

Instructions. The positive and negative CS instructions were short vignettes. Each vignette stated the alien’s name (either “Zop” or “Bim”) followed by the name of their home planet (either “Emer” or “Anite”), and a short description of the alien’s past behaviour (either positive or negative). We created six instructions overall (three positive and three negative), two of which were adapted from Mann et al. (2020) which were successful in updating implicit CS evaluations after evaluative conditioning (See Appendix A for all instructions). We conducted a pilot study (Pilot 1; n = 223) to ensure the positive and negative instructions were successful in instilling positive and negative evaluations of the alien stimuli. All positive and negative instructions successfully caused the aliens to be evaluated as likable and unlikable, respectively (the full procedure and results of this pilot are reported in the supplementary materials). The CSp was always paired with one of the three negative instructions, while the CSu was always paired with one of the three positive instructions. Which two instructions the participant received, which name was given to the CSp and CSu, and which planet name was used for the CSp and CSu was counterbalanced across participants.

Procedure. At the beginning of the experiment participants were instructed that the aliens were from two separate groups. All four aliens from the CSp and CSu group were presented separately above the text “These are aliens from the planet Anite/Emer”. Participants then completed baseline explicit evaluations of the CSp and CSu, followed by the three GSp and three GSu. Before the acquisition training began, they were informed they would view a series of pictures and were instructed to watch the screen and pay attention. Acquisition training consisted of 18 CSp-pleasant US pairings and 18 CSu-unpleasant US pairings (36 trials in total; randomised). Within each trial, the CS was presented for 2s and was immediately followed by a pleasant or unpleasant US for 2s, with a 2s inter-trial interval (black screen). After acquisition, participants evaluated the CSs followed by all GSs. In the affective priming task, a white fixation cross was presented for 500 ms followed immediately by the prime (GSp or GSu) for 200 ms, and the target word which remained on screen until a response was made. Participants were instructed to press “I” for positive words and “E” for negative words, and these instructions remained on screen throughout the task. Incorrect responses resulted in a red “X” displayed on screen (500 ms). Participants completed 20 practice trials before the main task. The main task consisted of 120 trials where each GS was paired with each target word once, resulting in four possible pairings (congruent: GSp-pleasant target, GSu-unpleasant target; incongruent: GSp-unpleasant target; GSu-pleasant target). Participants were then informed that they would be presented with important information about the aliens and were instructed to read each piece of information carefully. The CSp was presented above the negative instruction, and the CSu was presented above the positive instruction (randomised order). Participants clicked a “Next” button when they had finished reading each instruction. The text “You will now receive a repeat of the information. Please pay attention” was displayed on the screen and the CSp and CSu instructions were repeated to ensure participants paid attention to the instructions. Participants then evaluated the CSs and GSs and completed a second affective priming task. This task was identical to the first priming task with the exclusion of the 20 practice trials. Participants then evaluated all 12 USs (6 pleasant, 6 unpleasant; randomised). Participants then completed a contingency assessment in which the four aliens from the CSp and CSu groups were presented, and participants were asked to select which alien was paired with pleasant and unpleasant pictures during the first part of the experiment.¹ Participants provided demographic information (age, sex, and ethnicity) before exiting the study.

Data Preparation and Analyses. Evaluations of the three GSp and three GSu were averaged into overall GSp and GSu scores and used in all explicit evaluation analyses. Reaction times and errors for the three GSp and three GSu were averaged into total scores and these scores were used in the priming analyses. Reaction times from trials where the participant incorrectly categorised the target word or had a reaction time slower than 1000 ms were scored as errors. Participants who made more than 30% of error during priming were removed from the analyses (31 participants). All statistics were conducted with IBM SPSS Statistics 27 (alpha = .05; Pillai’s trace statistics of the multivariate solution). The raw data, stimulus materials, and analysis scripts are available at https://osf.io/akqbw/?view_only = 7d3c33a6e02c444aa530180acf015608

Results

Preliminary analyses

US Evaluations. A paired samples t test confirmed that mean evaluations of pleasant USs (M = 7.93, SD = 0.90) were more pleasant than the mean evaluations of the unpleasant USs (M = 2.06, SD = 1.30), t(248) = 48.16, p < .001, d = 3.052.

Contingency Awareness. 31 participants (12.1%) were able to verbalise the contingencies at the post-experimental contingency assessment. 130 (50.8%) selected stimuli from the same group as the correct CSs, even if they were not able to identify the specific aliens that were paired with the pleasant and unpleasant USs. The large percentage of participants who failed the contingency assessment is likely due to the difficulty in recalling acquisition training after reading the CS information and completing the two priming tasks. Due to the large number of participants who failed the contingency awareness assessment, we did not examine whether excluding these participants would change the pattern of results as excluding these participants would result in a large reduction of power.

Explicit evaluations

Explicit CS and GS evaluations are displayed in Figures 2 and 3, respectively, and were analysed with separate 2 CS/GS Type (CSp/GSp, CSu/GSu) × 3 Phase (baseline, post-acquisition, post-instruction) repeated measures ANOVAs.

Figure 2. Mean CS evaluations measured at baseline, acquisition, and post-instruction in Experiment 1. Error bars represent the standard error of the mean.

Figure 3. Mean GS evaluations measured at baseline, acquisition, and post-instruction in Experiment 1. Error bars represent the standard error of the mean.

CS Evaluations. Main effects of CS, F(1, 251) = 71.49, p < .001, ηp² = .222, and phase, F(1, 251) = 13.27, p < .001, ηp² = .096, were moderated by a CS × Phase interaction, F(2, 250) = 331.59, p < .001, ηp² = .726. Follow-up analyses of the interaction revealed no difference in CS evaluations at baseline, F(1, 251) = 0.19, p = .662, ηp² = .001. After acquisition, the CSp was evaluated as more pleasant than the CSu, F(1, 251) = 229.49, p < .001, ηp² = .487, which reversed after the instruction manipulation, such that the CSu was evaluated as more pleasant and the CSp, F(1, 251) = 578.59, p < .001, ηp² = .697.

GS Evaluations. A main effect of phase, F(2, 250) = 8.24, p < .001, ηp² = .062, was moderated by a GS × Phase interaction, F(2, 250) = 95.56, p < .001, ηp² = .433. Follow-up analyses of the GS × Phase interaction revealed no difference between GSs at baseline, F(1, 251) = 0.20, p = .655, ηp² = .001. The GSp were evaluated as more pleasant than the GSu after acquisition, F(1, 251) = 131.95, p < .001, ηp² = .345, which reversed after the instructions, such that the GSu were evaluated as more pleasant than the GSp, F(1, 251) = 70.58, p < .001, ηp² = .219. The main effect of GS did not reach significance, F(1, 251) = 1.25, p = .264, ηp² = .005.

CS Evaluations vs. GS Evaluations. We compared the size of the differential evaluations between the CSs and GSs following acquisition and the instructions to examine whether participants could differentiate the CSs from the GSs. The differential evaluations were calculated by subtracting the CSu/GSu score from the CSp/GSp score. A larger differential score indicates a larger difference between CSp/GSp and CSu/GSu evaluations. A positive differential score indicates that CSp/GSp was evaluated as more pleasant than the CSu/GSu and a negative differential score indicates the CSp/GSp was evaluated as more unpleasant than the CSu/GSu. Separate paired samples t tests were used to compare differential CS evaluations with differential GS evaluations at baseline, acquisition, and after the instructions. At baseline there was no difference between differential CS and GS evaluations, t(251) = 0.12, p = .915, d = 0.007. After acquisition, differential CS evaluations (M = 2.81, SD = 2.94) were larger than differential GS evaluations (M = 1.68, SD = 2.33), t(251) = 8.71, p < .001, d = 0.548. After the instructions, differential CS evaluations (M = −5.31, SD = 3.50) were larger than differential GS evaluations (M = −1.37, SD = 2.58), t(251) = 18.12, p < .001, d = 1.141.

Implicit evaluations

Reaction times were subjected to a 2 GS (GSp, GSu) × 2 Target Valence (pleasant, unpleasant) × 2 Phase (post-acquisition, post-instruction) within-participants repeated measures ANOVA. Tests of normality indicated that the reaction time data violated assumptions of normality, therefore reaction times were inverted, as recommended by Ratcliff (1993). The figures and priming scores show untransformed RTs for ease of comprehension. Mean RTs are displayed in Figure 4. Main effects of phase, F(1, 203) = 13.10, p < .001, ηp² = .061, and target valence, F(1, 203) = 54.72, p < .001, ηp² = .212, and an interaction of GS × Target Valence, F(1, 203) = 6.21, p = .013, ηp² = .030, were moderated by a GS × Target Valence × Phase interaction, F(1, 203) = 5.02, p = .026, ηp² = .024. The GS × Target Valence × Phase interaction was followed up by calculating priming scores ([incongruent: GSp-unpleasant target + GSu-pleasant target] – [congruent: CSp-pleasant target + GSu-unpleasant target]), which were significantly larger post-acquisition (M = 8.94, SD = 40.30) compared to post-instruction (M = 1.62, SD = 43.07), t(203) = 2.24, p = .026, d = 0.157. At post-acquisition, the priming score was larger than zero, t(203) = 3.52, p = .001, d = 0.246, indicating a difference in GS valence. At post-instruction, the priming score was not significantly different from zero, t(203) = 0.59, p = .556, d = 0.041, indicating that a difference in GS valence was no longer present.

Figure 4. Mean RT for congruent pairings (GSp + positive word; GSu + negative word) and incongruent pairings (GSp + negative word; GSu + positive word) measured at post-acquisition (Time 1) and post-instruction (Time 2) in Experiment 1. Error bars represent the standard error of the mean.

Discussion

In Experiment 1, we examined whether instructions about the CS would update implicit GS evaluations after evaluative conditioning. As predicted, EC effects generalised for both the explicit and implicit measures where the GSp were evaluated as more likable than the GSu. The CS instructions caused a reversal of explicit GS evaluations, such that the GSu was now evaluated as more likable than the GSp. The priming effect observed after acquisition was reduced and no longer present after the instructions, indicating no difference in the implicit evaluations of the GSp and GSu. It is possible, however, that we did not see differential evaluations in the implicit measure after the instruction manipulation because of the nature of the pre–post design. Repeating the priming task in a short period of time could have reduced its sensitivity due to floor effects or the repeated presentation of the GSs in the first priming task could have acted as extinction training as the GSs were presented equally often with positive and negative word USs.

Experiment 2

In Experiment 2, we used a between-participants design to investigate whether instructions about the CSs can update explicit and implicit GS valence after evaluative conditioning. After the EC procedure, one group received negative information about the CSp and positive information about the CSu (valence instruction group), while the second group received neutral information about both CSs (neutral instruction group). Implicit GS evaluations were measured once after the CS information was provided using an affective priming task. We hypothesised that there would be a difference in implicit evaluations between groups, where the neutral instruction group would evaluate the GSp as pleasant and the GSu as unpleasant, but the valence instruction group would not evaluate the GSp and GSu differently. To avoid the high number of contingency fails observed in Experiment 1, contingency awareness was measured after acquisition using explicit US expectancy ratings. It has been suggested that assessing contingency immediately after acquisition provides a more accurate measure of contingency awareness, as responses are less likely to be affected by a decline in memory (Shanks & St. John, 1994).

Method

Participants. 847 participants (467 male; 364 female; 16 missing data) from the United States aged between 17 and 76 (M = 38.85, SD = 11.47) were recruited from Amazon’s Mechanical Turk and were compensated $2.80 (USD). The research was approved by the Curtin University Human Research Ethics Committee. Participants were randomly assigned to the valence instruction or neutral instruction condition. An a-priori power analysis using G*Power indicated that a sample of 788 participants would be required to detect a small effect (d = .2) with 80% power using an independent samples t test at an alpha of .05. We recruited 850 participants to account for error and attrition in the priming data. Ten participants left the study before the affective priming task, and 145 were excluded due to making more than 30% errors. This resulted in 690 (valence instruction group: 350; neutral instruction group: 340) participants in the implicit evaluations analyses and 847 (valence instruction group: 423; neutral instruction group: 424) participants in the explicit evaluations analyses.

Apparatus/Stimuli

Explicit Measures. CS, GS, and US evaluations were measured as per Experiment 1. US expectancy was measured with a 7-point Likert scale where participants are asked to predict which pictures will follow the CS (1 = unpleasant pictures will always follow, 4 = no pictures will follow, 7 = pleasant pictures will always follow).

Implicit Measures. Implicit GS valence was measured as per Experiment 1.

Conditioning Stimuli. All CS, GS, and US stimuli were identical to Experiment 1.

Instructions. The positive and negative instructions were identical to Experiment 1. We created six neutral instructions, which did not state the alien’s name as we hypothesised that creating a name and identity for the alien may increase their likability. Instead, the neutral instructions only stated the alien’s home planet (i.e. “Anite” or “Emer”) followed by a short description of the planet (i.e. weather, terrain, location; See Appendix B). We conducted a second pilot study (Pilot 2; n = 225) with the six neutral instructions to ensure the neutral instructions did not cause any considerable changes in alien evaluations. Evaluations of the aliens paired with the neutral instructions became slightly more pleasant from pre-instruction (M = 5.40, SD = 1.50) to post-instruction (M = 5.91, SD = 1.40). Although this difference was statistically significant, it was numerically small compared to evaluative changes instilled by the positive and negative instructions, therefore we deemed these neutral instructions suitable as the control condition. The full procedure and results of this pilot are reported in the supplementary materials.

Procedure. The baseline evaluations, acquisition procedure, and post-acquisition evaluations were identical to Experiment 1. After participants evaluated the CSs and GSs after acquisition, they also completed a US expectancy assessment for the CSp and CSu (randomised). The valence instruction group were then presented with the CSp with the negative instruction, and the CSu with the positive instruction (as per Experiment 1) while the neutral instruction group received the CSp and CSu with two different neutral instructions. Both groups received a repeat of the instructions. Participants then evaluated the CSs and GSs and completed the affective priming task (identical to the first priming task in Experiment 1). Participants evaluated all 12 USs and provided demographic information.

Data Preparation and Analyses. The data were prepared and analysed as per Experiment 1.

Results

Preliminary analyses

Age did not differ between the valence instruction group (M = 39.27, SD = 11.51) and the neutral instruction group (M = 38.42, SD = 11.43), t(821) = 1.07, p = .286, d = 0.091. A chi-square test for goodness of fit (with α = .05) revealed no group differences in sex ratios between groups (valence instruction group: 227 male, 189 female; neutral instruction group: 240 male, 175 female), χ² (2, N = 844) = 0.90 p = .343.

Contingency Awareness. Participants with US expectancy ratings above 4 for the CSp, and below 4 for the CSu were considered aware of the experimental contingencies. 560 participants (66.1%) indicated contingency awareness after acquisition. This did not differ between groups, χ²(1, N = 847) = 1.13, p = .287. Excluding participants who were not aware of the experimental contingencies did not alter the pattern of results, therefore the entire sample is included.

US Evaluations. US evaluations were subjected to a 2 US type (pleasant, unpleasant) × 2 Group (valence instruction group, neutral instruction group) mixed model ANOVA. A main effect of US was detected, F(1, 831) = 5496.58, p < .001, ηp² = .869, revealing that the pleasant USs (M = 7.94, SD = 0.96) were evaluated as more pleasant than the unpleasant USs (M = 2.24, SD = 1.68). The US × Group interaction did not reach significance, F(1, 831) = 3.37, p = .067, ηp² = .004.²

Explicit evaluations

Explicit CS and GS evaluations are displayed in Figures 5 and 6, respectively, and were analysed with separate 2 Group (valence instruction group, neutral instruction group) × 2 CS/GS Type (CSp/GSp, CSu/GSu) × 3 Phase (baseline, post-acquisition, post-instruction) mixed model ANOVAs. The GSp and GSu represent the mean evaluations of the three GSs from each of the CSp and CSu groups.

Figure 5. Mean CS evaluations of the valence instruction and neutral instruction groups measured at baseline, acquisition, and post-instruction in Experiment 2. Error bars represent the standard error of the mean.

Figure 6. Mean GS evaluations of the valence instruction and neutral instruction groups measured at baseline, acquisition, and post-instruction in Experiment 2. Error bars represent the standard error of the mean.

CS Evaluations. Main effects of CS, F(1, 845) = 8.72, p = .003, ηp² = .010, phase, F(2, 844) = 5.23, p = .003, ηp² = .013, and group, F(1, 845) = 28.33, p = .032, ηp² = .005, and interactions of CS × Group, F(1, 845) = 368.41, p < .001, ηp² = .304, Phase × Group, F(2, 844) = 40.91, p < .001, ηp² = .088, and CS × Phase, F(2, 844) = 537.35, p < .001, ηp² = .560, were moderated by a CS × Phase × Group interaction, F(2, 844) = 634.51, p < .001, ηp² = .601. Follow-up analyses of the CS × Phase × Group interaction revealed no difference in CS evaluations at baseline in both groups, both F(1, 845) < 1.90, p > .168, ηp² < .003. After acquisition, both groups evaluated the CSp as more pleasant than the CSu, both F(1, 845) > 236.91, p < .001, ηp² > .218. After the instruction manipulation, the neutral instruction group continued to evaluate the CSp as more pleasant than the CSu, F(1, 845) = 143.45, p < .001, ηp² = .145. This pattern was reversed in the valence instruction group, such that the CSu was evaluated as more pleasant than the CSp, F(1, 845) = 1320.26, p < .001, ηp² = .610.

GS Evaluations. Main effects of GS, F(1, 845) = 73.78, p < .001, ηp² = .080, and phase, F(2, 844) = 22.45, p < .001, ηp² = .051, and interactions of GS × Group, F(1, 845) = 83.02, p < .001, ηp² = .089, Phase × Group, F(2, 844) = 4.39, p = .013, ηp² = .010, and GS × Phase, F(2, 844) = 154.78, p < .001, ηp² = .268, were moderated by a GS × Phase × Group interaction, F(2, 844) = 141.74, p < .001, ηp² = .251. Follow-up analyses of the GS × Phase × Group interaction revealed no difference in GS evaluations at baseline in either group, both F(1, 845) < 0.15, p > .700, ηp² < .001. After acquisition training, both groups evaluated the GSp as more pleasant than the GSu, both F(1, 845) > 136.56, p < .001, ηp² > .138. After the instruction manipulation, the neutral instruction group continued to evaluate the GSp as more pleasant than the GSu, F(1, 845) = 115.95, p < .001, ηp² = .121, whereas the valence instruction group evaluated the GSu as more pleasant than the GSp, F(1, 845) = 142.12, p < .001, ηp² = .144. The group main effect did not reach significance, F(1, 845) = 1.24, p = .266, ηp² < .001.

CS Evaluations vs. GS Evaluations. Separate paired samples t tests were used to compare the magnitude of differential CS evaluations with differential GS evaluations at baseline, acquisition, and after the neutral instructions (in the neutral instruction group) and after the positive/negative instructions (in the valence instruction group). At baseline, there was no difference between differential CS (M = 0.04, SD = 1.42), and GS evaluations (M = 0.004, SD = 1.32), t(846) = 0.61, p = .545, d = 0.464. After acquisition, differential CS evaluations (M = 2.32, SD = 2.99) were larger than differential GS evaluations (M = 1.39, SD = 2.49), t(846) = 13.51, p < .001, d = 0.464. In the neutral instruction group, differential CS evaluations (M = 1.80, SD = 2.73) were larger than differential GS evaluations (M = 1.34, SD = 2.42) after the neutral instructions, t(423) = 4.83, p < .001, d = 0.234. In the valence instruction group, differential CS evaluations (M = −5.45, SD = 3.40) were larger than differential GS evaluations (M = −1.48, SD = 2.69), t(422) = 23.47, p < .001, d = 1.141, after the valenced instructions.

Implicit evaluations

Participants who made more than 30% of errors during priming were excluded (144 total), resulting in 350 participants in the valence instruction group and 340 in the neutral instruction group. Tests of normality indicated that the reaction time data violated assumptions of normality, therefore reaction times were inverted, as recommended by Ratcliff (1993). Mean non-inverted RTs are displayed in Figure 7. Reaction times were subjected to a 2 GS (GSp, GSu) × 2 Target Valence (pleasant, unpleasant) × 2 Group (valence instruction group, neutral instruction group) between-participants mixed model ANOVA. A main effect of target valence, F(1, 688) = 184.46, p < .001, ηp² = .211, and a GS × Target Valence interaction, F(1, 688) = 12.37, p < .001, ηp² = .018, were moderated by a GS × Target Valence × Group interaction, F(1, 688) = 4.93, p = .027, ηp² = .007. The GS × Target Valence × Group interaction was followed up by calculating a priming score ([incongruent: GSp-unpleasant target + GSu-pleasant target] – [congruent: CSp-pleasant target + GSu-unpleasant target]), which was significantly larger in the neutral instruction group (M = 7.59, SD = 41.52) than the valence instruction group (M = 2.43, SD = 42.61), t(688) = 2.28, p = .027, d = 0.169. In the neutral instruction group, the priming score was significantly larger than zero, t(339) = 4.00, p < .001, d = 0.217, indicating that the GSp was evaluated as more pleasant than the GSu at an implicit level. In the valence instruction group, the priming score was not significantly larger than zero, t(349) = 0.93, p = .353, d = 0.050, indicating that there was no difference in implicit evaluations between the GSp and GSu.

Figure 7. Mean RT for congruent pairings (GSp + positive word; GSu + negative word) and incongruent pairings (GSp + negative word; GSu + positive word) measured for the neutral instruction and valence instruction group in Experiment 2. Error bars represent the standard error of the mean.

Discussion

In Experiment 2, we used a between-participants design to examine whether CS instructions update implicit evaluations of other group members after evaluative conditioning. Using a between-subjects design allowed us to only measure implicit valence once, and therefore avoid any effects of repeating the affective priming task. We also added a measure of US expectancy after acquisition to assess contingency awareness directly after conditioning. This resulted in a higher percentage of participants demonstrating contingency awareness, compared to Experiment 1, however, contingency awareness did not influence the results.

The explicit evaluations followed the same pattern as in Experiment 1, in which CS and GS evaluations reversed after the valenced instructions were presented. As we expected, the affective priming results indicated that the GSp was evaluated as more pleasant than the GSu in the neutral instruction group while there were no differences in implicit GS evaluations in the valence instruction group. This finding indicates that the positive and negative instructions were successful in reducing the generalisation of previous learning, but not in reversing generalised implicit evaluations.

General discussion

In the current investigation, we examined whether presenting diagnostic instructions about an individual group member that had previously been subject to evaluative conditioning would influence the implicit and explicit evaluations of other group members. We conducted this research across two experiments using fictional alien groups, where one group member was conditioned as a CS and the other group members were used as generalisation stimuli. In Experiment 1, we used a pre–post design where GS evaluations were measured before and after the CS instructions were presented. In Experiment 2, we used a between-participants design where one group received positive and negative CS instructions and a control group received neutral CS instructions. Implicit evaluations were measured once after the instructions were presented.

In Experiment 1, explicit and implicit evaluations of the GSp and GSu were pleasant and unpleasant, respectively, after the conditioning procedure. After the CS instructions were presented, explicit GS evaluations reversed such that the GSp became negative while the GSu became unpleasant. After the instructions, there was no longer a difference between GSp and GSu in implicit evaluations. In Experiment 2, both explicit and implicit evaluation of the GSp and GSu were pleasant and unpleasant, respectively, after the neutral instructions were presented, indicating that the neutral instructions had no effect on GS evaluations. Conversely, in the group who received the valenced CS instructions, explicit GS evaluations reversed, while implicit GS evaluations were eliminated, such that there was no difference between GSp and GSu evaluations. These findings demonstrate, for the first time, that instructions about a specific individual (CS) can influence the implicit valence of other group members (GSs) who were not present during conditioning or the target of instructions.

The current findings are consistent with accounts that posit that evaluations are formed via propositions (i.e. an understanding of how the CS and US are related). Propositional theories propose that both explicit and implicit evaluations are sensitive to CS instructions (see De Houwer, 2014), which is partially consistent with our findings. We observed a reversal of explicit evaluations (where the GSp became unpleasant and the GSu became pleasant) after the CS instructions, however, implicit evaluations did not reverse, but were eliminated. This was expected, based on the findings from Mann et al. (2020), in which positive diagnostic information presented after negative conditioning did not result in a reversal in implicit evaluations (i.e. negative becoming positive), but rather implicit CS evaluations became neutral in each of the individual experiments. According to De Houwer (2014), dissociations between explicit and implicit evaluations do not occur due to differences in how these evaluations are formed, but rather because of differences in the retrieval of information during assessment. For example, when receiving new information that contradicts prior learning (i.e. instructions after evaluative conditioning), self-report measures allow individuals time to consider these two opposing propositions and to decide which is true or holds more validity. Conversely, implicit measures require very fast responses, which does not allow the individual to control which of the two conflicting beliefs will influence their responding. As a result, implicit evaluations will reflect the incomplete retrieval of information (Van Dessel et al., 2019), which may explain why we observe an elimination of implicit evaluations, rather than a reversal.

The current findings provide evidence that reversing CS valence via instructions can override the generalisation effects instilled via evaluative conditioning, even when the instructions are specific to the individual CS only. This indicates that GS evaluations simply mirror CS evaluations, such that any changes in CS evaluations will be reflected in changes in GS evaluations. The instructions may have caused a reversal of GS evaluations because of their high diagnosticity. Glaser et al. (2015) proposed evaluations instilled by information that is both extreme and negative should generalise more readily compared to positive information since it is more salient and could be potentially threatening. The negative CS information did cause previously positive GS evaluations to become negative, but in Experiment 1, the magnitude of the change in GS evaluations did not differ compared to the stimulus that was paired with the positive information. In Experiment 2 however, we did observe a stronger reversal for the GSp (paired with negative information) compared to the GSu (paired with positive information).³ Based on theories of cognitive consistency (for a review see Harmon-Jones, 2007), Glaser et al. (2015) also proposed generalisation may occur because individuals are motivated to avoid inconsistent cognitions such as different evaluations of members from the same group. For instance, if the evaluation of the CS is reversed via instructions that are inconsistent with prior conditioning, then GS evaluations will also change in line with the CS evaluations to re-establish consistency among evaluations. However, we still do not know whether dissociations between CS and GS valence would be seen in more complex scenarios. For example, future studies should examine how groups are evaluated when two group members hold opposing valence acquired via different pathways (i.e. if one alien was conditioned as positive, but another alien from the same group was paired with negative information, how would the remaining group members be evaluated?). We chose fictional alien groups to avoid pre-existing attitudes about real life social groups influencing the conditioning procedure, however, this minimises the ecological validity of the findings. Future research should extend the current research to examine whether diagnostic instructions can also affect the evaluation of more realistic groups formed using either minimal group paradigms or social category cues such as race, age, and gender.

It is worth nothing that providing group membership information about the two groups before conditioning likely enhanced generalisation. We do not see this as a limitation as people are usually provided with information about the group memberships of individuals they encounter or can infer them readily. It could also be argued that the perceptual similarity among the aliens within each group could have made it difficult for participants to tell the GSs apart from the CSs. Explicit CS evaluations were larger than GS evaluations after acquisition and the instructions in both experiments, which indicates that participants could tell the group members apart during the conditioning task. If participants were not able to distinguish the CSs from the GSs we would expect to see no difference in how the CSs and GSs are evaluated. We cannot make this comparison for implicit evaluations (as we did not measure implicit CS evaluations), but given that participants appeared to be aware of the differences between group members in the explicit ratings, we believe that this is likely also true for the priming task. Although each GS was only presented for 200 ms during the affective priming task, we believe that 200 ms is long enough to note the distinct physical characteristics of the image and therefore we believe it was sufficient time for the participants to be able to note the distinct physical features of each individual group member and recognise them as different aliens than the CSs. A limitation of the current study is the contingency awareness measure in Experiment 1 resulting in a small number of participants (12.1%) who were able to identify the correct CS-US contingencies. This was likely due to memory decay and difficulty identifying the correct CSs when all aliens were presented on screen at the same time. The strong conditioning and generalisation effects observed in Experiment 1 contradict the results of the contingency assessment, therefore the high number of incorrect responses is likely a result of measuring contingency during the post-experimental questionnaire and does not necessarily reflect a lack of contingency awareness during conditioning. In Experiment 2, contingency awareness (as indicated by the US expectancy measure) did not affect generalisation, but this measure may have also been impacted by participants’ uncertainty about future CS-US pairings.

The current findings demonstrate that diagnostic CS instructions presented after evaluative conditioning can change the explicit and implicit valence of generalisation stimuli, here other members of fictitious groups of aliens. This indicates that generalised evaluations instilled via evaluative conditioning are malleable and can be updated to reflect changes in CS valence. This finding is encouraging as it suggests that generalised negative evaluations of a group acquired via evaluative conditioning can be eliminated by providing positive information about one group member. On the other hand, the findings also suggest that when a group is evaluated as positive, this positive evaluation of the group can be undone if negative information is provided about a single group member. The findings contribute to the growing body of evidence that evaluative learning generalises to other similar stimuli and extends previous work to demonstrate generalised valence can be influenced by CS instructions.

Author note

The raw data, stimulus materials, and analysis scripts are available at https://osf.io/akqbw/?view_only = 7d3c33a6e02c444aa530180acf015608 (this link will be made public after acceptance of the manuscript). This study was not preregistered.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This work was supported by an Australian Government Research Training Program Scholarship to Rachel Patterson and grant DP180111869 from the Australian Research Council to Ottmar Lipp.

Notes

1 We measured contingency awareness as a manipulation check to determine whether participants were paying attention to the experiment. Our aim was not to examine the role of contingency awareness as a moderator in conditioning or generalisation.

2 Follow-up analyses of this marginal interaction revealed that the pleasant USs were evaluated as slightly more pleasant in the valence instruction group (M = 8.02, SD = 0.96) compared to the neutral instruction group (M = 7.86, SD = 0.96), F(1, 831) = 6.44, p = .011, ηp² = .008.

3 We analysed whether the change in GS evaluations from post-acquisition to post-instruction differed between the GS paired with negative information (GSp) and the GS paired with positive information (GSu) by calculating a difference scores (GSp post-acquisition minus GSp post-instruction and GSu post-acquisition minus GSu post-instruction) which were then subjected to a paired samples t-test. In Experiment 1, the difference between GS evaluations at post-acquisition compared to post-instruction did not differ between the GSp (M = 1.53, SD = 1.93) and the GSu (M = -1.52. SD = 1.92), t = 0.223, p = 0.824, d = 0.014. In Experiment 2 (in the valanced instruction group only), the size of the GSp reversal (M = 1.52, SD = 2.09) was larger than the size of the GSu reversal (M = 1.33, SD = 2.02), t(422) = 2.49, p = .013, d = 0.121.

Appendices

Appendix A

Positive Instructions

This alien is named Bim. He belongs to an alien race from the planet Anite. Bim has devoted the last 10 years of his life to caring for homeless aliens. Bim works very hard at his day job in order to earn extra money so that he can donate it to the local homeless shelter. He spends every weekend doing volunteer work in which he helps build new houses for homeless aliens.

This alien is named Bim. He belongs to an alien race from the planet Anite. Bim is an animal welfare worker who has devoted his life and career to seeking out and helping abused or mistreated wildlife on planet Anite. Bim has passionately and effectively pursued this work for many years, taking pay cuts and even putting himself in harm’s way several times in order to help animals in need. (Adapted from Mann et al., 2020).

This alien is named Bim. He belongs to an alien race from the planet Anite. Bim recently rushed into a burning spaceship to save a baby alien, just as the fire took a turn for the worse. Running through the flames to reach the wailing infant in the spaceship, he grabbed the baby and rushed outside to safety a split-second before the spaceship exploded. The baby was unharmed, and Bim was hailed as a hero. (Adapted from Mann et al., 2020).

Negative Instructions

This alien is named Zop. He belongs to an alien race from the planet Emer. Zop purposely set fire to the largest forest on planet Emer. Thousands of Emer’s unique animals were killed in the fire, causing several species to become extinct. Zop said that he did not feel any remorse for his actions nor did he care about the death and destruction he had caused.

This alien is named Zop. He belongs to an alien race from the planet Emer. Zop has physically abused and neglected his pet animals. Many of his pets were found severely malnourished and some had also sustained injuries that required surgery. When questioned about his actions, Zop admitted that he never cared about the wellbeing of his pets and does not feel remorse for his actions.

This alien is named Zop. He belongs to an alien race from the planet Emer. Zop beat his girlfriend after a small argument. After sustaining severe injuries, Zop’s girlfriend managed to get away but eventually she succumbed to her injuries and died. Zop admitted that he did not feel remorse for his actions, nor did he care about the death of his girlfriend.

Appendix B

Neutral Instructions

This is an alien from the planet Emer. Emer is a planet located in a galaxy that spans 40 million miles. The weather on Emer is usually moderate with more sunny days than rainy days. During the day it can be quite warm, but it usually becomes colder during the night. Emer typically experiences approximately 3–4 days of snowfall each year.

This is an alien from the planet Emer. Emer is a medium-sized planet located near the centre of a large galaxy. Emer is made up of 60% water and 40% land. The land on Emer can be very flat in some areas, whereas other areas of the planet contain large mountains. The land in the flat areas is typically smooth and sandy, while the land in the mountains is hard and rocky.

This is an alien from the planet Emer. Emer is a planet located on the outskirts of a small galaxy. The ground on Emer contains soil that can grow a variety of edible and inedible plants. In some areas, only small shrubs and bushes grow. The soil in other areas can grow large trees and many different plant species.

This is an alien from the planet Anite. Anite is a planet located within a galaxy 100 million miles away. Anite has its own moon and revolves around a large sun located at the centre of the galaxy. Each day on Anite last approximately 17 h, consisting of 11 h of sunlight and 6 h of moonlight. One year on Anite consists of 280 days.

This is an alien from the planet Anite. Anite is a planet located in a galaxy slightly smaller than the Milky Way. Anite is a medium sized planet, with a surface area of 800,000 square miles. It is the third largest planet in the galaxy. The closest planet to Anite is approximately 10 million miles away.

This is an alien from the planet Anite. Anite is a planet located within a large galaxy. This planet contains many oceans and lakes. The largest lake covers a surface area of approximately 10,000 square miles. The oceans on Anite contain salt water while the lakes contain fresh water.