This is an open access article distributed under the terms of the Creative Commons
Attribution License (To function successfully in social life, people need to form impressions of others. These impressions are functional, in the sense of helping perceivers determine who they should interact, negotiate, trade, compete, or cooperate with, and also who to ignore or even actively avoid. This insight has been articulated and refined over many years in Wojciszke and Abele's model of impression formation (Abele & Wojciszke, 2014; Wojciszke & Abele, 2008). One implication of this insight is the fundamental role of moral traits in impression formation (Wojciszke et al., 1998). As often emphasized by Wojciszke and colleagues, information about morality plays a key role in various phases of person perception because the implications for the perceivers of other traits (especially competence-related traits, like intelligence) depends on whether the target is moral or immoral (Wojciszke, 2005).
One central moral trait that people evaluate in others is trust. Again, the perceiver's interest in trust can be derived from Wojciszke's functional perspective. After all, the perceiver can directly benefit from interacting with a trustworthy agent or suffer from interacting with an untrustworthy agent. More broadly, many social relationships rely on the assumption that our partners will keep their word, act in a reciprocal fashion, and conform to norms (written and unwritten). If an individual is trusted, others are more likely to give them access to resources and cooperate with them (Baron, 2015; Nowak & Sigmund, 1998). On a higher, political level, trust allows voters to delegate certain people as leaders and representatives, thus supporting basic democratic institutions (Putnam, 2000). Economically, most daily transitions (from self-checkout lines, monthly paychecks, loans, to bank savings) require trust in institutions and individual actors (Milinski et al., 2002).
Reflecting the fundamental role of trust, many social psychologists have devoted attention to investigating its role in social impressions and behaviors. So, how do people form impressions of trust? With acquaintances, people's trust primarily reflects the history of personal interactions (Harris & Corriveau, 2011; Rezlescu et al., 2012). With strangers, there is some "default" trust (Camerer, 2003). All of this, however, is qualified by social knowledge, including reputation and stereotypes. For example, some occupations, social positions, and roles (e.g., nurses, doctors, and teachers) are generally trusted while others (e.g., politicians, car salesmen, and thieves) are generally distrusted (Fiske & Dupree, 2014). In language, many terms regarding traits contain cues about trust and the correlated dimension of warmth: a gentle teacher is trusted more than an aggressive thief (Abelson et al., 1982).
Initial evaluations of trust also rely on more peripheral features, including non-verbal cues. For example, research on "thin-slice" interactions shows that structural facial features, such as roundness, and emotional features, such as a smile, increase trust (Krumhuber et al., 2007; van’t Wout & Sanfey, 2008; Willis & Todorov, 2006). This research also shows that the trust dimension is highly related to valence (Todorov et al., 2008). There is an interesting debate on to what extent such peripheral cues relate to the actual "ground" truth, and if so, why (Li et al., 2019). Some researchers challenge any veridicality of these perceptions, and suggest that any correlation between, say, facial "trustworthiness" of a candidate and election outcomes may result from a self-fulfilling prophecy – trustworthy-looking people are elected to positions of trust (Todorov, 2017). Other researchers suggest that faces carry some signal that is related to actual behavior – people who look untrustworthy tend to be associated with more corrupt actions – but, again, this correlation could be a result of differential monitoring of "suspicious-looking" individuals (Lin et al., 2018). In any case, all this research suggests that peripheral cues make a difference and are worth studying.
Our Account: Consistency of Cues
In this paper, we review our work on how perceivers form trust judgments and behaviors from basic cues. However, in contrast to dominant approaches, we show that when evaluating trust, perceivers attend not only to the individual trust cues, but also to the internal consistency of those cues. Consequently, when individual cues fit together, the target is trusted more than when the fit between cues is low. As such, our account proposes a key role for the consistency of cues in trust impressions. Although our account deals with consistency of low-level, rudimentary cues, it fits with a broader functional perspective on impression formations. After all, the assessment of consistency ultimately serves the goal of predicting who is worth interacting with and with whom we can form coalitions and predictable reciprocal relations (Bowles & Gintis, 2011; Wojciszke & Abele, 2008).
Our account has several implications. First, it implies that the final trust judgment is more than the sum of its parts and cannot be fully predicted from the evaluative and descriptive characteristics of individual cues. As such, it challenges models that primarily link trust to specific affective or descriptive content of individual cues. Second, our account assigns a key role to subjective experiences derived from objective inconsistency. As such, our account proposes that a sense of processing fluency or of subjective coherence plays a mediating role in the perception of trust.
Next, we elaborate on the concept of consistency and related concepts such as coherence, balance, and expectancy violation. Furthermore, we discuss how a subjective sense of fluency and a sense of coherence relate to the classic psychological notion of consistency.
Consistency, Balance, Coherence, Expectation Violation
One of the classic hypotheses in social psychology is that consistency "somehow" matters for social judgments (for a review, see Gawronski & Strack, 2012). This idea is expressed in many major theories in social psychology, so it may help to present a brief overview of some of these major models.
One of the earliest explorations of consistency is Asch's (1946) work showing that personality impressions are not simply additive but a function of overall fit (consistency) between multiple traits. As such, the meaning of a trait (e.g., intelligent) changes as a function of whether it is presented in the context of other traits (e.g., cold vs. warm). In short, people form impressions as a Gestalt, with different individual elements fitting (or not) into the global impression. As we discuss later, Asch (1946) primarily saw the effects of fit as involving changes in meaning ("clever" means different things in the context of a professor or a criminal). But he pioneered the idea of thinking of impression formations as going beyond individual features.
Heider's (1958) balance theory directly emphasizes the importance of consistency in the system of multiple elements. Those elements are typically specified at the level of "triads" of people or groups (P, O, X), with signs (+/-) reflecting their mutual preference. A balanced triad is, for example, when all people like each other (P+O, P+X, O+X). When a triad is unbalanced (P+O, P+X, O-X), people experience "tension" which acts as a motivational force to restore balance. This theory is an important precursor to later developments as it suggests that pure relational, "formal" (content-independent) constraints play a role in social-cognitive processing. Of course, since its onset, the theory has been challenged by examples of triads that are "formally" unbalanced, but psychologically plausible (e.g., Bill Clinton + Hilary Clinton, Bill + Monica Lewinsky, Hilary – Monica).
The notion of consistency is, of course, at the core of dissonance theory, which proposed that two cognitive elements are inconsistent if one follows from the opposite of the other (Festinger, 1997). Though there are important differences between balance and dissonance theories – besides moving the idea of balance "inside the head", the dissonance theory also states relations in terms of logical implications. Still, the general notion is that consistency (whether in positive/negative sign, or in logical implication) matters, and that the state of inconsistency generates psychological discomfort.
More contemporary research tends to view consistency in terms of formal models (Nowak et al., 2020). Thus, consistency seeking can be thought of as a process of maximal satisfaction of multiple constraints, or an overall fit between multiple elements (Shultz & Lepper, 1996; Thagard & Verbeurgt, 1998). Consistency seeking can also be conceptualized in terms of predictive processing frameworks, where the perceiver's goal is to build and verify an internal model of the world. As such, inconsistency represents expectancy violation, or a "prediction error" (Friston, 2012; Kaaronen, 2018). In that sense, the predictive framework integrates well with the broader social cognitive notion that the actor's goal is to figure out who is worth interacting with (build a model of the social world), and the notion of trust is a form of consistency (predictability), especially in the moral domain (Wojciszke et al., 1998). We will return to these ideas later, and also to the question of whether that inconsistency "per se" produces negative tension, or the reaction is always dependent on other goals (Kruglanski et al., 2018). As we argue later, our position can be relatively easily reconciled with these recent criticisms.
Finally, while much of the current work, including the studies reviewed here, focuses on relatively low-level, elemental forms of inconsistency, some approaches examine consistency (or coherence) at higher, more narrative levels. In law, consistency within a witness' testimony enhances credibility (Pennington & Hastie, 1992). In science, credibility comes from consistency with evidence and explanatory coherence (Thagard & Verbeurgt, 1998). In life, consistency and coherence enhance meaning (Heine et al., 2006; Heintzelman et al., 2013).
Evaluative vs. Descriptive Consistency
One distinction often drawn in this literature is between evaluative and descriptive consistency. Characterizing someone as a "beautiful hater" is evaluatively inconsistent because it simultaneously contains a highly positive and highly negative characterization. In contrast, characterizing someone as a "tall dwarf" is descriptively inconsistent because one characterization implies the opposite of the other (logically, or practically, as in the case of a "slim cook"). Empirically, both evaluative and descriptive consistency can influence social judgments. For example, people pay attention to the evaluative homogeneity, as opposed to heterogeneity, of the context in which moral behavior occurs. If participants learn about a target's positive (or negative) behavior in a context where others show consistent behaviors, they form stronger moral character impressions than in a context where others show mixed behaviors (Lammers et al., 2018). Regarding descriptive consistency, banners on a website generate more trust when they are congruent versus incongruent with the content of the page (Tang et al., 2014). In terms of our framework, it does not really matter whether the source of inconsistency is descriptive or evaluative, as both kinds of inconsistency can generate "conflict" which can then lead to similar effects on the underlying processes (sense of processing difficulty, sense of unpredictability, sense of incoherence, etc.).
Objective vs. Subjective Consistency
One key distinction in the literature is that objective inconsistency should not be equated with subjective inconsistency (a felt sense of dissonance, or an experience of incoherence). For example, a person may not be subjectively bothered by simultaneously believing p and not-p (e.g., "nobody goes to that restaurant because it's too crowded"), by objectively incoherent logic (e.g., "all fish are black, sharks are black, therefore sharks are fish"), or by inconsistent beliefs and behaviors (e.g., "I am a lung cancer doctor and I smoke"). Objective inconsistency is a mismatch between how each individual element is externally assessed on its logic. However, as Asch (1946) and Festinger (1997) emphasized, inconsistency has a psychological meaning, and for a subjective experience to arise, the perceiver must personally see the elements as contradictory, and attach some weight to the issue. Indeed, research shows that what matters in cognitive conflict is the subjective inconsistency, rather than objective contradiction (van Harreveld et al., 2015).
The Affective Nature of Inconsistency
When inconsistency is noticed, and the perceiver cares about it, it may trigger a particular subjective feeling. Originally, Heider (1958) and Festinger (1997), proposed that the feeling is tension or arousal. They also often added that this arousal has a negative tinge, or that it feels bad. However, there are some debates about the issue of whether inconsistency "per se" triggers negative affect. After all, much of the dissonance literature is based on studies where the experience of dissonance and the resulting steps to reduce inconsistency could be explained by presentational and self-esteem concerns. However, there are studies where unpleasant affect was elicited by simple cognitive inconsistencies, such as neutral sentences with low-probability endings (Levy et al., 2018) as well as studies where positive affect, including its physiological manifestations, can be triggered by simple prime-target matches (Winkielman & Cacioppo, 2001). The relatively automatic, spontaneous, and early-stage nature of the consistency-detection process is suggested by studies where participants showed facial responses (measured via EMG) to conceptual (in)consistencies, even in tasks where there was no judgment – just passive reading of neutral words (Topolinski et al., 2009). Also, Topolinski and Strack (2015) showed immediate negative facial responses to surprising trivia statements, again outside of any mental elaboration or judgmental task.
Still, some researchers point to situations where inconsistencies, or violations of expectations feel positive (Kruglanski et al., 2018). Indeed, many “surprises” bring us great joy – from unexpected guests at birthdays to surprisingly good news from the doctor. Consider an example of a plane arriving an hour earlier than the posted schedule. The only feeling a traveler may notice is joy from the positive outcome. From our perspective, this is not deeply problematic as any negative tinge from "expectancy violation" is often dwarfed by the pleasure of the positive outcome (e.g., getting home early, a surprise birthday guest, unexpectedly good test score, etc.). Furthermore, there is a difference between simple violations of expectations (surprise party, earlier arrival) that can be accommodated in our world model, and inconsistencies that more fundamentally challenge our world understanding. We will return to this issue later.
This debate about the nature of feelings from cognitive dissonance mirrors discussions, held since James (1890), about the nature of "cognitive experiences" that communicate overall quality of mental processing. In much current research, the broad term for such an experience is fluency – a sense of processing efficiency. Indeed, fluency is assumed to underlie more specific "structural experiences" such as a sense of coherence, integrity, fit, or rightness (Reber et al., 2002; Topolinski & Strack, 2009; Whittlesea, 2002; Winkielman, Ziembowicz, & Nowak, 2015). It must be said though that the relation between the subjective experience and the actual processing efficiency (e.g., measured by reaction time) is complex. Some evidence points to an occasional mismatch between an actual processing speed and subjective feeling of fluency (Reber et al., 2004). Other evidence highlights the possibility that low-level factors increasing objective processing speed (e.g., efficient coding) may enhance impressions without first translating into any subjective sense of effort, or subjectively experienced negative affect (Ryali et al., 2020). While these issues are complex, there is certainly plenty of evidence that the sense of fluency (and the resulting affect) is used to make judgments, even if it induced by incidental, background stimulus features (Winkielman, Ziembowicz, & Nowak, 2015; Ziembowicz et al., 2013).
In short, the theoretical reasoning guiding our research predicts that (i) inconsistency serves as an input into trust judgments, (ii) subjective, rather than objective inconsistency, is a predictor of judgments, (iii) evaluative judgments, including trust, decrease with greater subjective inconsistency, and (iv) even incidental inconsistency can influence judgments. In the next section, we present several lines of research consistent with these postulates.
Empirical Work: Consistency of Simple Text Cues and Visual Cues Increases Trust
We have conducted two experiments that suggest that consistency of simple cues increases trust (Nowak et al., under review). The first experiment shows this in the domain of simple text cues. The second experiment shows this in the domain of visual cues.
Experiment 1
In our first experiment, we examined the effects of verbal coherence in the description of individuals on trust toward these individuals. Each description consisted of a pair of an adjective and a noun. Our main hypothesis was that coherent adjective noun adjective pairings would result in higher trust than incoherent pairs.
We first ran a pilot study to generate consistent and inconsistent descriptions of target individuals. To generate consistent descriptions, we gave pilot participants a number of adjectives, or adjective phrases, for example “fat” or “pale”, and asked them to write a noun or a noun phrase that would fit the adjective. The most frequent parings, e.g., “fat cook”, or “pale anemic”, were used as the coherent pairs. To create incoherent phrases, we used the antonyms of the adjectives or adjective phrases, e.g., “slim cook” or “tanned anemic”.
The materials obtained in the pilot study were then used in the main study, with three separate groups of participants, and each group evaluating a different aspect of stimulus materials (to assure independence of each type of evaluation). The first group judged the degree of trust towards the target described by each pair of words (coherent and incoherent). The second group of participants rated the subjective coherence of pairs (degree of subjective fit of coherent and incoherent pairs). The third group of participants was used to isolate the main effects of valence of the nouns and adjectives and trust generated by them from the interaction effect of a pair's objective coherence. To do so, we asked these participants to rate the valence of each individual adjective and noun, and the amount of trust associated with each individual adjective and each noun. Objective coherence was defined as the degree of similarity between ratings of individual adjectives and individual nouns on dimensions of valence and trust.
The analysis of results revealed that the subjective coherence of adjective-noun pairs strongly predicted trust. Subsequent mediation analyses have shown that the trust toward a described individual depended on the subjective coherence of adjective-noun pairs, in addition to trust ratings of individual adjectives and nouns. Subjective coherence, in turn, depended on the objective evaluative coherence (valence match between the noun and the adjective). Interestingly, subjective coherence did not depend on the valence, or trust associated with the adjective and the noun, neither on the objective coherence of trust toward the adjective and the noun. In sum, the mediation models have shown that subjective coherence of noun-adjective pairs was a strong predictor of trust, and that it depended on the match in objective valence. When subjective coherence was included in the model, the only variables directly influencing trust were trust associated with the noun, and trust associated with the adjective.
In conclusion, this experiment suggests that adding a positive, but mismatching adjective to a neutral or a negative noun can introduce incoherence, which may paradoxically decrease trust toward the object of judgment. For example, even though the adjective "slim" is rated as more positive and trustworthy than the word "fat", a "slim cook" is trusted less than a "fat cook."
Experiment 2
Our second experiment focused on the effect of visual coherence on trust judgments. The experiment specifically tested if visual consistency influences trust even if it comes from incidental, background sources. To create a set of visual stimuli of varying coherence, we chose a set of pictures containing a person and changed them in different ways to introduce incoherence. For instance, a picture of a businessman in a suit at the office was edited to show the same businessman positioned on a beach. A nurse in the hospital, in the coherent condition, was presented in the incoherent condition as a nurse in a desert. As another method, incoherent elements were introduced into the picture (e.g., impossible figures), lighting was changed to incoherent sources of light, or one part of the face would be sharp and another part of the face would be soft.
The whole experiment consisted of three individual participant samples. The first participant sample rated both the original pictures and the modified ones for coherence. The second sample rated trust toward the person in the picture on a four-item Likert scale. In the third sample, the measure of trust was the amount invested in a trust game if it was played with the person in the picture.
The analyses were done on a picture-by-picture basis, where the mean ratings of consistency of the picture assessed by the first sample were used as the predictor. The analyses have revealed that subjective consistency significantly predicts trust both as measured by the trust scale and by the amount invested in the trust game. Additionally, we found (unsurprisingly) that some individuals (e.g., firefighters) are more trusted than others (e.g., businessmen).
In sum, the results of both experiments show that coherence operationalized as subjective consistency of verbal descriptions (Experiment 1) or visual displays (Experiment 2) uniquely predicts trust in the target, regardless of whether trust is measured by a single Likert scale, a multiple-item scale, or behavior – investment in a trust game. Additionally, both experiments have shown that a unit of subjective consistency is worth about a unit of valence and trust – as indicated by the relative strength of each statistical predictor. In fact, consistency can be a stronger predictor of trust than valence, the factor traditionally regarded as fundamental to trust. The second experiment has shown that this relationship holds even if coherence concerns accidental visual background information. Additionally, both experiments have shown that the trust also depends on the trust toward the categories of individuals, but it operates via a different causal route than subjective coherence.
Empirical Work: Consistency of Facial Expression Cues Increases Trust
As mentioned earlier, one source of initial impressions about trustworthiness are facial features. These impressions are tightly connected to emotion perception. For example, people judge a face as more trustworthy when it has features resembling smiles and less trustworthy when it has features resembling anger (Oosterhof & Todorov, 2008). However, as we argue here, people’s evaluative judgments, including trust, depend not only on individual features but also on the consistency between these features. If the features are inconsistent, or ambiguous, this creates disfluency (processing difficulty), which then lowers evaluative judgments.
These effects have been initially demonstrated in the face domain, using dimensions such as gender and ethnicity (Halberstadt & Winkielman, 2014; Owen et al., 2016). Typically, participants favorably evaluate mixed-gender (Male-Female) or mixed-ethnicity individuals (e.g., Asian-Caucasian), presumably reflecting the "beauty-in-averageness" effect. However, such mixed individuals lose their appeal if participants first need to categorize targets by the relevant category (e.g., gender or ethnicity). This decline reflects the fact that mixed individuals are now more difficult to process (less fluent) than non-mixed individuals.
Later work extended this to emotion stimuli, such as facial expression (Olszanowski et al., 2018; Winkielman, Olszanowski, & Gola, 2015). For example, participants viewed pictures of male and female faces. The picture ranged from an image of pure anger, through an image of mixed emotion (blended anger and happiness), to an image of pure happiness. Participants' task was to judge the individuals in the images on a dimension of attractiveness or a dimension of trust. Before giving their judgments of each individual, participants were first asked to categorize it either on a dimension of emotion (angry vs happy) or a control dimension (e.g., gender, male vs female). Results showed that participants judged mixed emotion stimuli (angry-happy blends) as relatively less trustworthy (and also less attractive) when they first needed to categorize the stimulus on the emotion dimension. That is, there was an evaluative cost associated with emotional ambiguity (incoherence). Unsurprisingly, the ambiguous stimuli were also categorized more slowly (longer RT), indicating lower processing fluency. Importantly – an implicit, physiological measures of affective responses (facial EMG) showed decreased smiling to those ambiguous and disfluent stimuli, suggesting a genuine reduction of positive affect. Finally, categorization fluency (RTs) statistically mediated the impact of stimulus ambiguity (mixedness) on evaluative ratings.
A recent study used EEG (electroencephalography) to better identify the brain mechanisms involved in the processing and evaluation of ambiguous emotional faces and to examine how these brain mechanisms relate to social impressions (Kaminska et al., 2020). Analyses of event related potentials (ERP) derived from EEG recordings revealed several interesting effects. First, emotionally ambiguous faces elicited greater amplitude of LPP (late positive potential) in frontal-medial sites. This response occurred around 560–660 ms, so relatively late in processing chain, especially considering that face detection is associated with an early N170 potential, and is modulated by emotional expressions around 200ms after stimulus onset. Second, enhanced responses to emotional ambiguity were obtained only in the emotion categorization task, but not in the control task (gender categorization). This means that ambiguity was not detected spontaneously. Third, and critically, these LPP responses from the frontal-medial cluster correlated with the reduction in trustworthiness judgments. Psychologically, these EEG results suggest that ambiguity (inconsistency) detection involves late, top-down processes that are integrated with processes underlying judgments of trustworthiness. Our preferred interpretation is that this frontal midline LPP reflects the mental effort of resolving ambiguity. In line with this possibility, Sun et al. (2017) found greater LPP to faces when participants had to categorize the faces on the ambiguous dimension. Aided by fMRI localization data, they interpreted these findings as reflecting cognitive effort generated by sources related to dACC, dmPFC and IFG – brain areas typically seen in studies on mental effort.
Faces, Bodies, and Context
The research we just discussed examined the inconsistency of the facial features composing an individual facial expression (e.g., a face containing mixed features of happiness and anger). However, it is worth noting that the phenomena describe here extend to other types of emotional inconsistency.
One type of emotional inconsistency is between different nonverbal channels (body, face, voice, etc.). People can quickly detect incongruence between emotional cues derived from the body and the face, such as seeing an individual with a happy body and an angry face (Meeren et al., 2005). Such face-body conflicts reduce trust, as suggested by a recent work from our lab (Rotteveel et al., under review). Future work may explore these ideas with other kinds of nonverbal inconsistencies and examine whether they represent low-level or higher-level processes. This is interesting because a classic cue to lying is inconsistency between different nonverbal channels (Uviller, 1993). Usually, perceivers' use of this cue is explained by their belief that truthful information leaks out via a less monitored channel. However, it is possible that the underlying process is more rudimentary and essentially involves perceivers experiencing low-level processing conflict. This idea would explain why perceivers sometimes experience a feeling or intuition of "wrongness", without being able to give a reason for this intuition.
It is also worth acknowledging that in real life nonverbal expressions containing mixed emotions in the face, or even across the face and the body, could be tolerable because people do occasionally experience mixed emotions (e.g., Larsen & McGraw, 2014). In fact, perceivers sometimes interpret mixed emotions as a sign of emotional complexity (Grossmann et al., 2016). However, one clear case of inconsistency occurs when an emotion displayed by the target plainly conflicts with context. This was recently explored in a study where participants viewed short movie clips with individuals being happy (vs. sad) during funerals or sad (vs. happy) during weddings (Kastendieck et al. 2021). Participants' evaluative responses showed not only that they perceived the person's emotionally inconsistent behavior as less appropriate, but also, that they had a lower sense of closeness with the individual. It would be interesting to extend this research to judgments of trust.
Implications
The research reviewed here examined the role of consistency in trust evaluation. We described several experiments on how people create trust impressions from brief verbal descriptions (pairs of words characterizing the target), visual scenes (pictures of targets on different backgrounds), and faces varying on gender, ethnicity, identity and emotion, as well as with their fit to other nonverbal channels (body). In all these studies, consistency positively predicted judgments of trust. These experiments highlight several key features of consistency.
First, the research with noun-adjective pairs shows that the negative effect of inconsistency is sometimes strong enough to override the benefits of positive verbal characterization. The experiments with visual scenes show that the target-background fit may be as important as the features of the target itself. In the research on pure and mixed faces, objectively happier but less featurally consistent faces are sometimes rated as less trustworthy than objectively angrier but more featurally consistent faces. All these three lines of research suggest that knowing the degree of consistency of the target sometimes predicts trust as well as or even better than knowing the mean valence of individual components. This is an important insight and goes beyond the content-related factors so far emphasized by current theoretical models of trust judgments. This fits with the classic insights from psychology that in impression formation the whole is not a simple sum of the contributing parts, but functions more like a "gestalt." It is important to emphasize again that our conclusion here is different from Asch (1946) who proposed that a mismatch can change the meaning of the pair (e.g., adding a positive adjective "clever" to a thief makes the compound target "clever thief" more negative than "dumb thief" because the former is more dangerous). We agree with this original insight as well as with the more contemporary work, discussed earlier in this article, on how the personal implication of competence traits (clever) depends on moral traits (thief), as elaborated in the functional account of impression formation (Abele & Wojciszke, 2014). However, we propose that at least some of the negative impression of "clever thief" is due to mere evaluative inconsistency of the combination, which creates processing difficulty and may serve as an independent source of negative affect and less favorable trust evaluations.
In short, our argument here is that impressions of a multidimensional target (e.g., a slim cook; a complex visual scene, a mixed emotional expression) reflect a fit between individual features. Why is such "mere fit" important? There are several theoretical possibilities, each with different assumptions about the underlying mechanisms and implications for boundary conditions.
First, as we have argued, some consistency effects may reflect processing effort (Winkielman, Ziembowicz, & Nowak, 2015; Zürn & Topolinski, 2017). Essentially, inconsistency is just a way of creating processing difficulty, which then leads to negative affect. This idea implies that if inconsistency is not disfluent, perceivers are not bothered by it.
Second, some approaches posit that consistency is closely related to the notion of prototypicality and frequency (Ryali et al., 2020). Slim cooks are atypical, office workers on the beach are unique, mixed-emotion or mixed-gender faces are relatively rare. As such, all such inconsistencies could generate a dislike via a standard association between typicality, familiarity, and positive affect (Zajonc, 1968). This idea implies that if inconsistency does not imply atypicality and rarity, perceivers should be fine with it.
Third, some authors argue that consistency or coherence reflects predictability. For example, slim cooks and accountants on the beach are surprising, as are features of a negative emotion on a happy face. There are multiple accounts of why unpredictability is bothersome. Mechanistically, unpredictability comes with a high energetic cost of updating a model of the world (Clark, 2013), which seems closely related to the idea of fluency. Epistemically, unpredictability shows that perceivers do not have a good model of their social world, which frustrates their goal of achieving certainty (Kruglanski et al., 2020). Related to this is the idea of inconsistency (or incoherence) leading to a particular type of causal reasoning (slim cooks probably do not like their own food, Thagard & Verbeurgt, 1998). All this contributes to a general sense that a perceiver cannot easily form a mental model of the target and thus cannot predict the target’s behavior – something that is essential to the basic meaning and function of trust. One implication of this approach is that inconsistencies that do not have implications for the perceiver's model of the world should be well-tolerated (Kossowska et al., 2018). Similarly, when the need to predict the world is low (e.g., when a person is planning to watch a movie and actively avoids any spoilers), inconsistencies should not evoke negative affect either (Kruglanski et al., 2020).
Once again, we emphasize that the predictive coding framework fits well with the functional perspective on social cognition where the actor's goal is to figure out who is worth interacting with – build a model of the social world (Wojciszke & Abele, 2008). From the functional perspective, one component of trust is a sense of consistency (predictability, reliability), especially in the moral domain. In other words, one goal of moral cognition is to figure out stability and reliability in the behavior of others (Bowles & Gintis, 2011), and also in one's own behavior, in the sense of consistency between one's own behavior and values (Wojciszke, 1987).
Of course, on the functional perspective, the other component of trust is forming impressions of others' prosocial goals, or at least the commonality of their goals with the perceivers' goals (even when the goals are wicked, see Gambetta, 2000 on mafia trust). In that sense, when we say that we "trust others", we say that we assume their cooperative intentions, at least when it comes to us. These two components of trust (predictability and prosociality) can be in interesting opposition. In the current review, we argue that perceivers do not trust inconsistent people, even at the cost of other positive traits. Our argument resembles a recent line of research showing that participants are willing to pay a price and forgo advantages just to have a consistent, predictable working or interaction partner (Gerten et al., 2022). Somewhat paradoxically, people may experience more "trust" in interactions and negotiations with predictable enemies, than with unreliable "friends". But clearly sometimes actors who are less "stable" but more "prosocial" will be trusted more. For example, a colleague who flipped from an enemy to a friend will probably be rated as more trustworthy than a colleague who is predictably antagonistic (so, trustworthy in the sense of consistency). In a similar example, people will probably trust a thief who returns a stolen wallet (inconsistent act), more than a thief who, predictably, keeps the stolen wallet. Future research will be needed to examine the scope of the account presented here. For example, it is possible that inconsistency matters less in contexts where target and behavior valence is unambiguously negative (so there is no attempt to assess or build trust). It will also be important to examine the extent to which the present account applies to different forms of trust, including epistemic trust (Koenig & Harris, 2007) and interpersonal trust (Fu et al., 2015; Li et al., 2020).
Conclusion
In this article, we described several lines of research on the determinants of trust in social impression. One inspiration for our work was the research on the central role of moral categories in impression formation (Wojciszke, 2005; Wojciszke et al., 1998) and the functional perspective on morality in social cognition (Abele & Wojciszke, 2014; Wojciszke & Abele, 2008). Our contribution to this rich literature is that some elements of trust depend on subjective consistency – a sense of fit between individual elements. As such, we described studies showing that trust impressions are not a simple sum of the contributing parts, but reflect a “gestalt.” Most broadly, our findings fit large integrative frameworks emphasizing the role of fluency of processing, and predictive coding in social cognition, which continue to generate insights in many domains of psychology.