Introduction

Emotional expressions surround us in everyday life—from a neighbor smiling at us on our way to work to an angry motorist in a traffic jam. However, not all emotional expressions are genuine. People can produce facsimiles of genuine emotional displays for strategic purposes, thereby deceiving others about their true emotions. The ability to determine if an emotional display is genuine or deceptive is vital in interpersonal communication. The current study aimed to identify individual traits predict this ability. Understanding the factors that may impede (or foster) accurate emotion perception in naturalistic settings is key to getting an accurate understanding of real-life emotion perception.

Different theoretical accounts can provide a basis for individual differences in emotion authenticity perception. The Basic Emotion Theory (BET; Ekman, 1992) assumes that people recognize basic emotions based on specific facial features, which are produced automatically and are discrete and “universal”. This framework predicts categorical differences between genuine and deliberate expressions due to the specific activation of facial muscles (i.e., categorical differences based on “reliable muscles”; Ekman, 2003), and a better ability to “read” such emotional cues will confer benefits to authenticity discrimination. This implies that such skills allow the individual to detect inconsistencies in emotional expressions if the sender is being disingenuous (i.e., producing a deceptive display). By contrast, the Behavioural Ecology View of emotion perception (BECV; Fridlund, 1991) proposes that emotional displays are social tools for situation-specific communication—in other words, they are flexible and goal-directed. Importantly, this framework does not treat authenticity as a function of facial configuration but one of sender intent. With respect to perception, individuals would differ in their ability to interpret social cues during communication.

The two theories imply different accounts of emotion authenticity perception, although the distinction is one of emphasis rather than mutual exclusivity. Authenticity judgments likely draw on both visible facial cues and the social context in which expressions are produced. From a BET perspective, authenticity detection can be framed primarily as a perceptual cue-detection problem: genuine and deliberate expressions may differ because genuine emotions are expected to produce characteristic facial configurations or reliable expressive markers. By contrast, the BECV places greater emphasis on facial displays as flexible social signals whose meaning depends on sender intent, production context, and communicative function (for a discussion, see Zloteanu & Krumhuber, 2021). In the present study, this distinction is important because authenticity was operationalized according to how the expression was produced, rather than according to the presence or absence of a particular facial configuration. Accordingly, the stimuli were defined by production method; accuracy was coded according to whether observers correctly classified expressions as genuine or deliberate as defined by sender intent, and individual differences in emotional awareness, empathy, and socio-perceptual mental-state inference were examined as predictors of variability in observers’ judgments of dynamic displays.

The scientific literature strongly points to three constructs of individual differences related to emotion perception: alexithymia, empathy, and Theory of Mind. Alexithymia is a trait associated with deficits in emotional awareness, manifesting as difficulties in identifying and describing one’s own emotional states, as well as recognizing emotions from facial displays in others (Jongen et al., 2014). There is a lack of studies examining the relationship between alexithymia and the ability to distinguish genuine versus deliberate emotional displays. Past studies primarily focused on individuals with Autism Spectrum Disorder and individuals with clinical and non-clinical eating disorders, often employing static emotional stimuli (cf., Brewer et al., 2015; Cook et al., 2013; Fujiwara, 2018; Ridout et al., 2010). Furthermore, little is known about if/how alexithymia affects the perception of realistic, dynamic emotional expressions.

Empathy has been found to relate to emotion recognition ability and the affective perception of the self and others, but care is needed with the interpretation of such results, as findings have been mixed and dependent on different subtypes of empathy (Besel & Yuille, 2010; Felisberti & Cropper, 2025; Konrath et al., 2014). This may be due to stimuli being too unrealistic and therefore unrelated to real-life emotion perception. Our study aims to measure perceptions of more realistic and authentic emotions to disambiguate previous findings.

Lastly, the ability to correctly attribute emotional states and beliefs to others, referred to as “Theory of Mind” (Premack & Woodruff, 1978), shows overlapping neural underpinnings with emotion perception (Mitchell & Phillips, 2015). Theory of Mind refers to the ability to attribute mental states to others (Premack & Woodruff, 1978). It develops between 3 and 5 years of age (Baron-Cohen et al., 1985; Wimmer & Perner, 1983), and individual differences in this trait persist up to adulthood (Baron-Cohen et al., 2001). There has been a replication crisis of implicit Theory of Mind tasks recently (Kulke & Rakoczy, 2018; Poulin-Dubois et al., 2018), possibly due to the unrealistic posed stimuli used in their paradigms (see Kulke et al., 2024, Packer & Moreno-Dulcey, 2022). Due to the relevance of Theory of Mind for emotion perception, individual differences should be studied with more realistic stimuli.

Such findings point to the need to study individual differences in the perception of authentic facial expressions of emotions. Moreover, research relating these traits to emotion recognition must consider realistic and dynamic expressions in their investigations, which are important for understanding emotion judgments (Zloteanu et al. 2018, 2021a, b).

The individual-difference measures were selected because they plausibly relate to different components of this judgment process. Theory of Mind should be relevant because authenticity judgments require observers to infer the sender’s likely mental or communicative state from expressive behavior. Empathy may support these judgments if affective responsiveness or perspective taking facilitates sensitivity to another person’s emotional display (cf. Zloteanu et al. 2021a, b). Conversely, alexithymia may hinder authenticity judgments if difficulties identifying and describing emotional states are associated with poorer interpretation of others’ expressions. These predictions are especially relevant under a BECV account because the task is not simply to identify the emotion category displayed, but to judge the likely authenticity and production context of the signal.

Our research considers the importance of accurately understanding the emotional state of others as an essential component of social interactions. It remains unclear how these three abilities relate to authenticity judgments, given the rather mixed and imprecise estimates of past research. Hence, it was important to explore the underlying relationships predicting accurate inferences of dynamic facial expressions to permit the construction of models that consider the elements of facial displays that play a role in judgment and in modulating performance. Because these traits were measured rather than manipulated, we treat them as associational predictors of judgment variability rather than as causal determinants of authenticity detection.

As per our pre-registered hypotheses (https://osf.io/fntr8), we expected empathy would be positively related to emotional authenticity judgments (H1) and that Theory of Mind would be positively related to emotional authenticity judgments (H2). In contrast, we expected the two alexithymia scales, General Difficulty Describing Feelings and General Difficulty Identifying Feelings, would be negatively related to emotional authenticity judgments, while no association was expected between the scale General-Externally Oriented Thinking and the emotional authenticity judgments (H3). We also expected differences in people’s emotional authenticity accuracy between the different expression types (deliberate and genuine) (H4). Finally, the three individual difference measures were predicted to relate to emotion perception judgments relating to perceived expression intensity, judgement confidence, and perceived genuineness of expression (H5). In sum, we aimed to investigate individual differences in perception of realistic genuine or staged expressions.

Method

Transparency and openness

We report how we determined our sample size, all data exclusions, all manipulations, and all measures in the study, and the study follows JARS (Appelbaum, et al., 2018). All data and analysis code are available at https://osf.io/dtrx6. Research materials are cited above. This study’s design was preregistered prospectively, before data were collected; see https://osf.io/fntr8/overview? view_only=cb30aaace7b6481db2c266219ecdeec6.

Participants

Six hundred and five (N = 605) participants were recruited using the Prolific platform (Mage = 26.63, SDage = 8.98; Male = 330, Female = 265, Prefer not to say = 10). No additional demographic information was recorded in line with GDPR recommendations for sensitive data. The sample size was estimated a priori, considering a small-to-moderate effect size, Cohen’s f = 0.02, at the traditional alpha value of 0.05, and 80% power. Participants were paid £3.50 ($4.70) for their time. The inclusion criteria were that individuals had to be at least 18 years of age and provide consent for all aspects of the study.

Materials

Three questionnaires were selected due to their psychometric properties (high reliability and validity) and their established use within the field, permitting transferable inferences across similar research.

Interpersonal reactivity index (IRI)

This questionnaire was developed by Davis (1983) to evaluate prosocial individual traits. It contains 28 self-assessed statements that contribute to four independent subscales: fantasy, empathic concern, personal distress, and perspective taking. It uses 5-point Likert-type items for ratings. The Fantasy subscale is often used as an index of empathy and identification with fictional situations and characters and the ability to be transported into narrative worlds (e.g., “I really get involved with the feelings of the characters in a novel”). The Personal Distress subscale is used to assess one’s own feeling of anxiety and unease in uncommon personal interrelationships (e.g., “In emergency situations, I feel apprehensive and ill-at-ease”). The Empathic Concern subscale assesses one’s empathy levels towards others (e.g., “I am often quite touched by things that I see happen”). The Perspective Taking subscale is used to assess the capacity to infer the points of view of others (e.g., “I believe that there are two sides to every question and try to look at them both”) and can be thought of as an important component of the Theory of Mind. The following Cronbach’s alpha (95% CI in brackets) were obtained: Fantasy (α = 0.80 [0.77, 0.82]), Personal Distress (α = 0.71 [0.68, 0.75]), Perspective Taking (α = 0.78 [0.75, 0.80]), and Empathic Concern (α = 0.84 [0.82, 0.86]).

Perth alexithymia questionnaire (PAQ)

The PAQ questionnaire is a self-report measure with 24 items used to assess alexithymia, a construct characterized by a subclinical inability to describe and identify one’s own emotions. It was designed by Preece and colleagues (2018) and it uses 7-point Likert-type items for ratings, from 1 for “strongly disagree” to 7 for “strongly agree”, whereby higher values indicate higher levels of alexithymia. PAQ has three constructs assessing Difficulty Identifying Feelings, Difficulty Describing Feelings, and Externally Oriented Thinking (EOT). Because DIF and DDF have valence-specific scores, PAQ has five subscales: Negative-DIF (N-DIF; 4 items, e.g., “When I’m feeling bad, I can’t tell whether I’m sad, angry, or scared”), Positive-DIF (P-DIF; 4 items, e.g., “When I’m feeling good, I can’t make sense of those feelings”), Positive-DDF (P-DDF, 4 items, e.g., “When something good happens, it’s hard for me to put into words how I’m feeling”), Negative-DDF (N-DDF, 4 items, e.g., “When something bad happens, it’s hard for me to put into words how I’m feeling”). The fifth subscale is the EOT and has 8 items (e.g., “I don’t pay attention to my emotions”).

The score of all PAQ questions can be used as an overall marker of alexithymia (ALEXI) and the five subscales can be combined into composite scores (not used in this study). The Cronbach alpha for the five subscales was: N-DIF, α = 0.86 [0.85, 0.88]; P-DIF, α = 0.83 [0.81, 0.85]; N-DDF, α = 0.86 [0.84, 0.88]; P-DDF, α = 0.87 [0.85, 0.88]; EOT, α = 0.90 [0.89, 0.92]; and ALEXI, α = 0.94 [0.94, 0.95].

Reading the mind in the eyes test (RMET)

The RMET (Baron-Cohen et al., 2001) is a relatively simple visual test widely used across different cultures to assess adults’ ability to recognize and understand the mental states of others from minimal cues arising from changes in expressions around the eyes alone. The test has 36 black and white photographs of the eye area of adult males and females. Each photograph is accompanied by four emotion words. For each set of eyes participants are asked to choose the word that best describes what the person in the picture is feeling or thinking. Although this test can be criticised for its simplicity, it offers various advantages for the current study, as it is widely used, can be used cross-culturally (Vellante et al., 2012) and is independent of verbal abilities (Baron-Cohen et al., 2001), making it particularly suitable for our study which aimed to recruit a large and diverse sample.

Dynamic emotional stimuli

The dynamic emotional stimuli consisted of video clips of surprise expressions developed and validated in previous work (Zloteanu et al. 2018, 2021a, b). The clips depicted encoders responding to a jack-in-the-box paradigm under four elicitation conditions: Genuine, Internal, External, and Rehearsed. In the Genuine condition, encoders’ spontaneous reactions were recorded as the toy unexpectedly released. In the Internal condition, encoders first experienced the event and were later asked to reproduce the expression by focusing on their remembered internal emotional state. In the External condition, encoders studied another person’s genuine reaction and then reproduced the observed behavioral display. In the Rehearsed condition, encoders reproduced their own earlier reaction after having experienced the genuine event. All clips were recorded in color, presented without sound, and lasted approximately 10 s.

Design and measures

The study used a repeated-measures individual-differences design, with repeated-measures variables for the type of expressions participants see, Genuine, Rehearsed, Improvised, Internal, and External, and the scores on the psychometric tests for empathy, Theory of Mind, and alexithymia as the individual difference characteristics. The dynamic emotion stimuli were taken from Zloteanu et al. (2018, 2021a, b). The videos have multiple subsets, each using slightly different instructions for how the senders should produce their displays, permitting an exploration of factors influencing others’ perception of expression authenticity. For a full description, see the above-mentioned studies.

The responses to the dynamic expressions were coded to create accuracy and bias (genuineness) dependent variables. Accuracy and genuineness were measured on a 5-point scale (note that this diverges from the preregistration that erroneously specified a 7-point scale). We used the same 5-point scale as previous research using the same stimuli (Zloteanu et al. 2018, 2021a, b). For accuracy, the responses on the authenticity scale were collapsed into three possible responses: 1–2 as “deliberate”, 3, as “don’t know”, and 4–5 as “genuine”. These scores were matched to the veracity of the expressions. If the rating and the expression veracity matched (e.g., the expression was genuine and the rating was “genuine”), then a score of + 1 was given; alternatively, a score of 0 was given. These scores were then summed across exemplars in each expression condition, producing per-condition accuracy scores. For genuineness (which serves as an ad hoc response bias measure), the scale values were recoded as -2 and − 1 for “deliberate”, 0 for “don’t know”, and + 1 and + 2 for “genuine”. These scores were summed across all expression conditions to produce a genuineness score.

Procedure

The study was conducted using the online Qualtrics platform. Participants were presented with a series of dynamic expressions – 10 from each of the 4 subsets (Genuine, Rehearsed, Improvised, Internal, and External; randomized) – and asked to judge each expression on several dimensions: expression intensity, expression authenticity, judgement confidence, perceived trustworthiness, and perceived attractiveness (all using a 5-point Likert-type scale). Following the videos, participants were presented with the three psychometric tests (in randomized order). Finally, they were debriefed.

Analysis plan

Data analyses were performed in the open-source JASP 0.19.1 software (JASP Team, 2024)Footnote 1, using a Bayesian Model Averaging (BMA) hypothesis-testing approach. For each outcome, BMA evaluates the full set of candidate models formed by all possible combinations of the specified predictors. Evidence is then averaged across this model space rather than being based on a single selected model, allowing predictor support to be quantified using inclusion Bayes factors (BF). Model selection is done on the basis of model Bayes factors (BFM) and per-predictor inclusion Bayes factors (BFincl.); absent strong theoretical insights, we relied on Jeffreys-Zellner-Siow (JZS) ‘default’ testing priors (i.e., multivariate Cauchy, r-scale = 0.50, covariate r-scale = 0.35) for the models. All models were run with at least 2,000 samples for numerical accuracy and 400 posterior samples for estimation to ensure an acceptable error rate under 10% for all BFs (van Doorn et al., 2021)Footnote 2. A Bayesian Analysis of Covariance (ANCOVA) was employed with the within-subjects factor of expression type, and RMET score, alexithymia score, and empathy (subscales) as predictors.Footnote 3

The models considered each of the four dependent variables separately. One ANCOVA considered the accuracy scores for the expression task; additional models considered the genuineness scores, intensity, and judgement confidence. Main effects for the condition factor were followed up with pairwise contrasts. All variables entered into the regression analysis were centered. The model used Bayesian contrasts to ensure equal weight was assigned a priori to all factors. For testing decisions, BFs > 3 or < 0.3 are considered as evidence for the H1 or H0, respectively. For interpretability, we report estimates from the highest posterior model, while model-averaged inclusion evidence is used to evaluate predictor support. To characterize model uncertainty, we additionally report the posterior support for the next highest-supported model when relevant. Full model-comparison output is available in the accompanying OSF materials.

Results

Descriptive associations

We provide a table with Bayesian Pearson’s correlations between the covariates used in our models (BFs use the default Beta(1,1) uniform prior over r ∈ [–1, 1]). While not pre-registered or used for subsequent inferences, given the large sample size collected, we consider these to have academic value for future research. See Table 1.

Table 1 Bayesian pearson correlations among individual-difference measures
Full size table

Expression classification accuracy

An ANCOVA was conducted on the classification accuracy scores of participants. Accuracy was treated as the ability to correctly detect that the facial displays were either genuine (i.e., a reaction to the jack-in-the-box) or deliberate (i.e., voluntarily produced in the absence of the jack-in-the-box). For each video subset, the accuracy can range from 0 to 10. The BMA procedure ran N = 128 model combinations for the predictors, model averaged R2 = 0.54, 95% CrI [0.50, 0.56], and produced strong evidence for the model containing Expression Type (BFincl. > 999.99) and RMET (BFincl. > 999.99) as optimal in explaining the variance in the data, receiving the most posterior probability after observing the data, P(M|data) = 0.47, and a substantial increase from the prior odds, BFM = 113.40:

$$\:Accuracy\:\sim\:ExpressionType\left(factor\right)+RMET\left(centred\right)$$

Our inferences are based on selecting this single model for interpretation of effect sizes and relationships, single model R2 = 0.54, 95% CrI [0.52, 0.57]. For comparison, the next highest-supported model only had a posterior probability of P(M|data) = 0.10 and differed only by the inclusion of IRI-PD, indicating that the main model uncertainty did not concern RMET. For a full list of competing models, see the JASP output files in the OSF project folder.

As there are 10 stimuli per set, we present the model-predicted accuracy (proportion correct) data as percentages (%) for easier interpretation. The intercept (grand mean) indicates an estimate of 62.4% (SD = 0.5%) correct identifications overall (i.e., task accuracy). Genuine facial displays were recognised with higher accuracy, + 21.2% points (pp.), 95% CrI [19.99%, 22.42%], over the deliberate expressions, while External expressions, + 2.30pp., 95% CrI [1.15%, 3.48%], were classified with higher accuracy than average, suggesting these were easier to identify as non-genuine. The other deliberate expressions showed poorer classification accuracy (i.e., more difficulty in correctly identifying the displays as non-genuine), Improvised, -2.81pp., 95% CrI [-3.97%, -1.65%], and Rehearsed, -20.66pp., 95% CrI [-21.87%, -19.44%]. The difference between conditions (pairwise contrasts) all indicated definitive evidence for accuracy differences (all BF10 > 999.99, after Bayesian multiple testing correction). This can be seen in Fig. 1.

RMET showed a large positive relationship with accurate authenticity discrimination of + 8.70pp., 95% CrI [+ 6.80%, + 10.70%] per scale unit point increase. See JASP output for details (https://osf.io/dtrx6).

Fig. 1
Fig. 1
Full size image

BMA posterior raincloud plot for facial display classification accuracy (count) based on Expression Type. Accuracy refers to the number of correctly classified expressions within each expression type, ranging from 0 to 10. Correct classification means identifying Genuine expressions as genuine and deliberate expressions as deliberate. Higher scores therefore indicate greater classification accuracy, not stronger ratings of genuineness

Expression genuineness

An ANCOVA was conducted considering ratings of expression genuineness, where positive values indicate participants believed the expressions were produced genuinely, while negative values indicate they believed they were deliberately produced. The result of the BMA of the 128 models produced a model averaged R2 = 0.75, 95% CrI [0.50, 0.79], and produced strong evidence for the model containing Expression Type (BFincl. > 999.99) and RMET (BFincl. > 999.99) as being optimal in explaining the data, receiving a posterior probability of, P(M|data) = 0.52, and a substantial increase from the prior odds, BFM = 136.09:

$$\:Genuineness\:\sim\:ExpressionType\left(factor\right)+RMET\left(centred\right)$$

Using the above model for inference finds a single model R2 = 0.75, 95% CrI [0.50, 0.79]. The next highest-supported model had a posterior probability of P(M|data) = 0.10 and differed only by the inclusion of PAQ. The model intercept indicates an overall negative genuineness perception of the expressions (Mgrand = -0.25, SD = 0.18), which is expected given the imbalance in conditions. Genuine facial displays were rated as appearing very genuine, 12.10, 95% CrI [11.79, 12.40]. Both External, -6.96, 95% CrI [-7.29, -6.64], and Improvised facial expressions, -5.42, 95% CrI [-5.75, -5.08], were rated as appearing less genuine, while Rehearsed expressions produced more uncertainty, 0.28, 95% CrI [-0.01, 0.60]. All pairwise differences had definitive evidence for a difference in genuineness (BF10 > 999.99, null controlled), see Fig. 2.

Scores on the RME test showed a (medium-to-large) negative relationship with the strength of ratings for genuineness, -0.17, 95% CrI [-0.23, -0.10]. Thus, people with high ToM tend to be less extreme in their decisions (i.e., make more conservative ratings).

Fig. 2
Fig. 2
Full size image

BMA posterior raincloud plot for ratings of genuineness based on Expression Type

Expression intensity

An ANCOVA was conducted considering ratings of expression intensity. The result of the BMA produced a model-averaged R2 = 0.55, 95% CrI [0.49, 0.63]. Unlike the previous models, the strongest evidence was for the model containing Expression Type, RMET, and PAQ, P(M|data) = 0.24, BFM = 41.04. Yet, the inclusion BF does not support PAQ as a stable predictor, BFincl. = 1.62. As such, for parsimony, we focus on the simpler model that includes only Expression Type (BFincl. > 999.99) and RMET (BFincl. > 999.99), which still has strong support, BFM = 254.32, and a substantial increase in the posterior odds, P(M|data) = 0.17:

$$\:Intensity\:\sim\:ExpressionType\left(factor\right)+RMET\left(centred\right)$$

The single model had an R2 = 0.61, 95% CrI [0.59, 0.63]. The model intercept indicates an overall high (positive) rating of expression intensity (Mgrand = 28.71, SD = 0.19). Genuine facial displays were rated higher in intensity, +2.78, 95% CrI [2.49, 3.00]. And as with genuineness, both External, -0.03, 95% CrI [-0.29, 0.24], and Improvised, -3.47, 95% CrI [-3.75, -3.19], were rated as appearing less intense, but the former is too uncertain. Rehearsed expressions were rated as appearing more intense, +0.75, 95% CrI [0.51, 1.01]. All pairwise differences had definitive evidence for a difference in intensity (BF10 > 665.21, null controlled), see Fig. 3.

The RME test shows a (small-to-medium) negative relationship with ratings of expression intensity, -0.16, 95% CrI [-0.23, -0.09]. Again, this might indicate an overall more conservative and tempered judgment process.

Fig. 3
Fig. 3
Full size image

BMA posterior raincloud plot for ratings of expression intensity based on Expression Type

Judgement confidence

Finally, an ANCOVA was conducted considering judgement confidence. The result of the BMA of the 128 models produced a model averaged R2 = 0.62, 95% CI [0.50, 0.65], and strong evidence for the model containing Expression Type (BFincl. > 999.99) and the Fantasy Scale (FS) of the IRI (BFincl. = 147.27) as the best model, receiving a posterior probability of P(M|data) = 0.30, and a substantial increase from the prior odds, BFM = 55.03:

$$\:Confidence\:\sim\:ExpressionType\left(factor\right)+I{RI}_{FS}\left(centred\right)$$

The single model had an R2 = 0.62, 95% CI [0.60, 0.64]. The next highest-supported model had a posterior probability of P(M|data) = 0.15 and differed only by the inclusion of IRI-PD. The intercept indicates an overall high level of confidence (Mgrand = 38.18, SD = 0.19). As with the other metrics, Genuine facial displays were more confidently rated, +1.96, 95% CrI [1.73, 2.20]. All deliberate expression conditions were rated with less confidence by comparison; External, -0.25, 95% CrI [-0.49, -0.00], Improvised, -0.64, 95% CrI [-0.90, -0.39], and Rehearsed, which had the lowest ratings, -1.07, 95% CrI [-1.29, -0.83].

The pairwise comparisons showed evidence for differences in confidence ratings between Genuine and all deliberate conditions, External (BF10 > 999.99), Improvised (BF10 > 999.99), and Rehearsed (BF10 > 999.99). External also showed a difference from Rehearsed (BF10 = 288.96). There was insufficient evidence to determine a difference between Improvised and either External (BF10 = 0.61) or Rehearsed expressions (BF10 = 0.68). See Fig. 4.

Fig. 4
Fig. 4
Full size image

BMA posterior raincloud plot for confidence ratings based on Expression Type

The IRI-FS sub-factor (Fantasy scale) shows a (small) positive relationship with judgement confidence, 0.13, 95% CrI [0.07, 0.20]. People who are more likely to transpose themselves into fictional situations rate their judgments as more accurate.

Discussion

The current study aimed to measure individual differences in the judgement of expression authenticity based on individual differences in RMET, empathy, and alexithymia. Participants’ ratings differed on whether expressions were genuine or deliberate. Higher scores on the RMET—which we used as an index of socio-perceptual mental-state inference—were associated with higher accuracy, lower ratings of genuineness, and lower ratings of expression intensity. Participants scoring higher on the IRI-FS fantasy scale were also found to be more confident in their judgments.

Our results corroborate H2, demonstrating that RMET is positively related to emotional authenticity judgments. However, H1 and H3 were not supported, as authenticity judgments were independent of empathy and alexithymia. As expected, there were differences in people’s emotional authenticity accuracy based on expression types (deliberate and genuine) (H4), replicating Zloteanu et al. (2018, 2021a, b). Genuineness, accuracy, and intensity of expression judgments were related to RMET, while confidence in judgments was only related to empathy (fantasy scale), partially corroborating H5.

The current study is the first to disentangle individual differences in expression authenticity perception in a large sample using genuine and deliberate dynamic expressions. These can be related to perceptions of expression genuineness, intensity, and even relate to judgement confidence. We also replicate findings that demonstrate how crucial it is to differentiate between different deliberate expressions in emotion research, as ratings and perceptions will differ based on the type of expression and the method used to produce it. Regarding authenticity, confidence was higher in the rating of genuine compared to deliberate expressions, in line with previous research (Zloteanu et al. 2018, 2021a, b), and shows that genuine emotions are processed differently by observers (even if these do not translate to more accurate ratings).

The theoretical contribution of the study is to clarify which individual differences appear relevant to authenticity judgments when expressions are dynamic and defined by their production method. The findings do not aim to test BET versus BECV. The findings aim to constrain how these frameworks are to be applied to authenticity perception. If authenticity detection were primarily supported by broad socio-emotional traits, one might expect empathy and alexithymia to show consistent associations with judgement. Instead, these measures showed negligible relationships. The strongest association was with RMET performance, suggesting that authenticity judgments may depend more on socio-perceptual inference from facial cues than on broad empathic disposition or self-reported emotional awareness. In this sense, the results support a more selective account of individual differences in authenticity perception: observers differ in their judgments, but not all traits commonly linked to emotion perception explain this variability.

RMET was related to accuracy, response bias, and perceptions of expression intensity. Looking closely at this pattern suggests that higher RMET was related to more conservative appraisals. This may be due to people with higher RMET scores having a greater understanding of the complexities and nuances of human expressive behavior, which may lead to a tempered judgement. As noted above, the RMET is a ToM measure that mainly focuses on emotion perception (see e.g., Vellante et al., 2012; Oakley, 2016). It may therefore particularly relate to the expression judgments measured in the current study. We decided to use the RMET measure as it is independent of verbal skills and widely used internationally, but it is not without its issues.

We highlight that the association of RMET should not be interpreted as evidence that broad Theory of Mind causes better authenticity detection, given the substantial limitations of the tool. Although the RMET is widely used as a measure of Theory of Mind, it is not a process-pure measure of broad mentalizing ability. Successful performance may depend on emotion recognition, sensitivity to subtle facial cues, vocabulary, and the ability to map facial information onto mental-state terms. This is particularly relevant in the present study because the authenticity task also required participants to judge dynamic facial expressions. The observed association may therefore reflect shared socio-perceptual demands across tasks rather than Theory of Mind specifically. Accordingly, we interpret the RMET effect as evidence that individuals who perform better on a facial mental-state inference task also show better, and more conservative, authenticity judgments. Future work should test whether this association remains when using broader Theory of Mind measures (e.g. Happé, 1994; Turner & Felisberti, 2017).

Alexithymia was unrelated to any expression judgement effects, with evidence from the Bayesian approach corroborating its negligible association. This is consistent with clinical work arguing for it being related to internal experiences more than external appraisal (Donges & Suslow, 2017). Alexithymia can be viewed as a theoretically valid experimental control (i.e., covariate) to consider in emotion research, but the current data does not suggest it provides much benefit. It should be noted that a diverse sample was recruited, which included only a few individuals who would meet the clinical threshold of alexithymia; therefore, an investigation of clinical cases may be interesting for future work.

Empathy was also unrelated to expression judgment accuracy, bias, or intensity, and its negligible effect was supported by the analyses. This contrasts with previous findings in the literature (Besel & Yuille, 2010; Konrath et al., 2014). While empathy is typically assumed to relate positively to more accurate judgments of emotional expressions, often research is not adequately set up to investigate such an association, as they employ small samples and fail to account for effect uncertainty (Maxwell, 2004). Additionally, previous research has also suggested that empathy relates more to speed of processing than to the judgement itself (Kosonogov et al., 2015). Thus, it may be that empathy is not an individual difference that relates to emotion authenticity judgments (Zloteanu et al. 2021a, b).

Interestingly, empathy was related to confidence in judgement: the higher the empathy scores on the fantasy scale, the more confident participants were in their judgments, even though their judgments were not more accurate. This finding was unexpected. One speculative explanation is that as individuals with higher scores tend to transpose themselves into fictional situations (such as posed emotion scenarios), they may rate their ability to judge others’ real emotional states more highly. Conversely, individuals with low empathy fantasy scores may be less confident as they perceive themselves as less capable of judging genuineness.

The current results are difficult to interpret under the BET (Ekman, 1992) framework, which argues for categorical differences between genuine and deliberate expressions (i.e., reliable muscles, Ekman, 2003), and a better ability to “read” emotional cues conferring benefits to authenticity discrimination. This was not the case in our data. However, because the present study did not code facial muscle movements directly, the findings do not rule out cue-based contributions to authenticity judgments (cf. Namba et al., 2021).

The pattern of results is more naturally aligned with a BECV of emotion perception (Fridlund, 1991), in which observers evaluate dynamic expressions as socially produced signals rather than as readouts of discrete internal states. Under this view, facial expressions are signals meant to communicate one’s affective state, and as such the concept of “authenticity” is related to the intent of the sender. Thus, individual differences in emotion perception and socio-cognitive skills are only beneficial if the sender is attempting to portray (send) specific emotions (signals). They would not confer any “decoding” benefits to spotting hidden information. Making predictions under the appropriate framework is important for constructing valid models for inference.

The results also support the importance of treating deliberate emotional displays as heterogeneous. Participants did not respond to all non-genuine expressions in the same way; accuracy, genuineness, intensity, and confidence differed across expression production types. This pattern is compatible with a Behavioral Ecology perspective in which facial displays are understood as social signals shaped by the sender’s goals, experience, and context. On this view, authenticity judgments are not simply judgments about whether a display contains the correct facial configuration, but judgments about the likely origin and communicative status of the display. Nonetheless, the findings do not rule out cue-based accounts outright. Rather, they suggest that future theories of authenticity perception must integrate both the production method of the expression and the observer-level differences that support interpretation of dynamic social signals.

Implications

The current data provide large-sample (precise) estimates for often-used individual difference measures in emotion research. The goal was to provide researchers with reliable estimates for theory of mind, alexithymia, and empathy when considering new investigations into emotion recognition and emotion authenticity judgments. The current analyses corroborate the pattern of judgments observed between genuine and deliberate dynamic expressions of surprise and extend the research to consider the associations with theoretically relevant individual differences.

Of all measured variables, RMET showed the strongest association with authenticity judgements. Individuals with better socio-perceptual mental-state inference skills were more accurate at identifying realistic emotional expressions as deliberate and showed more tempered judgments of expression genuineness and intensity. This is particularly interesting, as many ToM paradigms seem to be unsuitable for truly measuring Theory of Mind (for an overview, see Kulke & Rakoczy, 2018; Poulin-Dubois et al., 2018). As many Theory of Mind paradigms use actors posing, rather than genuine expressions (see Packer & Moreno-Dulcey, 2022), these paradigms may be confounded: based on the current study, people scoring higher on ToM-related skills may be more likely to identify the paradigms as fake. They may therefore show less ToM in unnatural paradigms despite their higher skills. Our study, therefore, provides one reason for the current replication crisis in ToM (for an overview, see Kulke & Rakoczy, 2018; Poulin-Dubois et al., 2018). Paradigms for measuring ToM and for measuring emotion perception should therefore be designed using genuine and naturalistic stimuli or real-life scenarios.

A practical implication is that researchers should be cautious when assuming that commonly used individual-difference measures will generalize across emotion perception tasks. Empathy and alexithymia are often theoretically linked to emotion recognition, but the present results suggest that their relevance to authenticity detection may be limited, at least for dynamic surprise expressions of the kind used here. Conversely, the association with RMET performance indicates that tasks involving facial mental-state inference may share important variance with authenticity judgments. This distinction matters for theory building: rather than treating emotion perception ability as a unitary construct, future work should distinguish between emotion categorization, authenticity detection, confidence, perceived intensity, and response bias.

Limitations and future directions

The present preregistered study used a large sample and dynamic stimuli to provide precise estimates of the associations between individual differences and emotional authenticity judgments. However, several limitations should be considered when interpreting the findings and when extending them to future work.

First, the interpretation of the RMET association requires caution. Although the RMET is widely used as a measure of Theory of Mind, it is not a “process-pure” measure of broad mentalizing ability. Performance may also reflect emotion recognition, sensitivity to subtle facial cues, vocabulary, or the ability to map facial information onto mental-state terms. This is especially relevant here because the authenticity task involved judgments of facial expressions. The observed association may therefore reflect shared variance in facial emotion perception or socio-perceptual processing, rather than Theory of Mind specifically. Future research should examine whether similar associations are found using broader Theory of Mind measures and non-facial mentalizing tasks.

Second, the present study focused on dynamic facial expressions of surprise. Surprise is relatively ambiguous in valence and can be interpreted positively or negatively depending on context. As such, empathy and alexithymia may play a larger role in authenticity judgments for emotions with clearer affective or motivational meaning, such as anger or fear. The present results should only be taken as evidence that these traits did not meaningfully explain authenticity judgments for the dynamic surprise expressions used here. Future work should examine whether the contribution of these traits varies across emotion categories, particularly those differing in valence, social salience, and motivational relevance.

Third, although the present study used dynamic facial expressions, emotion authenticity in everyday life is often conveyed through multiple channels. Vocal prosody, body movement, interpersonal context, and real-time interaction may all shape authenticity judgments. Future studies should examine multimodal and interactive displays, as well as expressions varying in intensity, spontaneity, and production context.

Fourth, the generality of the findings should be examined across populations and cultural contexts. The present sample was non-clinical, so it remains unclear whether the observed associations would differ in populations with higher levels of alexithymia, reduced empathy, or other socio-cognitive difficulties. Cultural variation is also likely to matter, as display rules, norms for emotional expression, and expectations about authenticity differ across social contexts. This is particularly relevant under a BECV, in which emotional displays are understood as socially situated signals.

Finally, the finding that fantasy empathy was associated with greater confidence, but not greater accuracy, raises questions about calibration in authenticity judgments. Future work could examine whether individuals who report greater imaginative engagement are more likely to feel confident in their interpretations of emotional displays despite showing no corresponding improvement in performance. This may help distinguish factors that support accurate authenticity detection from those that increase subjective certainty.

Conclusion

In summary, this preregistered study provides large-sample evidence that judgments of genuine versus deliberate expressions are systematically associated with RMET performance (an index of socio-perceptual mental-state inference), but not with alexithymia or empathy. The differing perceptions of the three deliberate expression types challenge the strong Basic Emotion Theory account in which authenticity depends primarily on reliable facial configurations, suggesting instead that production method can generate displays that appear genuine to varying degrees. By recruiting a large sample and applying Bayesian model averaging, we provide precise estimates that clarify which individual-difference measures do, and do not, explain variability in authenticity judgments, while highlighting the importance of dynamic stimuli and theoretically explicit operationalization of authenticity in emotion perception research.