Parsing Interactive Social Cognition: Mentalizing Accuracy and Propensity in Dyadic Contexts

Abstract

Successful social interaction requires attending to and accurately processing others’ thoughts and feelings, a social cognitive skill known as mentalizing. To date, however, mentalizing has most frequently been assessed in nondyadic contexts, leaving open questions as to whether variability in real‐world interactive social cognitive performance is due to trait‐level differences or to properties of particular interactions. The current study examined mentalizing about one's social partner in both close social dyads (n = 50 dyads) and stranger dyads (n = 52 dyads). Within these dyadic contexts, we measured empathic accuracy—or the ability to accurately infer another's thoughts and feelings—and mind‐mindedness, the propensity to spontaneously discuss another's mental states. We found that for both pre‐existing close dyads and stranger dyads, the empathic accuracy of one partner significantly correlated with the empathic accuracy of the other partner, suggesting that empathic accuracy may be better conceptualized as a property of the specific social interaction rather than solely an individual trait. In contrast, across both dyad types, one partner's level of mind‐mindedness did not relate to their partner's mind‐mindedness. Further, a noninteractive measure of mentalizing accuracy did not show dyadic concordance. Individual levels of empathic accuracy, mind‐mindedness, and noninteractive mentalizing accuracy were also uncorrelated. These findings underscore the importance of taking a multifaceted approach to measuring social cognition that considers the role of social context.

1. Introduction

Social interaction is core to human experience, yet many psychology studies use noninteractive paradigms to examine social cognition. Despite significant research and theoretical discussion about individual differences in social cognition (e.g., Apperly, ²⁰¹²; Bolis, Dumas, & Schilbach, ²⁰²³; Heerey, ²⁰¹⁵; Kennedy et al., ²⁰¹⁸; Meins, Fernyhough, Johnson, & Lidstone, ²⁰⁰⁶; Olderbak & Wilhelm, ²⁰¹⁷), open questions remain regarding the extent to which real‐world social cognitive performance reflects relationship‐invariant individual‐level traits or instead is influenced by dynamic, interaction‐specific experiences. Similar to what research on joint action and interpersonal synergies suggests (e.g., Riley, Richardson, Shockley, & Ramenzoni, ²⁰¹¹; Sebanz, Bekkering, & Knoblich, ²⁰⁰⁶; Tollefsen & Dale, ²⁰¹²), observed social cognitive performance is likely determined by co‐occurring processes and constraints across multiple levels and timescales. The processes impacting social cognition could include (1) relationship‐and interaction‐independent qualities of the individual (which we term independent social cognition); (2) emergent properties of the social interaction (emergent social cognition); and (3) relationship‐specific processes, such as the effects of familiarity or even seeking relationship partners who are similar to us on individual‐level traits (convergent social cognition; see Bolis et al., ²⁰²³ and Heerey, ²⁰¹⁵ for related discussion).

The dynamic interplay of these factors may also depend on how a particular component of social cognition is assessed, and such task constraints may be especially salient when considering mentalizing, or the ability to attend to and process the mental states of others. Mounting evidence suggests that different mentalizing measures show minimal relations to each other, indicating that multiple cognitive processes may be captured under the umbrella of mentalizing (e.g., Quesque & Rossetti, ²⁰²⁰; Schaafsma, Pfaff, Spunt, & Adolphs, ²⁰¹⁵; Warnell & Redcay, ²⁰¹⁹). For example, recent work by Pomareda, Devine, and Apperly (2024) suggested that mentalizing accuracy (i.e., accurate inferences about others’ mental states) and mentalizing propensity (i.e., spontaneous attention to others’ mental states) may capture different underlying processes, underscoring the importance of considering multiple tasks when measuring mentalizing.

Although mentalizing is traditionally assessed in noninteractive contexts (e.g., by reasoning about characters in third‐party stories), a growing body of research has examined relations between such mentalizing measures and dyadic interactions (e.g., McElwain, Ravindran, Emery, & Swartz, ²⁰¹⁹; Shaw et al., ²⁰¹⁷; Viana, Zambrana, Karevold, & Pons, ^{2023, 2016}). For example, Markiewicz, Rahman, Apperly, Mazaheri, and Segaert (2024) measured participants’ ability to infer mental states from third‐party videos and then had dyads complete a cooperative game, finding that both partners’ mentalizing ability influenced interactive success. Similarly, developmental work has found that when children play in pairs, both children's mentalizing abilities can impact coordination and communication (Etel & Slaughter, 2019), although other research has found that children's mentalizing abilities do not relate to mental state language produced in dyadic interaction (Bates, Lester, Nickalls, Gibson, & Fink, 2025).

Together, these studies suggest that mentalizing may impact and, in turn, be impacted by dyadic contexts, yet most of this research used noninteractive measures of mentalizing, which fail to capture the effects of the dyadic processes that emerge as interaction unfolds. Broadly, social interactions are joint actions (Garrod & Pickering, 2009; Sebanz et al., ²⁰⁰⁶) which can produce a variety of emergent properties such as motoric interpersonal synergies (Riley et al., 2011), synchronized smiling (e.g., McNaughton, Moss, Yarger, & Redcay, ²⁰²⁴), coordinated turn‐taking (e.g., Zhang, Perzolli, Esposito, Venuti, & Edelman, ²⁰²⁵), linguistic alignment (e.g., Howes, Healey, & Purver, ²⁰¹⁰), affective coregulation (e.g., Hilpert et al., ²⁰²⁰), common referential ground (e.g., Hanna, Tanenhaus, & Trueswell, ²⁰⁰³), joint attention (e.g., Kidwell & Zimmerman, ²⁰⁰⁷), and even neural alignment (e.g., Shamay‐Tsoory, Saporta, Marton‐Alper, & Gvirts, ²⁰¹⁹). Little work has investigated how multiple components of mentalizing manifest in dyadic contexts, where both individual and dyadic processes and constraints interact (cf. Ramenzoni, Davis, Riley, Shockley, & Baker, ²⁰¹¹). From this theoretical perspective, mentalizing arises from the complex, nonlinear interplay between individual processes (e.g., verbal ability, emotion recognition ability) and dyadic processes (e.g., linguistic alignment, behavioral coordination), as well as task‐specific parameters for the particular mentalizing measure (e.g., focus on accuracy vs. propensity, time‐scale of responses). Thus, in order to understand mentalizing ability, such abilities must be assessed in dyadic contexts.

In the current study, we examined interactive mentalizing in dyadic contexts for both close social dyads and stranger dyads. Participants completed written mentalizing accuracy and mentalizing propensity measures about their dyadic partner, measures which have been previously studied and operationalized in real‐world interpersonal contexts: empathic accuracy and mind‐mindedness. Empathic accuracy is the accurate identification of thoughts, feelings, and other mental states of another person (Ickes, Stinson, Bissonnette, & Garcia, 1990) and mind‐mindedness is the propensity to consider other people as entities with independent mental states (i.e., with their own thoughts, feelings, desires, and emotions; Meins, Fernyhough, & Harris‐Waller, ²⁰¹⁴).

1.1. Empathic accuracy as a measure of mentalizing accuracy in dyadic contexts

The first construct of interest to the current study—empathic accuracy—reflects the accuracy of one's mental state attributions when such attributions are explicitly solicited (Ickes et al., 1990). In the interactive version of this paradigm (Ickes et al., 1990), two participants are videotaped while they engage in a conversation. Subsequently, each member of the dyad watches the videotape, pausing at specific moments to report their own thoughts and feelings and then rewatches the tape to make inferences about their partner's thoughts and feelings. These reports (the target's self‐reported internal states and the perceiver's inferences) are then compared in order to calculate the overall empathic accuracy performance score for each member of the dyad. We note that although the term empathic accuracy is widely deployed for this task in the literature, participants make explicit verbal guesses about mental states broadly, including both feelings and propositional thoughts, akin to common mentalizing tasks (see Arioli, Cattaneo, Ricciardi, & Canessa, ²⁰²¹; Bzdok et al., ²⁰¹², Kanske, Böckler, Trautwein, Parianen Lesemann, & Singer, 2016, and Schurz et al., ²⁰²¹ for discussions about delineating mentalizing and empathy).

The empathic accuracy task has been applied in a variety of relationship contexts, including couples (e.g., Simpson, Ickes, & Blackstone, ¹⁹⁹⁵; Simpson, Orina, & Ickes, ²⁰⁰⁴; Kilpatrick, Bissonnette, & Rusbult, ²⁰⁰²), same‐sex friends (e.g., Stinson & Ickes, ¹⁹⁹²), and even strangers (e.g., Blanke et al. ²⁰¹⁵; Ickes et al., ¹⁹⁹⁰). Results indicate that the nature of the relationship between the perceiver and the target plays a crucial role in empathic accuracy performance, such that friends and dating partners are better at reading each other's thoughts and feelings compared to strangers (Stinson & Ickes, 1992; Simpson et al., ¹⁹⁹⁵). These findings are consistent with research suggesting that empathic accuracy in everyday interactions enhances relationship quality and has positive relational outcomes like emotional support (e.g., Sened et al., ²⁰¹⁷; Simpson et al., ²⁰⁰⁴), and better communication (Ickes et al., 1990; but see Simpson et al., ¹⁹⁹⁵; Ickes, ¹⁹⁹⁷).

Despite the crucial role of empathic accuracy in interpersonal dynamics, only a few studies have examined the concordance between partners’ levels of empathic accuracy. One study found that empathic accuracy scores of stable couples were significantly correlated, whereas the empathic accuracy scores of couples who ended up breaking up were uncorrelated (Simpson, Ickes, & Blackstone, 1995), and another paper found that friends, but not strangers, showed significant correlations in empathic accuracy (Stinson & Ickes, 1992). The authors argued that high empathic accuracy is not attributable to the amount of information exchanged in the interaction or to similar personalities, but instead, it is attributable to “intersubjective knowledge” that friends have acquired in their previous interactions, resulting in a feeling of closeness, consistent with convergent social processes and familiarity factors. A more recent study by Berlamont and colleagues (2023) also found that romantic partners had greater levels of similarity in their levels of empathic accuracy than would be expected in a random sorting of individuals, but did not explicitly examine real stranger dyads. Research is also limited in terms of whether empathic accuracy levels show consistent individual differences across contexts, with one study finding a cross‐target consistency in empathic accuracy, even for strangers (Marangoni, Garcia, Ickes, & Teng, 1995), and another finding that levels of empathic accuracy in real‐world interactions instead showed minimal correlations with empathic accuracy toward individuals in videos (Zikic et al., 2025).

1.2. Mind‐mindedness as a measure of mentalizing propensity in dyadic contexts

Research on dyadic mind‐mindedness is even sparser than for empathic accuracy. Mind‐mindedness has been predominately studied in parent−child relationships (Fishburn et al., 2017). Such studies have examined parental mind‐mindedness, or a parent's tendency to use mental state attributes to describe their infant or young child (Meins et al., 2006). There is compelling evidence that parents who are more mind‐minded toward their children have children with better social cognition who are more likely to be securely attached (McMahon & Bernier, 2017). However, much less is known about mind‐mindedness in adult social relationships. Importantly, although scoring schemes for mind‐mindedness in parent−child interactive contexts often include both propensity and accuracy (e.g., whether the mental state the parent inferred appeared accurate based on the infant's behavior; McMahon & Bernier, ²⁰¹⁷), studies of adult mind‐mindedness typically only score the percentage of statements that reference mental states, rather than the accuracy of those statements.

One of the few studies that investigated mind‐mindedness outside of parent−child contexts found that adult participants who were more mind‐minded when describing one close relationship (close friend) were more mind‐minded in their other close relationships (romantic partner; Meins et al., ²⁰¹⁴). In contrast, mind‐mindedness in close relationships did not correlate with mind‐mindedness when thinking about and describing famous people or works of art. This result was interpreted to mean that mind‐mindedness was a quality of close relationships rather than a trait. This study, however, only examined one‐way, or unidirectional, mind‐mindedness (i.e., how Partner A thinks about Partner B). Limited research has examined the concordance of mind‐mindedness scores within parental pairs, with mixed findings (e.g., Colonnesi, Zeegers, Majdandžić, van Steensel, & Bögels, ²⁰¹⁹; Foley, Devine, & Hughes, ²⁰²³; Lundy, ²⁰¹³). Such research, however, involves the concordance of parents when reporting on the child, rather than two partners reporting on each other.

In sum, the current literature on mind‐mindedness and empathic accuracy does not fully explore the extent to which social cognition is a property of the interaction, the relationship, or a particular individual. Some research examining other facets of social cognition has touched on these questions. Extensive research on friendships has found similarity on a variety of constructs related to social cognition, including aggression (e.g., Mariano & Harton, ²⁰⁰⁵; Poulin et al., ¹⁹⁹⁷). Additionally, studies on parent−child interactions have found relations between parental use of mental state language and children's theory of mind (e.g., Ruffman, Slade, & Crowe, ²⁰⁰²). There is also evidence of interpersonal neural similarity in friends when viewing audiovisual movies compared with more distant relationships (Parkinson, Kleinbaum, & Wheatley, 2018; although see McNabb et al., ²⁰¹⁹). Overall, however, no studies to our knowledge have directly tackled whether the accuracy and propensity of mentalizing about one's social partner show concordance across dyadic contexts.

1.3. The current study

In the current study, we examined levels of empathic accuracy and mind‐mindedness in close dyads and stranger dyads. Each dyad completed the empathic accuracy task (i.e., interacting and then guessing their partner's thoughts and feelings from a recording of that interaction) followed by the mind‐mindedness task (i.e., writing a description of their partner). We also collected one of the most commonly used measures of noninteractive mentalizing accuracy, in which participants reasoned about the mental states of third‐party individuals (Reading the Mind in the Eyes; Baron‐Cohen, Wheelwright, Hill, Raste, & Plumb, 2001), in order to examine whether this type of mentalizing ability showed convergence in close dyads and to provide a reference point for our interactive measures via a task with no clear dyadic properties.

Based on existing theories, three main patterns of results were possible for each social cognitive task (empathic accuracy, mind‐mindedness, and Reading the Mind in the Eyes). First, there might be no correlation between partners’ scores in either the close or stranger dyads. This would be evidence that a strong contribution to performance on these measures is trait‐level social cognition influenced by individual processes (i.e., individual social cognition). Second, social cognitive scores might be correlated in close dyads but not in stranger dyads. This would provide evidence for a role of familiarity and relationship‐specific processes in social cognition, potentially due to individuals either seeking out similar social partners or social partners coming to resemble each other over time (i.e., convergent social cognition). Finally, both close dyads and stranger dyads might show social cognitive concordance, where one partner's scores correlate with the other partner's scores within each dyad, suggesting the role of dyadic processes specific to that interaction (i.e., emergent social cognition). As the stranger dyads were randomly created and had no relationship history, this pattern of results would indicate that idiosyncratic properties of a particular interaction can be a strong driver of social cognitive performance. Importantly, the concordance between partners’ social cognitive abilities is separate from how high or low these abilities are. That is, a dyad in which both members score low on empathic accuracy still shows higher concordance than a dyad where one partner scores high and one partner is average.

Our core research question was what pattern of dyadic concordance would emerge across each of the three mentalizing tasks, with a particular focus on our interactive measures. Given the limited research on the topic, we did not have strong a priori hypotheses about which pattern of results would emerge, but it was plausible that the third‐party mental state inference task would show the lowest concordance levels given its noninteractive nature. Additionally, given the mounting evidence for the heterogeneity of mentalizing processes (e.g., Schaafsma et al., ²⁰¹⁵; Warnell & Redcay, ²⁰¹⁹), we also examined both individual and dyad‐level interrelations between mentalizing tasks, predicting that these tasks would not be related. Finally, we conducted a preliminary examination of potential mechanisms that could be driving dyadic mentalizing performance, including enjoyment of the specific interaction (for both close and stranger dyads) and relationship quality (for close dyads).

2. Method

2.1. Participants

The sample consisted of 50 close dyads and 52 stranger dyads (N = 204 total participants) between the ages of 18 and 31 years with an average age of 20.05 years (SD = 1.82 years). Close dyads (24 close friend pairs and 26 romantic pairs) consisted of 2 male‐male, 29 female‐female, 16 male‐female, and 3 pairs with two nonbinary members, with an average relationship length of 3.18 years (SD = 3.13 years). Stranger dyads consisted of 11 male‐male, 18 female‐female, and 23 male‐female pairs. Our sample size of approximately 50 dyads within each relationship type is comparable to other studies that have examined dyadic social cognition (e.g., Alkire, McNaughton, Yarger, Shariq, & Redcay, ²⁰²³). Close dyads were recruited via email, web postings, and flyers posted on campus, and participants received monetary compensation. Participants for the stranger dyads were recruited via flyers and via the online psychology subject pool. Those recruited via flyers received monetary compensation, and those recruited through the online psychology subject pool received course credit. We ensured that the dyads corresponded to their selected category (i.e., they self‐identified as close friends or romantic partners or were actually strangers) and that all participants were native English speakers with normal or corrected‐to‐normal hearing and vision. All participants provided informed consent, and all procedures were approved by the local Institutional Review Board.

2.2. Procedure

In the current study, both close and stranger dyads attended a 2‐h session during which they participated in a semi‐structured videotaped interaction followed by the empathic accuracy task, mind‐mindedness assessment, relationship/interaction quality assessment, noninteractive mentalizing task, and a battery of other questionnaires beyond the scope of the current study.

2.2.1. Assessing empathic accuracy

In order to assess mentalizing accuracy in dyadic contexts, we selected a widely used and well‐established paradigm: empathic accuracy (Ickes et al., 1990). We used the standard unstructured interaction paradigm (Ickes et al., 1990); however, we modified the procedure in two ways: (1) close social dyads and stranger dyads consented to being filmed prior to the start of the study, and (2) the topic of the conversation was neutral to accommodate both sets of dyads. Immediately upon arrival of the dyad at the testing space, the experimenter escorted them into the observation room, where they sat side‐by‐side. Dyads were informed that their interaction would be recorded, but the camera was set up unobtrusively behind a screen. Dyads were then encouraged to naturally engage in a conversation to discuss a neutral prompt (“How is your semester going so far?”). At this point, the experimenter left the dyad in the room and let the interaction go on for 6 min. After the interaction, the experimenter terminated the videotaping.

After the participants had completed the videotaped interaction, they were individually seated in separate rooms. Each participant was asked to independently view the videotape and to accurately and candidly report all the thoughts and feelings they experienced during the 6‐min interaction. The experimenter particularly emphasized that these were thoughts and feelings that they had while the interaction was taking place and not the thoughts and feelings which occurred to them for the first time while viewing the videotape. The experimenter also informed them that the thoughts and feelings they wrote down would not be seen by their interaction partner. The participants were instructed to stop the recording when they recalled having had a specific thought or feeling during the interaction (referred to as a “tape‐stop”). They recorded each thought or feeling by entering the time the specific thought/feeling occurred based on the digital readout (e.g., 4 min and 3 s) and the specific content of the thought/feeling entry expressed in a sentence form (e.g., “I was feeling happy about doing well on my chemistry test”).

After each partner had recorded their own thoughts and feelings based on the video interaction, the experimenter noted the time stamps of the tape‐stops for the other partner (i.e., Partner A's time stamps were written down on a form for Partner B's and vice versa). Then, each partner in the dyad was given this list of the specific times. The experimenter instructed their study partner to watch the videotaped interaction again, this time pausing the video at each of the given tape‐stops. At each tape‐stop, the partner inferred their study partner's thoughts and feelings by specifying their specific inference of their partner's thought/feeling in a sentence form (e.g., “They were feeling happy about doing well on their chemistry test”).

We employed the empathic accuracy coding system developed by Ickes et al. (1990). Two trained coders rated the similarity between Partner A's self‐reported thoughts and feelings and Partner B's inference of Partner A's thoughts and feelings. A 3‐point coding scale ranging from 0 (essentially different in content, i.e., none of the interaction partner's thoughts were correctly inferred) to 1 (somewhat the same content, i.e., some of the interaction partner's thoughts were correctly inferred) to 2 (essentially similar in content, i.e., most, if not all of the interaction partner's thoughts were correctly inferred) was employed. We then summed these values and divided by the maximum number of accuracy points that could be obtained for that interaction. That is, if Partner A reported six internal states, the maximum score that Partner B could earn was 12. If Partner B correctly guessed three states and partially guessed three states, they would earn 9 total points, or an empathic accuracy of 75%. Responses were coded by two independent coders with high levels of interrater reliability (Kripendorff's alpha = 0.78). Disagreements were resolved by the senior author.

2.2.2. Mind‐mindedness assessment

After completing the empathic accuracy assessment, participants completed an assessment of how mind‐minded they were toward their partner. We followed the procedure of Meins et al. (2014). Specifically, participants were asked to provide written descriptions of each other, measuring roughly 7−10 typed lines in response to the following prompt: “Think about the person who came in with you to the study today. Please use the box below to tell us about this person.”

These mind‐mindedness responses were coded per the manual provided by Meins et al. (2006) in order to determine the percentage of mentalistic statements (mentioning their partner's thoughts, feelings, beliefs, and desires) versus nonmentalistic statements (e.g., physical traits). Specifically, each response was split into clauses and then coded into categories: (1) mentalistic, (2) behavioral, (3) physical, and (4) general. Mentalistic statements referred to interests, beliefs, desires, knowledge, or metacognition (Meins et al., 2006). Behavioral statements referred to behaviors and observable personality traits (e.g., “plays soccer” or “laughs a lot”). Physical statements referred to physical attributes such as appearance, age, or gender. General statements were comments referred to the partner but did not fit into any of the above categories. Statements were coded into categories by two independent coders with high interrater reliability (Krippendorff's alpha = 0.77). Disagreements between coders were resolved by the senior author.

2.2.3. Noninteractive mentalizing accuracy

We also included one of the most widely used measures of noninteractive mentalizing accuracy: the Reading the Mind in the Eyes (Baron‐Cohen et al., 2001). Participants viewed 36 black‐and‐white photos of the eye region of faces and selected each depicted person's mental state from a list of four choices. We calculated an accuracy percentage for each participant. We note that this particular measure is one of the most widely used adult mentalizing assessments (e.g., Kittel et al., ²⁰²²), in spite of criticism of both its psychometric properties and its ability to capture mental state understanding (e.g., Baker, Peterson, Pulos, & Kirkland, ²⁰¹⁴; Higgins, Ross, Langdon, & Polito, ²⁰²³, Higgins, Savalei, Polito, & Ross, 2026).

2.2.4. Conversational enjoyment measures

One potential mechanism driving interactive social cognition may be enjoyment of the interaction. To examine this, stranger and partner dyads completed a questionnaire that included items about the 6‐min interaction that they engaged in at the beginning of the testing session. For stranger dyads, we assessed each participant's perceptions of the quality of the overall interaction using six questions on a 1–5 Likert scale (e.g., “How much would you want to talk to this person again?”). These measures demonstrated high relations with each other (Cronbach's alpha = 0.80), so we averaged them together into a single measure of perceived interaction enjoyment. Close dyads answered a subset of two of these questions (i.e., “How did you feel about the length of time you had to talk today?”; “How would you rate the conversation you had today?”), which we also collapsed into a single measure. Due to the fact that these questions were added later in data collection, only a subset of close dyads answered them (in 26 close dyads, both partners completed these measures, and in one close dyad, only one partner answered the questions).

2.2.5. Relationship quality measures

Our relationship measures were only assessed in close dyads. All dyads completed three measures: (1) Inclusion of the Other in the Self (IOS) scale, in which participants chose which set of increasingly overlapping circles best represented them and their partner (Aron, Aron, & Smollan, 1992; Gächter, Starmer, & Tufano, ²⁰¹⁵); (2) the emotional and intellectual subscales of the Personal Assessment of Intimacy in Relationships (PAIR) questionnaire (Schaefer & Olson, 1981); and (3) a measure of overall relationship satisfaction from The Perceived Relationship Quality Component (Fletcher, Simpson, & Thomas, 2000).

The emotional PAIR subscale is a six‐item measure that asks participants to report the extent to which a particular statement describes their relationship (e.g., “My partner listens to me when I need someone to talk to,” “I sometimes feel lonely when we're together.”) using a 5‐point scale (1 = does not describe me/my relationship at all to 5 = describes me/my relationship very well). The PAIR intellectual subscale is also a six‐item measure that asks participants to report the extent to which particular statements describe their relationship (e.g., “I feel it is useless to discuss somethings with my social partner,” “We have endless number of things to talk about”). Composite scales were calculated by averaging the scores across both subscales. Because scores on the two subscales were highly correlated (r(98) = .63, p<.0001), we averaged the subscales together for subsequent analyses. Because the Perceived Relationship Quality Component produced minimal variability, with almost 50% of participants scoring at ceiling (n = 47), we did not use this as a measure of individual differences in subsequent analyses.

2.3. Analyses

To address our core research question of social cognitive concordance across dyad types, we first examined intraclass correlation coefficients for our interactive measures of mentalizing accuracy (empathic accuracy) and propensity (mind‐mindedness), as well as our noninteractive measure of mentalizing accuracy (Reading the Mind in the Eyes) for both close and stranger dyads. Next, to examine interrelations between these social cognitive tasks, we used individual‐level analyses as well as dyad‐level analyses employing the Actor‐Partner Interdependence Model (APIM; Cook & Kenny, 2005). APIM is a widely used analytic approach that allows for simultaneous estimations of actor effects (e.g., impact of Partner A's mind‐mindedness on their own empathic accuracy) and partner effects (e.g., impact of Partner B's mind‐mindedness on Partner A's empathic accuracy). After examining relations between social cognitive tasks, we used APIM analyses to examine whether conversational enjoyment or relationship quality could explain performance on any of our social cognitive variables.

APIM analyses were conducted using the APIM Shiny Application (Stas, Kenny, Mayer, & Loeys, 2018), and all other analyses were conducted in SPSS 29.0, using scripts from Alferes and Kenny (2009) to calculate intraclass correlation coefficients.

3. Results

3.1. Descriptive analyses of interactive mentalizing measures

Overall, there was great variability in levels of empathic accuracy and mind‐mindedness in both stranger and close dyads (Table 1). Close dyad participants had significantly higher mind‐mindedness values than stranger dyad participants (t(202) = 3.64, p < .001, d = 0.51), but there was no effect of dyad type on empathic accuracy (t(202) = 1.79, p = .075, d = 0.25). There were no effects of gender or age on an individual's level of empathic accuracy or mind‐mindedness, either in the full sample or stratified by close versus stranger dyads (ps>.12), and so we did not include those variables in subsequent analyses.

Table 1.

Descriptive statistics

	Close dyads	Stranger dyads	Comparison
Empathic accuracy (%)	34.08 (18.66) 0–85.71	29.56 (17.44) 0–80.00	t(202) = 1.79 p = .075
Mind‐mindedness (%)	30.44 (19.38) 0–91.67	21.05 (17.48) 0–71.43	t(202) = 3.64 p<.001
# Mental	4.27 (3.17) 0–12	1.82 (1.83) 0–10	t(202) = 6.81 p < .001
# Behavioral	3.96 (2.93) 0–19	3.56 (2.05) 0–9	t(202) = 1.14 p = .26
# Physical	0.94 (1.68) 0–8	1.35 (1.84) 0–9	t(202) = −1.65 p = .10
# General	4.04 (2.87) 0–15	1.55 (1.70) 0–7	t(202) = 7.58 p < .001
# Total	13.21 (5.70) 3–33	8.27 (4.14) 1–21	t(202) = 7.11 p<.001

Note. Values are mean (standard deviation), followed by the range of minimum−maximum. Empathic accuracy refers to the percentage of the partner's internal states that were correctly inferred. Mind‐mindedness refers to the percentage of descriptors of one's partner that referenced mental states.

3.2. Dyadic properties of social cognition

3.2.1. Concordance in social cognition

We first examined intraclass correlation coefficients to assess dyadic concordance for our measures of interactive mentalizing. We found a strong correlation between the empathic accuracy of the two dyad members and this correlation was similar in magnitude whether partners formed a close dyad or a stranger dyad (close: ICC(1,1) = 0.51, F(49,50) = 3.10, p<.001; stranger: ICC(1,1) = 0.54, F(51,52) = 3.36, p<.001; Fig. 1a). There was no concordance, however, between levels of mind‐mindedness for either close or stranger dyads (close: ICC(1,1) = 0.18, F(49,50) = 1.44, p = .21; stranger: ICC(1,1) = 0.13, F(51,52) = 1.30, p = .35; Fig. 1b). Dyadic concordance was only present for mentalizing accuracy in interactive contexts; there was no dyadic concordance on the Mind in the Eyes for either close or stranger dyads (close: ICC(1,1) = −0.15, F(49,50) = 1.34, p = .31; stranger: ICC(1,1) = −0.13, F(48,49) = 1.29, p = .38).

Fig. 1.

Dyadic concordance in interactive mentalizing.

Note. Concordance between (a) empathic accuracy and (b) mind‐mindedness in close and stranger dyads. Partner A and Partner B labels were randomly assigned within dyads. Reference line is Partner A = Partner B (x = y). Intraclass correlation coefficients are ICC(1,1).

3.2.2. Interrelations among social cognitive tasks

On an individual level, empathic accuracy and mind‐mindedness were not related within the full sample (r(202) = −.024, p = .73), or within the subsamples of individuals who participated as part of close (r(98) = −.10, p = .34) or stranger (r(102) = −.013, p = .90) dyads. Our noninteractive measure of mentalizing accuracy (Reading the Mind in the Eyes) was not significantly correlated with empathic accuracy or mind‐mindedness in the full sample (empathic accuracy: r(197) = .057, p = .42; mind‐mindedness: r(197) = .058, p = .42), in the close sample (empathic accuracy: r(98) = .12, p = .24; mind‐mindedness: r(98) = .13, p = .21), or the stranger sample (empathic accuracy: r(97) = −.02, p = .87; mind‐mindedness: r(97) = −.04, p = .71).

We next used APIM analyses to further explore dyadic relations between our interactive mentalizing tasks, but we did not find any significant actor or partner effects such that one's mind‐mindedness was not a significant predictor of their own empathic accuracy or their partner's empathic accuracy (Fig. 2). Similarly, we did not find significant actor or partner effects in APIM models examining the Reading the Mind in the Eyes and either empathic accuracy or mind‐mindedness for either close or stranger dyads (ps > .14).

Fig. 2.

Dyadic interrelations for interactive mentalizing measures.

Note. The figure provides estimates of both actor and partner effects for both close and stranger dyads. Dyad members were treated as indistinguishable for analyses. Results indicate that one's mind‐mindedness was not a significant predictor of their own empathic accuracy or their partner's empathic accuracy. Dyadic concordance within tasks was measured using partial intraclass correlations. Coefficients are standardized beta values. 95% confidence intervals for the actor and partner effect estimates were as follows: close actor [‐.27, .11], close partner [‐.28, .09], stranger actor [‐.18, .19], stranger partner [‐.35, .02]. **p<.01; ***p<.001.

3.2.3. Exploring the impact of conversational enjoyment and relationship quality on dyadic social cognitive performance

Given evidence that interpersonal affiliation may be related to social cognitive performance (e.g., Gönültaş, Selçuk, Slaughter, Hunter, & Ruffman, ²⁰²⁰), we first examined whether our composite conversational enjoyment measure related to interactive social cognitive measures in either close or stranger dyads. Using APIM, we found there were no actor or partner effects for conversational enjoyment on either mind‐mindedness or empathic accuracy in either stranger or close pairs (ps > .1). Suggesting that these null findings were not simply due to measurement issues with the conversational enjoyment measure, we did find conversational enjoyment showed significant dyadic concordance in both stranger and close dyads (close: ICC(1,1) = 0.32, F(39,40) = 1.94, p = .04; stranger: ICC(1,1) = 0.45, F(51,52) = 2.64, p < .001). Additionally, we found that relationship quality was related to conversational enjoyment in close dyads, again suggesting that conversational enjoyment captured meaningful variability in interaction quality (see Supplementary Materials for full results).

We next examined whether relationship quality was related to either interactive mentalizing measure for our close dyads. We found no actor or partner effects of either Inclusion of Other in Self scores or the composite PAIR score on empathic accuracy or mind‐mindedness (ps > .1). Finally, we repeated our analyses with the Reading the Mind in the Eyes for both close and stranger dyads but there were no significant actor or partner effects in models examining Reading the Mind in the Eyes and conversation enjoyment or relationship quality (ps > .3).

Given that conversational enjoyment and relationship quality did not show actor or partner effects with our mentalizing measures, we conducted an exploratory follow‐up analysis to examine if the mechanisms driving empathic accuracy concordance within dyads were present when the empathic accuracy task was completed by third‐party raters viewing the dyads’ videos. We had a team of n = 34 research assistants—blind to the participants’ empathic accuracy scores and partner status—complete the empathic accuracy task for a quasi‐random selection of approximately one‐fifth of the total dyads (split between close and stranger dyads; see Supplementary Materials for full information about video selection and third‐party coding procedures). This selection of dyads had similar overall empathic accuracy to the full sample (mean empathic accuracy = 37.84, SD = 21.68) and similar dyadic concordance to the full sample (ICC(1,1) = 0.63, F(18,19) = 4.42, p = .002). Each research assistant completed the empathic accuracy task for a subset of these 19 dyads, estimating the original participants’ mental states at the corresponding tape stops.

Even when averaging third‐party ratings to account for large individual differences in empathic accuracy (consistent with extensive research on variability in third‐party empathic accuracy when watching videos; e.g., Zaki, Bolger, & Ochsner, ²⁰⁰⁸; Zikic et al., ²⁰²⁵), there was no evidence for dyadic concordance in the averaged empathic accuracy ratings of third‐party raters (collapsed dyad type: ICC(1,1) = −0.17, F(18,19) = 1.40, p = .48; close: ICC(1,1) = 0.18, F(9,10) = 1.45, p = .57; stranger: ICC(1,1) = −0.42, F(8,9) = 2.47, p = .22). At the individual level, there were also large differences in third‐party empathic accuracy for each scored dyad participant (i.e., target), with individual evaluators also showing inconsistent patterns of performance across targets (e.g., it was not the case that one evaluator consistently had the highest empathic accuracy; see Supplementary Materials for full analyses including details on the variability of third‐party ratings). Thus, taken together, these results indicate that third‐party raters, at least for these video stimuli, do not produce concordant empathic accuracy scores for dyad members.

4. Discussion

In the current study, we measured the interrelations between different components of social cognition in both close and stranger dyads. We selected one measure focused on interactive mentalizing accuracy (empathic accuracy) and one focused on interactive mentalizing propensity (mind‐mindedness), as well as a noninteractive mentalizing accuracy measure. We considered three theoretically grounded patterns of results: a minimal concordance between partners’ social cognition in both stranger and close dyads would be evidence for independent processes, dyadic concordance between social cognitive abilities in close but not stranger dyads would be evidence for convergent processes, and, finally, high levels of concordance in both close and stranger dyads would be evidence for emergent processes. We found high dyadic concordance across dyad types for interactive mentalizing accuracy, but no significant concordance for either dyad type for interactive mentalizing propensity or noninteractive mentalizing. These findings suggest that there are dyadic influences on interactive mentalizing, and these influences may show task‐specific effects, underscoring the need for future work to decompose the specific and dynamic interrelations of independent, convergent, and emergent social cognitive processes across timescales and contexts.

Our primary findings showed strong support for empathic accuracy being influenced by dyadic, emergent properties of the interaction, such that both stranger and close dyads showed high levels of concordance in empathic accuracy (i.e., if Partner A displayed high levels of empathic accuracy toward Partner B, so did Partner B toward Partner A, even if they had never met previously). We investigated whether conversational enjoyment or relationship quality (for close dyads) impacted empathic accuracy, but found that they did not. Interestingly, pilot data from third‐party observers suggest that third‐party coders do not show concordance in their empathic accuracy scores of dyad members, indicating that the interactive properties driving concordance between partners may be difficult to observe from a recording after the fact.

There are numerous candidate mechanisms that might be responsible for high dyadic concordance in empathic accuracy, even between two individuals who had never met prior to their conversation. Research on third‐party empathic accuracy (e.g., inferring mental states about videotaped individuals) has found that verbal information, from both linguistic content and other cues like prosody, is particularly important (Hall & Schmidt Mast, 2007; Zaki, Bolger, & Ochsner, ²⁰⁰⁹) and that the target's expressivity and the perceiver's empathy interact to produce variability in empathic accuracy (Zaki et al., 2008). Beyond these more individual‐level factors, emergent properties of dyadic interaction could play a role, including synchrony in facial expressions or bodily movements, joint attention, affective coregulation, linguistic alignment, smooth turn‐taking, or other physiological or neural synchrony. Future research should rigorously compare various methods for estimating these constructs, such as frame‐differencing techniques for estimating nonverbal synchrony (Paxton & Dale, 2013; Ramseyer & Tschacher, ²⁰¹⁴), as well as methods to quantify prosodic alignment and overall interactive rhythm (e.g., Fusaroli & Tylen, ²⁰¹⁶) and more recent approaches to quantifying multimodal synchrony (e.g., Grafsgaard, Duran, Randall, Tao, & D'Mello, ²⁰¹⁸; see Ohayon & Gordon, ²⁰²⁵ for review). Such studies could then use techniques such as dominance analysis to systematically determine which parameters jointly foster or inhibit empathic accuracy across dyad members or experimentally manipulate sources of information available (e.g., audio‐only interactions).

Future work should also utilize a round robin design (e.g., Bonito & Kenny, ²⁰¹⁰; Warner, Kenny, & Stoto, ¹⁹⁷⁹) to help separate out the effects of the particular participant from the emergent properties of the dyad, as the current design cannot precisely titrate the relative contribution of individual versus dyadic factors. For example, a recent study by Bates et al. (2025) with children found that dyadic partners showed high concordance in the use of mental state language when interacting, but that each child's use of such language was not stable when moving from one dyadic interaction to the next. Unfortunately, the dyads in the current study were not asked to consent to having their videos used as stimuli for follow‐up studies. Future research with new dyads should ensure that video recordings are viewable by a large pool of third‐party raters, particularly given the wide variability in third‐party empathic accuracy scores seen even within the preliminary analysis of the current dataset.

Unlike empathic accuracy, our measure of interactive mentalizing propensity (i.e., mind‐mindedness) showed no concordance in either set of dyads, providing evidence that this construct may reflect stronger contributions of trait‐level independent social cognition. In contrast to the current paradigm, most developmental research on parental mind‐mindedness has included both appropriateness and frequency (e.g., Bigelow, Power, & Dadgar, ²⁰²³, Meins et al., 2002; see Aldrich, Chen, & Alfieri, ²⁰²¹ for review). Future research should share Partner A's mind‐mindedness response with Partner B in order to get feedback on accuracy. One possibility is that such accuracy, as opposed to the percentage of mental statements, would show higher levels of concordance, consistent with the empathic accuracy findings. Alternatively, as mind‐mindedness involves a more coherent reflection on the interaction as a whole (vs. a time‐stamped inference), it may be less influenced by moment‐by‐moment emergent interactive properties, attesting to the interplay of dyadic and individual task constraints in shaping mentalizing.

Importantly, our noninteractive measure of mentalizing accuracy—Reading the Mind in the Eyes—did not show concordance in either dyad type, suggesting that beyond a distinction between propensity and accuracy, the interactive context of a mentalizing measure may alter performance. This is consistent with existing research showing limited relations between empathic accuracy in third‐person versus first‐person contexts (Zikic et al., 2025) and with increasing emphasis in social cognitive and social neuroscience research on considering the temporal dynamics and impacts of interactive processes (e.g., Redcay & Schilbach, ²⁰¹⁹; Schilbach & Redcay, ²⁰²⁵). Given concerns about the interpretability of the Reading the Mind in the Eyes in particular (e.g., Higgins et al., ^{2023, 2026}), it will be important to repeat this work using other noninteractive measures that produce variability in adult mentalizing (e.g., Movie for the Assessment of Social Cognition; Dziobek et al., ²⁰⁰⁶).

For both empathic accuracy and mind‐mindedness, as well as our noninteractive measure of mentalizing accuracy, we did not find stronger concordance in close versus stranger dyads, suggesting that the impacts of familiarity may manifest in more subtle ways. The lack of close dyad concordance for all measures is surprising given a long history of research on similarities between close social partners, including work on how perceived similarity can foster relationships (e.g., Martin, Jacob, & Guéguen, ²⁰¹³) and findings that dyadic concordance in social cognitive skills may be more predictive of friendship quality than skill levels alone (Bolis, Lahnakoski, Seidel, Tamm, & Schilbach, 2021). One possibility is of a task‐specific effect such that discussing a neutral topic placed weaker demands on familiarity‐based processes than a more emotional topic. Similarly, convergence may be higher when studying dyads who had known each other for longer time windows, as our close dyads had only known each other for an average of 3 years and were all young adults. Broadly, the interplay between independent, dyadic, and task‐based factors is likely timescale‐dependent, and studying these constructs across both short and long timescales is essential. Our cross‐sectional study cannot capture these dynamics, and longitudinal designs are needed to better capture how mind‐mindedness and empathic accuracy levels are impacted by familiarity and social context.

In addition to our core research findings about dyadic concordance, we also found no relations between our interactive mentalizing measures at either the individual or dyadic level. That is, Partner A may describe Partner B with many mental state terms but fail when it comes to accurately identifying Partner B's thoughts and feelings at specific timepoints. Similarly, our noninteractive mentalizing accuracy measure was not related to either dyadic mentalizing measure. Our findings are congruent with previous research showing that different components of social cognition tap into different adult competencies (Barreto, Fearon, Osório, Meins, & Martins, 2016; Pequet & Warnell, ²⁰²¹) and highlight the importance of considering both the specific social cognitive assays selected and the contexts in which they are deployed.

Although the empathic accuracy paradigm is widely used, one limitation is that the timescale on which participants infer thoughts and feelings is different than in the original interaction. That is, participants reported on their own mental states and guessed at their partner's mental states after the interaction had concluded. In contrast, some third‐party studies of empathic accuracy (reasoning about a video recording) have used real‐time dials to report emotional valence (e.g., Zaki et al., ^{2008, 2009}), which may allow for more contemporaneous judgments. There may be tradeoffs, however, in that explicitly asking participants to reflect on their own or another's mental state during an interaction changes how the conversation unfolds. Future research should continue to examine how to more synchronously assess dyadic mentalizing accuracy, particularly when such mentalizing is more implicit in nature.

In summary, this research breaks new ground in the study of social cognition and attests to the importance of measuring multiple facets of social cognition in dyadic contexts. Measures of social cognition that have predominately been discussed as measures of individual capacity or qualities of a long‐standing relationship may be influenced by emergent properties of a particular interaction. Our findings underscore the complex nature of social cognitive constructs in relation to real‐world social interactions.

Supporting information

Supplementary information 

Data availability statement

Data are available at: https://osf.io/mkzjf/.