Theory of mind development from adolescence to adulthood: Testing the two-component model

The ability to infer mental and affective states of others is crucial for social functioning. This ability, denoted as Theory of Mind (ToM), develops rapidly during childhood, yet results on its development across adolescence and into young adulthood are rare. In the present study, we tested the two-component model, measuring age-related changes in social-perceptualandsocial-cognitiveToMinasampleof267participantsbetween11and 25 years of age. Additionally, we measured language, reasoning, and inhibitory control as major covariates. Participants inferred mental states from non-verbal cues in a social-perceptual task (Eye Test) and from stories with faux pas in a social-cognitive task (Faux Pas Test). Results showed substantial improvement across adolescence in both ToM measures and in the covariates. Analysis with linear mixed models (LMM) revealed speciﬁc age-related growth for the social-perceptual component, while the age-related increase of the social-cognitive component fully aligned with the increase of the covariates. These results support the distinction between ToM components and indicate that adolescence is a crucial period for developing social-perceptual ToM abilities.

Theory of Mind (ToM) denotes the ability to represent and understand mental states, intentions, and feelings of others, and to predict one's own and others' behaviour. ToM is fundamental for social interactions and adaptive social behaviour (for a meta-analysis see Slaughter, Imuta, Peterson, & Henry, 2015). Recent studies suggest that an improvement in ToM abilities occurs at any age, including adulthood, provided ToM stimulatingexperiences, such as social and conversational inputs (Peterson & Wellman, 2018;Pyers & Senghas, 2009). While the vast majority of ToM research considers early childhood, adolescence is a likewise important period for socio-emotional development (for reviews, see Fuhrmann, Knoll & Blakemore, 2015;Steinberg, 2005). During adolescence, social interactions represent the key context for the construction and manifestation of self-and social understanding (Carpendale & Lewis, 2004;Zerwas, Balaraman, & Brownell, 2004). The social world of adolescents becomes richer and wider and relationships outside the family become more important (Collins & Laursen, 2004;Larson & Verma, 1999).
Alongside evidence for parallel age-related improvements in ToM and in cognitive functions, there are stable inter-relations between these constructs across development (Rakoczy, 2017). Different aspects of language including semantics, syntax, and pragmatics have been identified as significant covariates of ToM development, particularly in verbal ToM tasks (e.g., Astington & Jenkins, 1999;Harris, de Rosnay, & Pons, 2005; for a meta-analysis see Milligan, Astington, & Dack, 2007). ToM tasks further require coordination and suppression of different perspectives (e.g., own versus other) which may explain the involvement of executive functions (e.g., inhibitory control) in ToM performance (Rakoczy, 2017). Moreover, individual differences in reasoning account for some, albeit small, variability in ToM (for a meta-analysis see Baker, Peterson, Pulos, & Kirkland, 2014). According to the twocomponent model, different cognitive functions are involved for the social-perceptual and the social-cognitive ToM component, and different developmental trajectories are postulated (Tager-Flusberg & Sullivan, 2000). However, empirical evidence is scarce and results are mixed, leaving an ambiguous state of evidence for the two ToM component model (Osterhaus et al., 2016).
Development at different rates would be evidenced in favour of the two-component account, since this would indicate involvement of different functions for either component. Second, development of each component should be separable from development in relevant general cognitive functions, since this suggests that a ToM component comprises a complex of specific ability.
The present study is an inquiry into the two putative ToM components. We aimed at testing whether there are different developmental trajectories of social-perceptual and social-cognitive ToM across adolescence into young adulthood, while controlling for agerelated change in language, executive functions, and reasoning as major covariates of general cognitive development with potential links to ToM (see above). Our approach is a first attempt to test the two-component model, since either component can claim an own domain of ability only if its development does not fully align with general cognitive development. It is also explorative in nature, since the age-dependency of socialperceptual and social-cognitive ToM measures has so far not been addressed across adolescence into young adulthood.

Participants
In total, 293 volunteers between 11 and 25 years of age participated in the study. The participants were recruited in middle-class schools and at the university via leaflets, informative letters, and emails. All participants were native speakers of the German language, and 21 of them were bilingual. All bilingual participants started to learn German language before 4 years of age. None of the participants reported severe head injuries or other impairments concerning perception, hearing, and cognitive and mental functions including psychiatric disorders, learning disabilities, or ADHD. The age-continuous sample consisted of 9-23 participants per year of age, having gender distribution balanced across ages. After outlier clearing, N = 267 valid cases, 175 female, entered statistical analysis. Detailed descriptions of the sample and outlier clearing methods are provided in the Supporting information.
Prior to the study, all potential participants and parents of adolescent participants were informed in written form about the study aims, methods, sources of funding, any possible conflicts of interest, and institutional affiliations of the researchers. Only participants who had returned their written agreement to be contacted about the study were included in the sample. Subjects received small monetary compensation or course credit for participation. According to the Declaration of Helsinki, written informed consent was obtained from all participants; for adolescents, consent was also obtained from their parents.

Eye Test (ET)
The German version of the Reading the Mind in the Eyes Testrevised ( Baron-Cohen et al., 2001;B€ olte, 2005) was used to measure participants' attribution of complex emotional and mental states based on non-verbal cues from the eyes. Participants were given a handout with definitions of all emotional concepts and were instructed to study them before the task started and to ask questions whether clarification was necessary. During the task, the handout was available and participants were alerted to using it. A total of 36 greyscale photographs of the eye region of different actors were presented to participants consecutively. Each photograph revealed a complex emotional or mental state, such as 'thoughtful' or 'worried'. With each photograph, four adjective descriptions of complex emotions or states were presented, one of them matching the expression depicted.
Participants were asked to select the adjective that matched the depicted expression best. The presentation was self-paced, and each photograph was presented until the participant responded. The proportion of correctly identified photographs was measured.

Faux Pas Test (FPT)
Faux pas recognition in conversation was tested using the Social Stories Questionnaire (SSQ, Baron-Cohen et al., 1999;Lawson, et al., 2004). Faux pas recognition requires (1) noticing that different story characters have different knowledge, (2) appreciating the emotional impact of a statement, and (3) knowledge about social norms in different social contexts (Baron-Cohen et al., 1999;Osterhaus et al., 2016). The SSQ consists of 10 short social episodes from different social contexts, each falling into three sections.
Across all stories, ten sections contain a blatant target utterance, ten contain a subtle target utterance, and ten contain no target utterance. For each section, participants indicate whether it contains faux pas and, if so, identify the utterance by marking it (out of 4-6 utterances). The number of correctly identified faux pas out of 20 is measured. A paperpencil version of the test is available at https://www.autismresearchcentre.com/arc_tests.

Language (LA)
A standard treasury of word test was used to measure verbal IQ (MWT-B, Lehrl, 2005). In 37 trials, participants are asked to detect the only real word from a five-word sequence that contains 4 artificial words. Test duration was~5 min. Detailed information on the MWT-B can be found in Lehrl (2005). The proportion of correctly identified words was measured.

Inhibitory control (IC)
Despite its suggested relevance to social functioning, executive functions are predominantly studied in non-social domains (Hill, 2004). To investigate inhibitory control as a core component of executive functions, the use of social stimuli may be more appropriate, particularly in adolescence (e.g., A€ ıte et al., 2018). In the present study, we used social stimuli in a perceptual competition paradigm (Schmitz, Cheng, & De Rosa, 2010) as a test of the ability to ignore irrelevant objects while focusing on target faces. The participants categorized face gender in compound face-house images. Two graded levels of opacity were applied on the distractor stimulus (35 and 65%). When faces and houses are overlaid in transparency, successful face gender categorization is possible only if the irrelevant house object can be efficiently suppressed. The task comprised two levels of opacity (low and high), two orientations (upright and inverted), and 30 replications, resulting in 120 experimental trials. Stimulus presentation was self-paced. The proportion of correct gender categorizations was measured.

Reasoning (RE)
A short version of Raven's standard progressive matrices task (Raven, 2000) was used to measure abstract non-verbal reasoning. In this test, all trials have a visual-geometric design with a missing piece. Subjects choose one out of eight elements to complete the matrix. Stimulus presentation lasts until the participant makes a selection. Forty matrices are presented, ordered by difficulty. There is an overall time limit of 10 min for the entire test. Participants are informed about the time limit prior to testing. The proportion of correct responses (including all trials that were not carried out due to the time limit) was measured.
Since proportion correct was measured in all five tests, higher test scores indicate better performance.

Procedure
All tests were computer-administered, conducted at the university laboratory in two onehour sessions held on two consecutive days. Participants responded by clicking the left or right computer mouse button or by mouse clicks on pre-defined arrays on-screen. Prior to testing, they were asked to carefully read the instructions and were free to ask questions. For each test, except for the Faux Pas Test, 5-10 practice trials were provided to familiarize participants with task requirements and response procedures. Up to three participants were tested at a time in the same experimental room, separated by movable partition walls. A research assistant was permanently present to supervise the procedure and to answer questions. Inquisit 4.0 (Millisecond Software, Seattle, WA, USA) was used for programming computer-based test administration.

Ethics statement
The experimental procedures for the project 2017-JGU-psychEK-009 (Johannes Gutenberg University) were approved by the local ethics committee of the (Johannes Gutenberg University) University.

Descriptive statistics
Statistical analyses used z-scores. Note this implies same means for Eye Test and Faux Pas Test, but reveals potentially different distribution across age. Moving averages for a fullyear integration window (see Figure 1, solid curves) illustrate an overall flatter course of Faux Pas Test scores compared to Eye Test scores. The latter reflected lower levels in early adolescence, followed by a steeper rise across teenage years. Table 1 shows zero-order Pearson correlations. Most tests correlated strongly, while the Eye Test and the Faux Pas Test correlated modestly, but significantly (r = .25***, r rank = .20**). All tasks showed strong correlations with age (rs ≥ .3, ps ≤ .001), pointing to age-related growth as a major common driver behind task correlations. The partial correlation of both ToM tests, controlled for age, fell short of statistical significance and indicated a negligible amount of shared variance, r xy.age = .112, t(265) = 1.92, p = .055; r 2 xy:age = :0125.  term, follow-up analyses with linear models (LM models) reveal the specific covariate and age effects for each individual ToM test. Following suggestions by Pinheiro and Bates (2000), model performance was compared with a likelihood ratio test. We had intercepts for participants as random effects in all LMM models (random subject factor) to account for the repeated measures structure of ToM. All other factors had fixed effects. All analyses, including likelihood ratio tests, were executed using R (R core Team, 2012) with lme4 package for LMM (Bates, M€ achler, Bolker, & Walker, 2015). Estimates of global model fit (R 2 ) were calculated using the MuMIn package by Barton (2013). Table 2 shows the results of the LMM analysis. Model M0 explained 22.2% of ToM variance and indicated the three covariates as highly significant ToM predictors. M1 (23.5% explained variance) revealed a significant Age parameter, indicating an age-related increase in ToM which is segregated from age-related improvements in language, inhibitory control, and reasoning. The likelihood ratio test showed that the 1.3% increase in model fit by adding the Age term was significant, v 2 (1) = 8.94, p < .005. A significant Age 9 ToM term in M2 revealed different Age effects for Eye Test and Faux Pas Test. The 1.1% increase in model fit for M2 (R 2 = .246) relative to M1 (R 2 = .235) was significant, v 2 (2) = 8.07, p < .02. The Age 9 ToM interaction of M2 is illustrated in Figure 2B. Either test survived the correction for repeated testing ( Sid ak-corrected error rate a SID = .0253 for a 5% overall alpha level).
The different age effects for Eye Test and Faux Pas Test are illustrated by the raw (unstandardized, b) slope estimates, which were b Age = .07 (Eye Test) and b Age = .02 (Faux Pas Test), respectively. The estimated slope of b = .07 corresponds to an about one standard (z) unit increase in the Eye Test performance across the studied age span, while the increase in the Faux Pas Test would be negligible (see also Figure 2). These results indicate a ToM-specific, age-related increase free of covariate effects only for the Eye Test, but not for the Faux Pas Test. Note. The family-wise error rate was a SID = .017 for a 5% overall alpha level ( Sid ak-correction).
ToM × Age ToM × Age

Discussion
The key findings of the present study are twofold. Our results reveal that the socialperceptual and social-cognitive ToM components dissociate in their development across adolescence and young adulthood and that they involve general cognitive functions to a different degree. We found an age-related increase in the social-perceptual ToM component (Eye Test) apart of improvements in language, inhibitory control, and reasoning. The age-related increase in the social-cognitive ToM component (Faux Pas Test), however, fully aligned with the development in these cognitive functions. A specific age-related improvement is thus evident only for the social-perceptual ToM component. This conclusion is, however, constrained by the possibility that not all relevant covariates have been identified for the latter ToM component. While our selection of covariates aimed at separating ToM components from most relevant dimensions of general cognitive development, it is possible that age-related growth in the social-perceptual component does not segregate from more specific covariates, for example, emotion recognition ability (Tousignant, et al., 2017). A test in the context of specific social cognition covariates is, at the time, outstanding. Recent longitudinal and cross-sectional studies show that ToM develops steadily across early and middle childhood (Peterson & Wellman, 2018;Wellman, Fang & Peterson, 2011). However, little is known about ToM changes across adolescence and during transition to adulthood. We observed a steady increase of average performance across adolescence for the social-perceptual component, while performance levels were volatile for the social-cognitive component, stabilizing not before young adulthood (see Figure 1). Discontinuous development even with retrograde local periods is known from other domains where availability of social norms is crucial (Luengo Kanacri, Pastorelli, Eisenberg, Zuffian o, & Caprara, 2013). However, the volatile performance in the socialcognitive component should be received with care, since this could likewise be an artefact of the specific samples of participants in mid and late adolescence (see also Limitations section below).
Advanced ToM tasks tap into distinct ToM components that may develop at different rates (Osterhaus et al., 2016;Tager-Flusberg & Sullivan, 2000). Tager-Flusberg and Sullivan (2000) suggested a distinction between social-perceptual and social-cognitive components in their two-component theory (see also Apperly & Butterfill, 2009;Meinhardt-Injac et al., 2018). The two components could be separated based on differential impairment in atypical populations (Williams Syndrome: Tager-Flusberg & Sullivan, 2000; Autism: Note. The table shows standardized (b) coefficients with their standard errors, tstatistic with significance level, multiple correlation coefficient, and determination coefficient. Schuwerk, Vuori, & Sodian, 2015). Results of the present study provide support for this model, as the data revealed (1) lacking of correlation between ToM components after controlling for age and (2) protracted development in the social-perceptual (i.e., Eye Test), but not in the social-cognitive component (Faux Pas Test). Similar results were reported by Devine and Hughes (2013) for children and adolescents (8-13 years of age), who showed that non-verbal Silent Films task performance, but not the Strange Stories task performance, was significantly influenced by age. In non-verbal material, the detection of the social cues is as relevant as their evaluation and interpretation (see also Geiger et al., 2019). However, the distinction between social-perceptual and socialcognitive components of ToM based only on task requirements is not always straightforward (see also Osterhaus et al., 2016;Tager-Flusberg & Sullivan, 2000). There are claims that the social-cognitive component of ToM has tighter connections to language and other aspects of cognition than the social-perceptual component (Tager-Flusberg & Sullivan, 2000). In line with this, we found that development of the socialcognitive component aligns with cognitive development, while only the social-perceptual component showed specific age-related increase beyond that in language, inhibitory control, and reasoning. This does not imply that these cognitive functions are not involved in the social-perceptual component (Meinhardt-Injac et al., 2018;Tager-Flusberg & Sullivan, 2000). Yet, it demonstrates that cognitive factors alone do not fully explain agerelated changes in this ToM component. Protracted development in social-perceptual ToM component may reflect increasing sensitivity to nuanced changes in emotional facial expressions across adolescence, that is, fine-tuning of perceptual discrimination (Rodger, Vizioli, Ouyang, & Caldara, 2015;Thomas, De Bellis, Graham, & LaBar, 2007). Such a finetuning is driven by new social developmental tasks and increasing experience with more complex emotional and mental states (Garcia & Scherf, 2015;Motta-Mena & Scherf, 2016). This hypothesis is further supported by evidence for protracted development in face perception and recognition (Fuhrmann et al., 2016), and by evidence for connections between perceptual processes and visual ToM tasks, including the Eye Test (Meinhardt-Injac et al., 2018).
Note that the distinction between social-perceptual and social-cognitive component crucially depends on the source of information (non-verbal vs. verbal). Another important differentiation proposed in ToM research concerns affective and cognitive ToM components (Sebastian et al., 2011;Shamay-Tsoory & Aharon-Peretz, 2007). Cognitive ToM (i.e., also termed 'cold' ToM) refers to the ability to represent one's own and others' thoughts, intentions, and desires. Affective ToM (also termed as 'hot' ToM) is concerned with the understanding of the own and others' emotional states and preferences. In this view, it does not matter whether affective and metal states are judged from non-verbal or verbal cues, or a combination of the two (see Yoni Task, Shamay-Tsoory & Aharon-Peretz, 2007), only content matters (e.g., thinking about something vs. liking something). Imaging studies have found large overlapping, but also differences in the neural networks involved in cognitive as opposed to affective ToM (Shamay-Tsoory & Aharon-Peretz, 2007). The activation of the neuronal structures, however, was not only modulated by content, but depended on the stimulus material and the source of information used in the test material (Schlaffke et al., 2015). This suggests that the two models might be complementary rather than competing. Possibly, information about affective states could stronger rely on non-verbal cues, compared to information about cognitive states and intentions. This remains to be clarified in future studies.
In the last years, there is increasing interest in bi-directional associations between ToM performance and real-world social behaviours. We know that social context shapes ToM development (Brizio et al., 2015), but also that social behaviour is predicted by ToM ability (Banerjee, et al., 2011). In school-aged children, advanced ToM ability has been associated with social competence (i.e., more cooperation and less conflict), prosocial behaviour, and peer popularity (see Slaughter et al., 2015, for a meta-analysis). Although social and environmental changes during adolescence and young adulthood are tremendous, at the time there is still little evidence on bi-directional associations with ToM development (Tamnes et al., 2018). Since ToM ability is crucial for coping with social and personal challenges that first arise in adolescence, studying its underpinnings and characteristics is an important endeavour, which deserves attention in future research.

Limitations
The current study has two limitations that should be acknowledged. We included a large sample of participants with a continuous age distribution in our study; nevertheless, it is important to keep in mind that cross-sectional age differences are only an approximation of true developmental changes (Molenaar, Huizenga, & Nesselroade, 2003). With the applied cross-sectional research design, the present study suffers from constraints in establishing valid (causal) temporal associations between learning experiences at a given age and the increase in ToM ability. Further, the test we used to measure inhibitory control consists of non-verbal social stimuli (faces) that may underestimate the role of inhibitory control for verbally presented ToM tests. Nonetheless, separating ToM from general cognitive development, the data provide a first approximation of how social-perceptual and socialcognitive ToM change across adolescence into young adulthood.

Data availability statement
The data that support the findings of this study are available from the corresponding author upon reasonable request. Zerwas, S., Balaraman, G., & Brownell, C. (2004 Supporting Information The following supporting information may be found in the online edition of the article: Figure S1. Regression-scatterplots for all five tests (z-scores). Figure S2. Distribution of standardized residuals (left) and plot of standardized residuals against predicted values (right) for the LMM models M0, M1 and M2. Table S1. Descriptive statistics of the sample.