(Why) Do Big Five Personality Traits Moderate Evaluative Conditioning? The Role of US Extremity and Pairing Memory

The definition of EC already suggests where personality differences might come from. Because EC is the change in liking of a CS due to its co-occurrence with a valenced US, personality differences in EC might be due to personality differences in the valence of the US. That is, the same US might be more or less extreme depending on the Big Five, which could contribute to personality differences in EC. This explanation raises three questions, which we will discuss consecutively:

1. How could personality differences in US extremity emerge in EC experiments?

2. How should US extremity influence EC?

3. How should the Big Five relate to US extremity?

A common paradigm to study EC is the picture-picture paradigm (Levey & Martin, 1975), where neutral CS pictures (e.g., faces, words, abstract forms) are presented with positive or negative US pictures (e.g., happy people, puppies, tasty food vs. violence, blood, rotten food). Many EC studies, including Vogel et al. (2019), select highly positive/negative US pictures from standardized databases, such as the International Affective Picture System (IAPS; Lang et al., 2008) or the Open Affective Standardized Image Set (OASIS; Kurdi et al., 2017).

But even though the selected pictures are highly positive/negative to most individuals, there are still substantial interindividual differences in the pictures’ valence (Ingendahl & Vogel, 2022). For instance, the picture in Figure 1 (i.e., picture I263 in the OASIS database) yields a mean valence rating of 5.85 on the 1-7 scale, thus being a prime candidate for being selected as a positive US (> +1SD compared to all other OASIS pictures). However, the standard deviation of the picture’s valence ratings is 1.14. Assuming a normal distribution of the valence ratings, approximately 16% of a random participant sample would evaluate it as neutral or even negative (below 4.7 on the 1-7 scale). These differences in the valence of the picture might also lead to differences in the evaluation of a CS paired with this picture.

Accordingly, one might spontaneously dismiss personality differences in EC caused by personality differences in US extremity as a mere artifact. Specifically, one could argue that the antecedent conditions for EC – pairings with valenced stimuli – are simply not the same for all people in such a picture-picture paradigm. Also, one might speculate that personality differences in EC vanish if researchers simply select US materials based on individual pre-ratings instead of normed ratings. Yet, as discussed next, the relationship between US extremity and EC is more complex and depends on the underlying cognitive processes.

EC is commonly defined as the change in liking towards a neutral CS due to its mere co-occurrence with a valenced US (De Houwer, 2007). This functional definition of EC leaves open a) in which direction the valence of the US impacts the liking of the CS (e.g., positive US positive CS or positive US negative CS), and b) which particular cognitive mechanisms are involved.¹ And indeed, previous theories on the cognitive processes behind EC make different assumptions about how US extremity should relate to EC effects:

According to associative or referential accounts, the CS becomes automatically linked to the US in memory (Baeyens et al., 1992; Gawronski & Bodenhausen, 2018). These associative links form under the Hebbian rule (“fire together wire together”) and thus depend only on the spatiotemporal proximity of CS and US. Upon recurrence of the CS, the mental representation of the US is activated, and thereby, its valence. More recent memory-based accounts propose that instead of automatic formation and activation, a conscious encoding and retrieval of US valence is necessary (Gast, 2018; Stahl & Aust, 2018). According to associative and memory-based accounts, CSs linked in memory to more extreme USs should elicit more extreme evaluative responses. That is, a CS linked in memory to a highly positive US valence should consequentially receive a more positive evaluation than a CS linked in memory to a moderately positive US valence.

Propositional accounts (e.g., De Houwer, 2018) argue that people draw inferences about the relation between the CS and the US. These inferences could range from “the CS causes the US”, “the CS co-occurs with the US”, to “the CS is opposite to the US”. In line with this account, EC depends on the relation people infer from the stimulus pairings (Unkelbach & Fiedler, 2016). For instance, a CS paired with a negative US can be evaluated positively if the relation between the two stimuli is negative (e.g., “the CS is opposite to the US”). As the default proposition for co-occurring stimuli is that they are positively related (Unkelbach & Fiedler, 2016), one could expect that in most situations, more extreme US valence leads to stronger EC. Yet, people also form propositions about evaluative features of the US (e.g., “This stimulus is positive”; De Houwer, 2018) that could deviate from the individual’s subjective evaluation of the US. For instance, some people might dislike dogs but might form a proposition based on the normative valence of the stimulus (e.g., “This dog is supposed to be positive”). Accordingly, propositional theories would be consistent with stronger EC for extreme USs, but can also account for other outcomes.

Last, according to the affect misattribution account, people confuse the source of their affective experience when presented with a CS-US pair. Specifically, they misattribute the affective response from the US to the CS (Jones et al., 2009). Affect misattribution is more likely if US valence is less extreme; thus, this account predicts that EC is stronger for less extreme USs (Jones et al., 2009; March et al., 2019; but see Mierop et al., 2019).

There is surprisingly little empirical evidence on how US extremity influences EC (De Houwer, 2011). Yet, recent research on the cognitive processes in EC has found good support for memory-based and propositional accounts (Corneille & Stahl, 2018; De Houwer, 2018; Hofmann et al., 2010; Moran et al., 2020, 2023), and thus, we relied on these accounts to derive our hypotheses. Accordingly, extreme USs should lead to stronger EC. Logically, the same should be the case for interindividual differences in US extremity: Individuals for whom the USs are more extreme should show stronger EC. Still, EC is often considered a multi-process phenomenon (Gawronski & Bodenhausen, 2018; Sweldens et al., 2010), and some individuals might be more prone to one or another mechanism. To account for this, we investigated not only how the Big Five relate to US extremity, but also personality differences in how US extremity relates to EC.

Differences in stimulus appraisals have been the focus of a large body of research, showing that personality traits are associated with the differential experience of positive/negative stimuli (see Augustine & Larsen, 2015). This is especially true for Neuroticism and Extraversion, but also for Agreeableness.

Costa and McCrae (1980, p. 673) bring it to the point: “Extraversion […] predisposes individuals toward positive affect, whereas Neuroticism […] predisposes individuals toward negative affect”. In line with this notion, plenty of findings have shown that Neuroticism is associated with stronger reactivity to negative stimuli and Extraversion with stronger reactivity to positive stimuli (e.g., Canli et al., 2001; Gross et al., 1998; Larsen & Ketelaar, 1991; Rusting & Larsen, 1997; Smillie et al., 2012). Accordingly, normatively negative USs should be more negative to people high in Neuroticism; and normatively positive USs should be more positive to people high in Extraversion.

Agreeableness is negatively associated with psychopathy (Decuyper et al., 2009; Stead & Fekken, 2014), a trait related to abnormal reactions to emotional stimuli (Hoff et al., 2009; Kiehl et al., 2001). Accordingly, people with higher Agreeableness show stronger behavioral approach reactions toward positive stimuli and stronger avoidance reactions toward negative stimuli (Bresin & Robinson, 2015; Czerwon et al., 2011; Finley et al., 2017; Ingendahl & Vogel, 2022). Another essential aspect of Agreeableness is compliance with social norms (Wilmot & Ones, 2022). Thus, people high in Agreeableness might be more likely to evaluate a stimulus in line with its normative valence. Consequently, normatively positive USs should be more positive, and normatively negative USs should be more negative to people high in Agreeableness.

These predictions were recently supported in a study by Ingendahl and Vogel (2022): They let a broad and heterogenous sample evaluate a typical US set of positive, neutral, and negative pictures from the OASIS and also assessed the Big Five. Their findings revealed more extreme evaluations for individuals with higher Neuroticism, Extraversion, and Agreeableness.

To summarize:

1. Personality differences in US extremity are likely to emerge in standard EC procedures.

2. Based on memory-based and propositional accounts, extreme USs should lead to stronger EC.

3. USs should be more extreme for people high in Neuroticism, Extraversion, or Agreeableness.

As a logical consequence, our main prediction in this research is that people high in Neuroticism, Extraversion, or Agreeableness show stronger EC effects, but that these stronger EC effects can be explained by more extreme US valence for people high in these traits. This is what we tested in the present study.

Therefore, we first predicted an EC effect (H1) which should increase at higher levels of Neuroticism (H2), Extraversion (H3), and Agreeableness (H4). However, these moderations should go together with more extreme US evaluations. To test this, we also assessed participants’ individual US evaluations and examined to what extent personality differences in US evaluations could account for personality differences in EC. Here, we predicted that people evaluate normatively positive stimuli more positively than normatively negative stimuli (H5). Based on the theoretical reasons explained above, we predicted that US evaluations become more extreme at higher levels of Neuroticism (H6), Extraversion (H7), and Agreeableness (H8). We expected that individual US evaluations mediate the EC effect (H9) and that the moderations of the EC effect in H2-H4 can be explained statistically by the more extreme US evaluations (H10a, H11a, H12a). In addition, we tested whether the personality traits moderate the association between individual US evaluations and CS evaluations (H10b, H11b, H12b). Our detailed research model is depicted in Figure 2.

Next to personality differences in US extremity, people might also differ in their memory for the stimulus pairings. Pairing memory is a crucial moderator in most EC accounts – with vastly different expectations. Current memory-based and propositional accounts postulate that a conscious encoding and retrieval of the US (more precisely, its valence) is even necessary for EC (De Houwer, 2018; Gast, 2018; Stahl & Aust, 2018). Referential/associative accounts require only an automatic mental connection between CS and US valence and thus only implicit memory (e.g., Baeyens et al., 1992). In contrast, the implicit misattribution account would even predict that people are less likely to misattribute affect if they recollect the stimulus pairings (Jones et al., 2009; March et al., 2019). In line with memory-based and propositional accounts, pairing memory is empirically the strongest moderator of EC, with stronger EC when participants can remember the CS-US pairings (Hofmann et al., 2010; Sweldens et al., 2014).

Due to the central role of pairing memory in EC, one might therefore speculate that some Big Five traits are associated with differences in the encoding or retrieval of stimulus pairings. However, it is difficult to come up with straightforward predictions here. For instance, one could argue that memory is better for emotional than non-emotional content (Grider & Malmberg, 2008; Hamann, 2001). Based on this memory advantage, one might speculate that more extreme USs are also better remembered. Accordingly, people for whom USs are more extreme – people high in Neuroticism, Extraversion, or Agreeableness – should also have better US memory. However, EC does not depend on memory for the US alone but on memory for the CS-US pairing. EC requires remembering the relation between two concepts which rather corresponds to associative memory or source memory. In these domains, evidence on emotion-enhanced memory is mixed (Bell et al., 2012; Symeonidou & Kuhlmann, 2022) or even shows worse memory for negative emotional content (Bisby & Burgess, 2014, 2017). Thus, it is unclear whether more extreme USs also lead to more accurate pairing memory (but see Mierop et al., 2019, for support of that idea).

Beyond that, previous research argued that Neuroticism and Extraversion foster the formation of associative networks selectively for negative/positive concepts (Robinson, 2007), which might benefit the encoding of CS-US pairings of that specific valence. Also, some findings indicate that the two traits are associated with valence-specific retrieval advantages of positive (Extraversion) and negative (Neuroticism) information (Mayo, 1983; Rusting, 1998, 1999). Yet, all of these findings are only loosely related to the memory for stimulus pairings in EC.

Overall, the theoretical (and empirical) fundament for personality differences in pairing memory is much weaker than for differences in US extremity. Due to the importance of pairing memory in EC, we nevertheless considered it worthwhile to investigate its role in personality differences in EC, but in an exploratory manner. Therefore, we also assessed participants’ pairing memory with a postconditioning memory test (Walther & Nagengast, 2006) and examined to what extent personality differences in pairing memory exist and how they relate to personality differences in EC (see Figure 2).

We preregistered our hypotheses, methods,² and analyses on the OSF: https://osf.io/y5n7p . All data, analysis scripts, and materials are in the following OSF directory: https://doi.org/10.17605/OSF.IO/3UJ7K

In a single-factor design, normed US valence (NUSV; positive vs. negative) varied within participants. The Big Five served as continuous covariates. To determine the sample size, we conducted an a priori power analysis with GPower (Faul et al., 2007). We took the mixed ANOVA design with two groups and two repeated measures. We considered this a conservative approximation for our primary analysis, where personality was a continuous (versus categorical) between-subjects predictor, and each NUSV level was measured ten times (versus once). Our goal was to obtain 90% power to detect a small effect size³ of f = .08 at the standard 5% alpha level for the between-within interaction, which resulted in N = 414 as a minimum sample size. Due to uncertainty regarding the actual effect sizes, we decided to oversample to N = 500. Note that this sample size is also sufficient to detect a significant mediation for correlations between predictor, mediator, and outcome as low as r = .18 (Schoemann et al., 2017).

511 German participants were recruited via the Respondi panel and compensated according to the panel’s incentive system (min. 1€) for a study of 20 minutes. Detailed descriptive statistics of our sample are displayed in Table 1. Overall, our sample was heterogeneous in gender, age, and education.

The experiment was structured in the following way: After an informed consent, participants first answered a Big Five questionnaire. Next, 20 US pictures (10 per NUSV level) were drawn randomly from our stimulus pool and evaluated by the participants. After that, the same 20 US pictures were paired with neutral CSs in a subsequent conditioning phase. Next, participants also evaluated the CSs. A memory test for the stimulus pairings followed. Finally, demographic information was assessed, and participants were thanked and debriefed about the study’s purpose. In the following, we will explain each experimental task step by step. A detailed list of all stimuli and screenshots of the respective tasks can be found in the OSF supplement.

We used the same instructions and rating scale (a seven-point scale labeled at each point) as in the original OASIS (Kurdi et al., 2017) and the study by Ingendahl and Vogel (2022). The 20 US pictures (10 per normed US valence) were rated in random order. Each picture was presented on its own slide. Below the picture, the heading “Valence” was shown together with the labeled scale (very negative, moderately negative, somewhat negative, neutral, …, very positive).

After having evaluated the US pictures, participants were exposed to an EC procedure. Participants were told that they would now see some sequences of brand names they were unfamiliar with, presented together with pictures. They were instructed to look at these sequences carefully and wait for further instructions. Next, 20 CSs were presented together with the US pictures. Ten CSs were conditioned positively, and ten CSs negatively. Each conditioning trial started with a blank screen of 250ms, followed by a CS-US pair presented for 2500ms. We counterbalanced between participants whether the CSs/USs were shown on the left/right side (see the OSF for screenshots). Each CS was conditioned five times, leading to overall 100 trials presented in random order. Each CS was always shown with the same US.

After this conditioning procedure, participants also evaluated the CSs in random order. Each CS was rated on the same seven-point scale as the US evaluation. On each slide, participants were asked, “How would you evaluate this brand name?”, and presented with the brand name and the rating scale.

After the CS evaluations, participants were told to recollect which specific picture had been paired with a brand name. The following memory test was based on the measure of Walther and Nagengast (2006). Each CS was presented on a single slide together with a matrix of four US pictures from the conditioning phase. Participants had to select the US picture the CS had been presented with. Two pictures were from positive NUSV, and two pictures were from negative NUSV. One of these pictures was the correct US. The position of the pictures varied randomly within the matrix.

We used the BFI-2 with 60 items to assess the Big Five (Danner et al., 2019). Note that the BFI-2 labels neuroticism as negative emotionality, but we use the more common term neuroticism here. Descriptive statistics, internal consistencies, and intercorrelations are displayed in Table 2.

We used a random subset of 36 fictional brand names (e.g., STAREBO, DEMADOS) as CSs that had been rated neutrally in a pretest. The full list is provided in the OSF directory. For each participant, we selected a random subset of 20 brand names.

We used the same 2 x 30 pictures as in the study by Ingendahl and Vogel (2022) as USs. Thirty pictures from the OASIS (Kurdi et al., 2017) with normed valence ratings higher than +1 SD (OASIS valence rating > 5.56 on a scale of 1-7) served as positive NUSV stimuli, and 30 pictures with normed valence ratings below -1 SD (2.86) as negative NUSV stimuli. The picture sets thus differed in their valence ratings reported in the manual, F(1, 58) = 3033.61, p < .001, M_neg = 2.40, M_pos = 5.88. The pictures had arousal ratings between -1SD (2.86) and +1SD (4.50). Positive and negative images overall did not differ in their arousal ratings, F(1, 58) = 0.05, p = .827. Each US set consisted of ten images depicting scenes, ten depicting persons, five depicting objects, and five depicting animals. We selected the 2 x 10 USs randomly for each participant, without any constraints regarding their category. We did not use pictures depicting extreme violence or nudity.

As the measures were nested within participants, we ran multilevel regression models with the R package lme4 (Bates et al., 2019). In all models, we used the highest converging random effect structure (Barr et al., 2013). The CS evaluations were standardized at the grand mean. In a first baseline model, we decomposed the variance of the ratings into variance between individuals and error variance by including random intercepts for the participant. Next, we included NUSV as a predictor in the model, coded with 1 (positive) and -1 (negative). As expected, NUSV had an impact on CS evaluations, β = .205, 95% CI [.177, .233], thus showing a robust EC effect (H1). Notably, the EC effect was heterogeneous across participants, as indicated by the standard deviation of the random slope, SD_NUSV = .069.

For our main model, we standardized the Big Five and entered them into the model with their two-way interactions with NUSV (see Table 3a). In this model, the interaction terms Personality x NUSV test whether interindividual differences in the EC effect (i.e., the variation in the random slope) can be explained statistically by the respective Big Five trait. Due to our coding scheme, regression weights can be interpreted as standardized weights. The results of this main model are visualized in Figure 3a and presented in detail in Table 3a.

The positive Neuroticism x NUSV interaction shows that the EC effect was descriptively stronger among individuals with high Neuroticism (H2). However, with p = .073, the interaction did not meet the preregistered significance level (α = .05). The Extraversion x NUSV interaction predicted in H3 was not significant, also the moderations by Conscientiousness and Openness. The Agreeableness x NUSV interaction was as expected in H4, such that individuals with high Agreeableness showed stronger EC effects. Thus, we replicated the main results obtained by Vogel et al. (2019), except for the moderation by Neuroticism, which was not statistically significant in the present study.

We analyzed the US evaluations with the same analytical approach as the CS evaluations (see Table 3b and Figure 3b). USs with positive normed valence were evaluated more positively than USs with negative normed valence (H5). Consistent with H6 and H8, this effect was more pronounced at higher levels of Neuroticism and Agreeableness, meaning more extreme US evaluations at higher levels of these traits. Simple slope analyses revealed that Neuroticism was related to more negative evaluations of normatively negative USs. Agreeableness was related to more extreme evaluations of both normatively positive and negative USs. In contrast to H7, there was no moderation by Extraversion. Additionally, a significant Conscientiousness x NUSV interaction revealed more extreme evaluations of the USs at higher levels of Conscientiousness.

Moderated Mediation Analysis. The previous analyses revealed similar moderations by personality on both CS and US evaluations. In line with the preregistration protocol, we thus conducted a multilevel moderated mediation analysis with the mediation package (Tingley et al., 2014) to test whether personality differences in US evaluations mediate personality differences in EC (H10a, H11a, H12a). To do so, we entered the US evaluations as a further predictor into the main model on the CS evaluations. To further test H10b-12b, we also entered the US Evaluation x Personality interactions. The results of this model can be found in Table 4a.

In line with H9, US evaluations had a strong positive effect on CS evaluations, and the EC effect was mediated by US evaluations, β = .136, 95% CI [.103, .170]. The effect of NUSV was reduced but still significant, β = .069, 95% CI [.040, .100], showing that US evaluations only partially mediated the EC effect.

Unexpectedly, the Neuroticism x NUSV interaction became stronger when controlling for the US evaluations and was now significant (see Table 4a). In contrast, the Agreeableness x NUSV interaction was considerably smaller but still significant. In line with the preregistration protocol, we thus computed conditional indirect effects at different levels of the two personality traits. These effects are displayed in Figure 4.

The indirect EC effect via US evaluations increased with higher levels of Agreeableness. Thus, more extreme US evaluations partially accounted for the stronger EC effect among agreeable individuals (H12a). However, the direct effect also increased, suggesting that the stronger EC effect for agreeable individuals could not be exclusively due to more extreme US evaluations. For higher levels of Neuroticism, only the direct effect increased, whereas the indirect effect became even slightly weaker. Thus, more extreme US evaluations could not account for the stronger EC effect for people high in Neuroticism found in this analysis.

In contrast to H10b, H11b, and H12b, the relationship between US evaluations and CS evaluations was not significantly moderated by any Big Five trait (see Table 4a).

Overall, participants identified the correct⁴ US for ~55% of the pairings, thus above chance levels (p_guessing = .25). We analyzed pairing memory with the same analytical approach as before but with a binomial regression due to the dichotomous format (“correct” = 1, “incorrect” = 0). The results are displayed in Table 3c and Figure 3c. Overall, pairing memory was slightly more accurate for negatively conditioned CSs. Apart from this, Agreeableness was associated with more accurate pairing memory.

As people high in Agreeableness showed stronger EC effects but also had more accurate pairing memory, we next tested whether the more accurate pairing memory could account for the stronger EC effect. To do so, we standardized pairing memory at the grand mean and added it into the main model, together with the two-way and three-way interactions with personality (see Table 4b). As evident from the positive Memory x NUSV interaction, EC effects were stronger when participants had accurate pairing memory. Also, the moderation by Agreeableness was stronger when participants had memory, as evident from the significant Memory x Agreeableness x NUSV interaction. Crucially, the moderation by Agreeableness became weaker when adding pairing memory into the model (Tables 3a/4b), implying shared variance of Agreeableness, pairing memory, and the EC effect. In addition, we observed an unexpected Memory x Openness x NUSV interaction, such that for people high in Openness pairing memory had a stronger moderating effect.

We also conducted a model with both US evaluations and pairing memory as predictors, shown in Table 4c. Overall, the results were very similar to the separate analyses, except that the Agreeableness x NUSV interaction was not significant anymore when controlling for both covariates. This implies that the stronger EC effect at high Agreeableness was not incremental to more extreme US evaluations and more accurate pairing memory combined.

Mediation Analysis. As the moderated mediation approach was not applicable here (because either a main effect or an interaction of Agreeableness emerged in the underlying models), we relied on a different analytical strategy here. Because the EC effect is the difference in evaluations between positively and negatively conditioned CSs, we computed such a difference score for each participant. From the previous results, one should expect that Agreeableness predicts this EC score and a participant’s pairing memory. We thus conducted a mediation analysis with Agreeableness as independent, the EC effect as dependent, and pairing memory as mediator variable while controlling for the other Big Five traits. The results are visualized in Figure 5; detailed results are provided on the OSF.

Pairing memory partially mediated the relationship between Agreeableness and EC, with a significant indirect effect, β = .091, 95% CI [.045, .143], and a significant direct effect, β = .113, 95% CI [.024, .200]. Thus, more accurate pairing memory partially accounted for the stronger EC effect among agreeable individuals.

We next computed a similar difference score with the US evaluations (positive-negative NUSV) to also control for interindividual differences in US extremity. Adding this score did not eliminate the indirect effect via pairing memory,⁵ β = .078, 95% CI [.030, .128]. Thus, the mediation by pairing memory was incremental to what could be explained by personality differences in US extremity. Crucially, the direct effect of Agreeableness was not significant anymore, β = .050, 95% CI [-.043, .142]. Thus, personality differences in US extremity and personality differences in pairing memory combined could fully account for the stronger EC effect at higher levels of Agreeableness.

In addition, we examined whether the effects of the Big Five were incremental to sociodemographic variables. Therefore, we repeated all previous analyses while controlling for gender and age. These analyses were preregistered, except for the analyses including pairing memory, as this measure was added after we preregistered the study. In all models, both control variables were allowed to interact with NUSV. The moderation by Neuroticism for the CS evaluations was gone entirely. Instead, there was a Gender x NUSV interaction, such that male participants showed weaker EC effects. All effects on US evaluations and pairing memory were robust against including the control variables, even the mediation analyses. Detailed results are provided on the OSF.

In the present study, we were interested in the incremental effects of each Big Five trait, that is, the relationship of a personality trait with EC that other correlated traits cannot explain. Yet, some researchers might be more interested in the mere bivariate relationship between a trait and the EC effect. Thus, we also computed bivariate correlations between the Big Five, the EC effect, US extremity, and pairing memory, which are presented in Table 2. All traits were significantly positively related to the EC effect with correlations between .09 and .20, except for Neuroticism which was negatively associated with the EC effect (-.03, not significant). A similar pattern emerged for US extremity and pairing memory. Overall, the correlations with Agreeableness were the strongest.

The Big Five can also be decomposed into more narrow personality facets. Our Big Five measure, the BFI-2, can be decomposed into 15 facets (three per dimension). In an exploratory manner, we also repeated our main models on CS evaluations, US evaluations, and pairing memory with the 15 facets (instead of the five dimensions). The results, including bivariate correlations, are presented in detail on the OSF. In essence, only Agreeableness-Compassion and Conscientiousness-Responsibility moderated EC effects in the multilevel analysis. They also had the strongest bivariate correlations with the EC effect, together with Agreeableness-Respectfulness. A very similar pattern emerged for US evaluations. For pairing memory, however, the pattern was more complex: Only Agreeableness-Respectfulness and Conscientiousness-Responsibility had a positive effect in the multilevel analysis, whereas the effect of Conscientiousness-Productiveness was negative. The bivariate correlations were strongest for Agreeableness-Respectfulness, Agreeableness-Compassion, and Conscientiousness-Responsibility.

In the present study, we examined whether Evaluative Conditioning (EC), a prominent effect in social and cognitive psychology, differs between individuals as a function of the Big Five. Our large-scale preregistered study revealed that EC increases at higher levels of Agreeableness, whereas the predicted moderation by Neuroticism was only significant in some of the analyses. The expected moderation by Extraversion did not emerge, as well as the moderations by Openness or Conscientiousness. Furthermore, people high in Agreeableness showed more extreme evaluations of the US pictures and more accurate memory of the stimulus pairings, which both accounted for the stronger EC effect for agreeable individuals. Even though people high in Neuroticism showed more extreme US evaluations, US evaluations (and also pairing memory) did not account for the stronger EC effect among people high in Neuroticism.

Next to extending and generalizing results from previous research (Vogel et al., 2019), our study reveals important insights into interindividual differences in EC and the Big Five traits Agreeableness and Neuroticism.

Our findings align with current propositional and memory-based accounts on EC (De Houwer, 2018; Gast, 2018). Consistent with previous evidence (Sweldens et al., 2014), and in contrast to the implicit misattribution account (Jones et al., 2009), more extreme US evaluations and more accurate pairing memory were positively related to EC. Going beyond previous findings that support these accounts on a situational level, the new insight from our research is that even personality differences in EC relate to personality differences in these two central constructs. In our study, the two combined accounted for the stronger EC effect among agreeable individuals. This further shows that theorizing on EC helps predict and explain changes in evaluations due to stimulus pairings – not only in which situation they occur, but also for which individuals.

Furthermore, our findings show which specific affective and cognitive mechanisms in EC relate to the Big Five. First, people systematically differed in US extremity in our study, but not in the relationship between US extremity and EC. Based on these findings, one could conclude that people differ in the antecedents for EC, namely, to what extent a positive/negative unconditioned stimulus is perceived as such. However, the relation between US extremity and EC seems independent of the Big Five. This suggests that once a stimulus is perceived as positive/negative, its effect on co-occurring stimuli does not depend on the Big Five. Thus, propositional and memory-based processes seem to operate similarly across different individuals. Furthermore, these findings speak against the possibility that some personality traits predispose individuals to affect misattribution (cf., Osorio et al., 2003).

Second, some people also have stronger pairing memory, which benefits EC. Thus, people with a disposition to form more accurate memory for stimulus co-occurrences are also more likely to change their evaluations of co-occurring stimuli.⁶ Again, this is perfectly in line with propositional and memory-based accounts on EC (De Houwer, 2018; Gast, 2018) and in contrast to the affect misattribution account (March et al., 2019). However, we also observed that the relationship between pairing memory and the EC effect increases with higher Agreeableness. This could imply that agreeable individuals not only have a more accurate memory of stimulus co-occurrences but also rely on this memory to a greater extent when forming an evaluative judgment. Another interpretation is that people differ not only in the strength but also in the type of CS-US memory links. For instance, agreeable individuals might be more prone to infer specific relations between CS and US (e.g., the CS causes the US) that further strengthen the EC effect (De Houwer, 2018).

To summarize, our study shows that personality differences in the central constructs in EC theories can account for personality differences in EC.

Our findings also offer valuable information about the Big Five, particularly Agreeableness. Personality research has often overlooked this trait (Graziano & Tobin, 2002). Yet, our findings emphasize that Agreeableness substantially relates to more extreme⁷ evaluations of affective stimuli, which has been suggested by previous research (Bresin & Robinson, 2015; Finley et al., 2017; Ingendahl & Vogel, 2022). Furthermore, our findings suggest that these more extreme evaluations can bare further consequences, specifically, more liking/disliking of co-occurring persons or objects in the social environment. This finding could also offer important insights into the role of Agreeableness in other domains, such as person perception. For instance, agreeable individuals are more prone to form positive perceptions of a stranger after brief interactions (Harris & Vazire, 2016). Agreeable individuals might be more likely to experience the interaction (US) positively and therefore evaluate the stranger (CS) more favorably.

Furthermore, our findings suggest that agreeable individuals form more accurate memory of stimulus co-occurrences in their social environment. Even though we did not predict this finding beforehand, one could speculate that relying on memory for positive/negative co-occurrences is a socially adaptive skill (Hütter et al., 2014). Detecting and memorizing co-occurrences of neutral and affective stimuli helps predict future exposure to these affective stimuli, thus enabling agreeable individuals to actively seek out or avoid them. Correspondingly, agreeable individuals have been shown to regulate their emotions by selecting positive situations and avoiding negative ones (Bresin & Robinson, 2015). Such a regulation strategy substantially benefits from accurate memory of which stimuli co-occur with positive/negative things in the environment.

Another important implication of our findings is that disagreeable individuals might form deviant associative network structures. Starting with a different evaluation of a US, stimuli linked in memory to this US will also obtain a different valence. Thus, the same objective learning episode seems to lead to different cognitive structures, depending on an individual’s Agreeableness. As evaluative responses influence other linked concepts in such a network (Vogel & Wänke, 2016; Walther, 2002), the different EC effects might also influence other concepts linked to the CS (e.g., stimuli linked to the CS via second-order conditioning). This interpretation of deviant associative networks relates well to previous research on Agreeableness, showing that social cues (e.g., the concepts of blame or aggression) might prime different thoughts and behaviors in disagreeable people (Meier et al., 2006; Meier & Robinson, 2004).

However, another interpretation of our findings is that agreeable individuals are more compliant with the experimental task and provide more reliable responses. EC paradigms like ours can be regarded as problem-solving tasks (De Houwer, 2018), and agreeable individuals might be more inclined to play the role of good subjects (Corneille & Lush, 2023). Even though we cannot exclude this alternative explanation, our analyses suggest that mere compliance with the experimental tasks might not fully explain the results for Agreeableness. First, in that case, controlling only for US extremity or pairing memory alone should have fully explained the moderation by Agreeableness, as each of these measures should share the same compliance component as the CS evaluation measure. Second, the facet-level analyses revealed that primarily the Agreeableness-Compassion facet was related to the US evaluations and EC, which is more in line with the interpretation that disagreeable individuals score high on psychopathy and thus show abnormal reactions to emotional stimuli (Decuyper et al., 2009; Hoff et al., 2009; Stead & Fekken, 2014). These effects were also incremental to the influence of the Conscientiousness – Responsibility facet, suggesting that the effect of Agreeableness was not exclusively due to more compliant or reliable responses.

Regarding Neuroticism, we observed only a small and marginally significant moderation. This moderation was stronger and significant when controlling for individual US evaluations. However, the zero-order correlation between Neuroticism and EC was descriptively opposite to the expected direction. In combination with the results of Vogel et al. (2019), these findings could imply that there might be a small moderation by Neuroticism that emerges only when controlling for the other Big Five traits. However, more research is necessary to be confident that Neuroticism moderates EC.

Yet, our results show that people high in Neuroticism are not better at memorizing the CS-US pairings. Also, even though Neuroticism is associated with more extreme US evaluations and, in particular, more negative evaluations of negative stimuli, this does not lead to stronger EC. In fact, our results suggest that the opposite is the case: When controlling for individual US evaluations, the moderation by Neuroticism became stronger. As one interpretation of this finding, people high in Neuroticism also have a heightened chronic focus on valence, which has been shown to strengthen EC (Gast & Rothermund, 2011). This chronic valence focus might play a lesser role for stimuli with strong emotional content, meaning those USs that received highly positive/negative evaluations. Thus, settings with more ambivalent USs might offer more room for the effect of Neuroticism (Bunghez & Sava, 2021).

Despite several strengths, such as a large and heterogeneous sample, this research is also compromised by shortcomings.

First, our study was inspired by previous research (Vogel et al., 2019), and our results imply the generalizability of personality differences beyond the specific paradigm of Vogel et al. (2019). However, due to the different materials and samples, we cannot say with absolute certainty that personality differences in US extremity and pairing memory could also account for the results of Vogel et al. (2019). Also, we relied on a single experiment with a highly controlled EC procedure which leaves open to what extent personality moderates EC in everyday life. Also, previous findings indicate that specific conditioning procedures strengthen different processes underlying EC (i.e., varying USs might enhance affect misattribution; Hütter & Sweldens, 2013; Sweldens et al., 2010), which might be considered in future research.

Second, we did not find stronger EC effects for high Extraversion that were incremental to the other Big Five traits. However, we neither found more extreme US evaluations for high Extraversion. This is inconsistent with previous findings on Extraversion and affect reactivity (Gross et al., 1998), mainly to Ingendahl and Vogel (2022), whose materials we adapted here. One possibility is that evaluating exclusively positive and negative stimuli creates a strong contrast experience where interindividual differences in affect reactivity have less influence.

Another possibility is that the relationship between Extraversion and affect reactivity is more complex. For example, Smillie et al. (2012) demonstrated that extraverts experience stronger positive affect primarily when rewards are being pursued. Our US pictures might be experienced as rewarding only when the overall frequency of positive pictures is low – for example if the evaluation task also incorporates neutral stimuli. Thus, future studies with neutral USs might reveal associations of Extraversion with US extremity and EC.

Our research shows that interindividual differences in EC are associated with the Big Five, specifically with Agreeableness and, to a lesser degree, Neuroticism. Furthermore, it shows that some of these personality differences in EC can be traced back to personality differences in specific affective and cognitive processes, namely personality differences in US extremity and pairing memory. Thus, our results demonstrate the role of the Big Five in fundamental psychological phenomena and offer essential insights into interindividual differences in attitude change.

Contributed to conception and design: MI, TV

Contributed to acquisition of data: MI, TV

Contributed to analysis and interpretation of data: MI, TV

Drafted and/or revised the article: MI, TV

Approved the submitted version for publication: MI, TV

This research was supported by the Open Access Publication Funds of the Ruhr University Bochum and a scholarship from the Graduate School of Economic and Social Sciences Mannheim to the first author.

The authors have no competing interests to declare.

All supplemental material can be found on this paper’s project page on the OSF: https://doi.org/10.17605/OSF.IO/3UJ7K

All the stimuli, presentation materials, participant data, and analysis scripts can be found on this paper’s project page on the OSF: https://doi.org/10.17605/OSF.IO/3UJ7K