Dynamic face processing impairments are associated with cognitive and positive psychotic symptoms across psychiatric disorders

Impairments in social cognition—including recognition of facial expressions—are increasingly recognised as a core deficit in schizophrenia. It remains unclear whether other aspects of face processing (such as identity recognition) are also impaired, and whether such deficits can be attributed to more general cognitive difficulties. Moreover, while the majority of past studies have used picture-based tasks to assess face recognition, literature suggests that video-based tasks elicit different neural activations and have greater ecological validity. This study aimed to characterise face processing using video-based stimuli in psychiatric inpatients with and without psychosis. Symptom correlates of face processing impairments were also examined. Eighty-six psychiatric inpatients and twenty healthy controls completed a series of tasks using video-based stimuli. These included two emotion recognition tasks, two non-emotional facial identity recognition tasks, and a non-face control task. Symptoms were assessed using the Positive and Negative Syndrome Scale. Schizophrenia and bipolar disorder groups were significantly impaired on the emotion-processing tasks and the non-face task compared to healthy controls and patients without psychosis. Patients with other forms of psychosis performed intermediately. Groups did not differ in non-emotional face processing. Positive symptoms of psychosis correlated directly with both emotion-processing performance and non-face discrimination across patients. We found that identity processing performance was inversely associated with cognition-related symptoms only. Findings suggest that deficits in emotion-processing reflect symptom pathology independent of diagnosis. Emotion-processing deficits in schizophrenia may be better accounted for by task-relevant factors—such as attention—that are not specific to emotion processing.


INTRODUCTION
Social cognition is increasingly recognised as a core deficit in schizophrenia 1 . One component of social cognition is the ability to extract emotional cues from faces. Impaired emotion recognition is associated with poorer social and occupational functioning in schizophrenia, and may also mediate the relationship between social functioning and broader neurocognitive deficits [2][3][4] . Furthermore, deficits in emotion recognition appear to precede the onset of psychosis, and may predict the conversion to schizophrenia in clinically high-risk populations 5 .
The exact mechanics that underlie impairments in emotionprocessing are an ongoing source of debate. There is consensus, however, that individuals with schizophrenia are significantly impaired in their ability to recognise emotions compared to healthy controls and other psychiatric disorders 6,7 . Meta-analyses estimate these effect sizes to be quite large, e.g., d = −0.91, d = −0.85, and g = 0.89 respectively 6,8,9 . It remains unclear whether these reported emotion-processing deficits are indeed specific to facial expressions, or whether they could be due to impairments in processing structural aspects of faces in general, or processing visual stimuli more generally. Notably, the ability to recognise the identity of a face may also be impaired in schizophrenia, although previous studies have produced varying results 10,11 . As facial emotion-processing and identity-processing are believed to be underlain by largely separate neural routes 12,13 , the comparison of these abilities allows us to better characterise these impairments and their relevant correlates.
The N170 is an event-related potential thought to relate to the structural encoding of faces and is increased when viewing faces compared to other complex objects. In patients with schizophrenia, the N170 is attenuated for both faces and other complex objects, although findings vary across studies (see paper by Salisbury et al., 2019 14 ). Overall, this suggests a generalised visual processing impairment that is not specific to faces, but may contribute to impairments on face-processing tasks.
Research in healthy populations suggests that using dynamic stimuli-rather than static images-to investigate emotionprocessing confers a range of advantages, including increased ecological validity and greater accuracy 15 . Despite this, only a handful of studies have employed dynamic face stimuli to investigate emotion-processing in schizophrenia. Deficits in recognising dynamic face-specific emotions have been reported by several studies of schizophrenia patients [16][17][18][19] . Notably, static and dynamic tasks may tap into different patterns of difficulties. In one study, performance on dynamic tasks correlated with greater positive symptoms, while performance on static tasks correlated with negative symptoms 17 . In another study, patients' performance on a dynamic emotion-recognition task correlated with IQ and other cognitive measures, while performance in healthy controls correlated with a face memory task and a social cognition task only. These results concur with research using static stimuli (e.g., Bediou et al., 2007 20 ) and lend weight to the argument that emotion-processing deficits in schizophrenia may be accounted for by cognitive deficits.
A further question is whether the face processing deficits observed in schizophrenia are shared by other psychiatric disorders. There is some evidence for impaired face processing in bipolar disorder, particularly for recognising emotion, although the degree and persistence of these deficits remain contentious 21 . Patients with bipolar disorder have shown impaired performance compared to healthy controls in correctly labelling facial emotions [22][23][24] as well as discriminating between different expressions [25][26][27] . Other studies have shown that patients with bipolar disorder require greater intensity to recognise emotions compared to controls 28,29 . In contrast, other studies report intact emotion recognition in bipolar disorder [30][31][32] or in certain patient subsets, such as euthymic patients 22 .
Studies indicate some degree of impairment in recognising facial emotion in major depressive disorder (MDD), although the extent of these deficits remains unclear. Several reviews suggest that, on balance, there are mild but significant impairments in emotion-processing in MDD [33][34][35] . In particular, MDD patients tend to show a bias towards perceiving neutral or ambiguous expressions as sadness, and show a reduced ability to recognise all basic emotions except for sadness 36 . In contrast, a number of experimental studies have reported no difference in emotionprocessing ability between patients with MDD and healthy controls 29,30,[37][38][39][40][41][42] . This discrepancy is possibly due to a lack of power in these studies, as a recent meta-analysis suggests that the overall effect size for these deficits in MDD is quite small (Hedges g = −0.16) 36 .
With respect to the anxiety disorders, one meta-analysis examined emotion-processing impairments across 40 studies 43 . A large weighted mean effect size was found for post-traumatic stress disorder (Cohen's d = −1.60), indicating substantial deficits in emotion recognition. However, only small or negligible effects were found for social phobia (d = 0.12), obsessive-compulsive disorder (d = −0.16), panic disorder (d = −0.25) and generalised anxiety disorder (d = −0.12). Whether these deficits indicate specific impairments in recognising emotion, or simply reflect more general difficulties with attentional control that typically accompany anxiety disorders 44 , remains to be established.
Face processing deficits may accompany specific symptoms which are found across disorders. For instance, emotionprocessing deficits-albeit using static stimuli-have been predominantly associated with negative symptoms in schizophrenia, and to a lesser degree positive symptoms, later age of illness onset and inpatient status 6,45,46 . In contrast, the relationship between identity recognition impairments and symptoms has not been widely studied, and evidence is mixed. Two studies have reported a negative correlation between identity recognition performance in schizophrenia and both positive and negative symptoms 47,48 while several report correlations with negative symptoms only 26,45,[49][50][51] . To date, no studies have assessed symptom correlates with identity processing using dynamic stimuli.
The current study aimed to address these issues by assessing participants with a range of psychiatric disorders (n = 86) and healthy controls (n = 20) using four different emotion and identity processing tasks (see Fig. 1). To ascertain whether these deficits generalise beyond face processing, performance was also compared on an equivalent task using non-face stimuli. Novel dynamic video-based stimuli were developed for use in all tasks (see Darke et al., 2019 52 for further details). The aims were (a) to explore whether impairments in emotion-processing can be explained by more general deficits in non-emotional face processing or non-face processing, (b) to assess whether face processing impairments are shared by other psychiatric disorders, and (c) to investigate symptom correlates of face processing ability across psychiatric disorders.

Demographics
Demographics and symptom ratings are shown in Table 1. Pearson's chi-square test revealed that the sex makeup of groups did not differ significantly as a function of diagnosis, X 2 (4) = 6.18, p = 0.186. One-way ANOVAs were performed with Group as a between-subjects factor and Age, Years of Education, and estimated FSIQ as within-subjects factors. A significant effect was found for Years of Education, F(4,101) = 2.50, p = 0.048, ŋ p 2 = 0.09. Post-hoc t tests (Bonferroni corrected) revealed a significant difference of 2.11 years between the Control and Schizophrenia group (p = 0.02). FSIQ estimates were also found to differ between groups, F(4,93) = 2.52, p = 0.047, ŋ p 2 = 0.10. Post-hoc t tests (Bonferroni corrected) revealed a significant difference of 8.38 points between the Control and Schizophrenia spectrum groups (p = 0.03). Age did not differ significantly between groups,

PANSS subscales
One-way ANOVAs were run with group as IV (excluding healthy controls), and Positive, Negative, and General Psychopathology scores as DVs. A significant main effect was found for Positive Symptoms, F(3,82) = 18.76, p < 0.001, ŋ p 2 = 0.41. Bonferroni corrected post hoc tests revealed, not surprisingly, that the Nonpsychosis group had significantly lower Positive symptom scores than all other groups (p = 0.002 to <0.001). The Other group trended towards having significantly lower Positive symptom scores compared with the bipolar group (p = 0.055). No other group differences approached significance.

Task performance in healthy controls
A repeated-measures ANOVA was conducted to determine whether difficulty varied across the five dynamic tasks in healthy controls. Briefly, accuracy for the Identity Discrimination task was significantly higher compared to the Sex Labelling and Emotion Discrimination tasks, however performance across all other tasks was of a comparable level. Within-group comparisons for the four patient groups can be found in the supplemental appendix. A main effect of task was found for each of the five within-group analyses (p < 0.001).

Group differences
Repeated-measures ANOVAs were conducted on raw accuracy data across morphing levels for each of the five groups (see supplementary appendix for analyses). As all groups showed the same pattern of performance, data for each task was collapsed across morphing levels for subsequent analyses.
No effect of group was found for Sex Labelling, (F(4,99) = 0.363, p = 0.84, ŋ p 2 = 0.01). An effect of morphing level was seen, with lower accuracy for 40% male/60% female faces (See Participants were asked to report whether the feature of interest was the same or different across the two videos (a, c, and e) or to label the emotion illustrated as disgust/fear (b) or the sex was male/female (d). Morphing of the video files was used to create different levels of task difficulty. Further details regarding the stimuli set and morphing procedure are provided in supplemental Figures 1-3 and the methods paper by Darke and colleagues 52 .
Supplementary Fig. 9E), however overall performance was consistent across all groups (see Fig. 2d).
Unexpectedly, a significant univariate effect of group was also found for Car Discrimination (

Response bias-c
Mean values of c ranged from 0.02 to 0.88 across groups and tasks. Multivariate ANOVA revealed no significant main effect of group, F(4,99) = 0.84, p = 0.67, ŋ p 2 = 0.04. Pairwise comparisons showed no differences between groups on any task, suggesting that response bias did not differ between groups, and is therefore unlikely to account for group differences in task performance.
Age, education and FSIQ Pearson correlations were conducted to determine if age, years of education, or FSIQ estimates predicted performance on any of the tasks. It was found that increasing age correlated with worsening d' for Emotion Labelling (r = −0.197, p = 0.04), but not other tasks. FSIQ estimates produced no significant correlations. Years of education correlated positively with performance on Emotion Discrimination (r = 0.260, p = 0.007) only. To determine if years of education could account for group differences on the Emotion Discrimination task, analyses were re-run excluding participants with fewer than 11 years of education (eliminating 13 patients in the schizophrenia group, 5 Bipolar disorder, 6 Other and 6 Nonpsychosis). One-way ANOVA revealed a significant mean effect of group F(4,70) = 10.64, p < 0.001, ŋ p 2 = 0.38. Post-hoc tests showed that the schizophrenia group performed significantly lower than the control (p < 0.001), Non-psychosis (p = 0.001) and Other groups (p = 0.02). The control group also outperformed the Bipolar group (p = 0.01). Thus, limited years of education was unlikely to account for the group differences shown on these tasks.

Illness duration and medication
Pearson correlations revealed that illness duration correlated negatively with task performance on Emotion Discrimination (r = −0.337, p = 0.002), Emotion Labelling (r = −0.224, p = 0.04) and Car Discrimination (r = −0.298, p = 0.005), and trended negatively with Identity Discrimination (r = −0.204, p = 0.06). Hierarchical regression analyses revealed that, after controlling for Age, Illness Duration continued to significantly predict performance on Emotion Discrimination (change in R 2 = 0.09, F(1,83) = 8.37, p = 0.005) and Car Discrimination (change in R 2 = 0.13, F(1,83) = 12.27, p = 0.001), but no longer predicted performance on Emotion Labelling (change in R 2 = 0.01, F(1,83) = 1.27, p = 0.26). This suggests that patients with a longer illness duration performed more poorly on all tasks except for Sex Labelling, regardless of group, however this cannot account for the group differences observed.
In recognition of the potential sedating effects of benzodiazepines 53 and antipsychotics 54 , correlations between medications and performance were examined. Benzodiazepine daily dose (n = 16) correlated negatively with performance on Emotion Labelling (r = −0.519, p = 0.04) and Car Discrimination (r = −0.535, p = 0.033) and trended towards significance for Identity Discrimination (r = −0.458, p = 0.07). Mean antipsychotic daily dose (n = 84) produced no significant correlations.
To determine if benzodiazepine use could account for the group differences in task performance, a MANOVA was run excluding the 17 inpatients who had taken benzodiazepines. As before, a significant main effect was found for group, F(4,83) = 2.50, p < 0.001, ŋ p 2 = 0.13 (Pillai's Trace) and univariate tests still revealed significant, or trending towards significant effects of group for four of the five tasks: Identity Discrimination (F(4,83) = 2.07, p = 0.09, ŋ p 2 = 0.09), Emotion Discrimination (F(4,83) = 11.22, p < 0.001, ŋ p 2 = 0.35), Emotion Labelling (F(4,83) = 7.01, p < 0.001, ŋ p 2 = 0.25), and Car Discrimination (F(4,83) = 3.96, p = 0.005, ŋ p 2 = 0.16). This suggests that although benzodiazepine use was related to poorer performance on some tasks, it cannot account for the group differences reported in this study.  Table 2 shows correlations between PANSS subscales and task performance. Positive Symptoms were negatively correlated with all tasks except Sex Labelling (rs = −0.31 to 0.41). General Psychopathology scores correlated negatively with Car Discrimination only (r = 0.37). No correlations with Negative Symptoms approached significance for any of the tasks. Spearman-rank correlations between individual items on the PANSS and task performance are shown in Table 3. When considered overall, it appears that classically positive symptoms -such as Delusions, Grandiosity, Suspiciousness, and Unusual Thought Content-correlated negatively with performance on the two Emotion tasks, and with Car Discrimination. In contrast, no positive symptoms correlated with the two Identity tasks (with the exception of Unusual Thought Content, which was positively correlated with Sex Labelling).

Correlations with clinical symptoms
The other trend to note is that performance on all tasks (except Sex Labelling) tended to correlate negatively with cognitive symptoms, such as Conceptual Disorganisation, Difficulty in Abstract Thinking, Stereotyped Thinking, Poor Attention, Disorientation, and Lack of Judgement and Insight. These items also appear to drive the association reported between General Psychopathology subscale scores and Car Discrimination performance. Interestingly, almost no correlations were found with affective symptoms such as Depression and Blunted Affect. The only exception was that higher Depression and Guilt Feelings was associated with better performance on Emotion Discrimination, but not other tasks.

DISCUSSION
The aim of the current study was to determine the diagnostic specificity and symptom correlates of impairments in facial emotion and identity processing using dynamic tasks. Results revealed that groups with bipolar disorder, schizophrenia, and 'other' (non-schizophrenia) psychotic disorders were significantly impaired on the emotion tasks compared to healthy controls, while patients with non-psychotic disorders were unimpaired. In contrast, all patient groups showed relatively intact performance on the two identity processing tasks. Unexpectedly, patients with schizophrenia and bipolar disorder also showed deficits on a non-face comparison task. Analysis of symptom correlates suggested that tasks of facial emotion, identity, and non-face discrimination were associated with different patterns of symptoms. These results revealed significant emotion-processing deficits in patients with bipolar I disorder which were comparable to patients with schizophrenia. These findings are consistent with previous studies of patients with bipolar disorder using static stimuli, which typically report emotion-processing deficits regardless of task design [22][23][24][25][26][27] . Unexpectedly, the current study showed no significant difference in performance between the bipolar and schizophrenia groups. This is contrary to previous research 26,31,34,[55][56][57] although several other studies have also produced a null result 30,41,58 . This finding may reflect a lack of power due to the small number (n = 16) of bipolar patients in this sample. Alternatively, this sample may have been more impaired due to the exclusion of bipolar II disorder (nonpsychotic) patients from this group, a subtype that in some cases has been shown to be less impaired in emotionprocessing compared to bipolar I patients 22,59 .
This study demonstrates an impairment in dynamic emotionprocessing in non-schizophrenia psychoses. This finding supports the idea that non-schizophrenia forms of psychosis produce similar deficits to those seen in schizophrenia, and likely involve similar neural mechanisms 60 . Moreover, the finding that emotion recognition was unimpaired in patients with non-psychotic disorders is consistent with studies reporting broadly intact performance in major depression 29,39,41,42 , certain anxiety disorders 43 , and borderline personality disorder 61 . Note that, although some meta-analyses suggest that these disorders are associated with mild deficits in emotion recognition (particularly misinterpretation of threatening faces) these are typically of a much smaller effect size than those seen in schizophrenia, and it is  H. Darke et al. possible that the current study lacked the power to detect subtler deficits 36,43 . Future work should aim to replicate these results in larger studies to ensure the findings are robust and representative of the broader population with these disorders. Our results suggest that, irrespective of diagnosis, patients with positive symptoms such as delusions, suspiciousness, and unusual beliefs are most likely to show emotion-processing difficulties compared to those with other symptoms. This result is at odds with the majority of past research using static stimuli, which predominantly report associations with negative symptoms 30,51,[62][63][64][65][66] However, this finding is consistent with that of Johnston and colleagues (2010) who found that positive symptoms correlated with dynamic emotion-processing, while negative symptoms correlated with static emotion-processing. Given that dynamic faces elicit different patterns of brain activation compared to static faces 67 , it is possible that positive and negative symptoms may have differential effects on these brain networks. It is worth noting, however, that subsequent studies using dynamic emotion tasks have not replicated this dissociation between positive and negative symptoms 16,18 .
Interestingly, the current study found that performance on two dynamic identity-processing tasks was not associated with positive symptoms or negative symptoms. The use of our dynamic stimuli that were designed specifically to examine identity processing in patients, provides important new evidence about the specificity of perceptual alternations seen across this psychiatric population.
The current study revealed no significant impairments in identity processing in any of the groups examined. This finding is consistent with previous studies in bipolar disorder 23,27,68 , major depression 37 , and some studies of schizophrenia 20,37,69 . Unexpectedly, it was found that the schizophrenia and bipolar groups were both significantly impaired on a non-face discrimination task compared to healthy controls. It is possible that the schizophrenia group showed preserved identity processing because the processing of non-emotional face information is mediated by rapid, largely automatic perceptual processes 70 . In contrast, discrimination of less familiar stimuli, such as cars, may require a greater level of cognitive effort or attentional control, and may therefore be more sensitive to subtler cognitive impairment.
The observation that cognitive-related symptoms correlated negatively with task performance suggests that patients with more generalised cognitive difficulties tended to be less accurate overall, regardless of stimulus type. This finding is consistent with previous studies indicating associations between cognitive factors and emotion-processing in schizophrenia 64,71,72 . This finding is not unexpected, particularly given the predominance of attentional difficulties in psychiatric disorders such as schizophrenia 73 . However, it does highlight the pervasive impact of generalised cognitive deficits, even on tasks that are intended to tap into specialised areas of perceptual deficit, such as emotion-processing. The current results align with the view that emotion-processing deficits in schizophrenia may be accounted for by more general task-relevant factors, such as attentional disturbance 74 . It is important to note, however, that our results do not rule out the presence of an overlapping emotion-specific deficit. Future studies would benefit from including standardised tests of cognition, particularly attention and visual working memory, to better control for these factors and clarify the relationship between aspects of cognitive impairment and emotion-processing ability.
In conclusion, using dynamic stimuli this study found that patients with schizophrenia and bipolar disorder showed similar deficits in emotion-processing, and non-face discrimination compared to healthy controls and patients with non-psychotic illnesses. Examination of symptoms across disorders indicated that positive symptoms of psychosis correlated with both emotion-processing performance and non-face discrimination across patients. Uniquely, we found that identity processing performance was associated with cognitive-related symptoms only. Findings align with the views that emotion-processing deficits may be accounted for by more general task-relevant factors, such as attentional disturbance seen in psychotic disorders, regardless of diagnostic category.

METHOD Participants
Eighty-six inpatients were recruited from an acute psychiatry unit in Melbourne, Australia. All participants were inpatients at the time of participation. Final diagnoses (DSM-IV criteria) were obtained from discharge reports provided by the treating psychiatrist and verified by the research psychiatrist (Sundram). Patients were categorised into four groups: 36 schizophrenia-spectrum (including schizophrenia, schizoaffective disorder, and first episode psychosis), 15 bipolar disorder (bipolar-affective disorder with a history of psychotic symptoms), 17 "other" psychotic disorders (including drug induced-psychosis, major depression with psychosis, borderline personality disorder [PD] with hallucinations, and schizotypal PD), and 18 non-psychotic disorders (including bipolar II disorder, major depression, generalised anxiety disorder, borderline PD, and situational crisisnone of whom had ever experienced symptoms of psychosis).
Twenty non-clinical controls were recruited via online advertising. All were free from neurological injury, psychiatric illness or substance use disorder by self-report, and were not taking psychoactive medication. All participants received monetary compensation for their time and gave written informed consent. Inpatient participant consent was signed in the presence of an impartial witness. This study was approved by Melbourne Health and University of Melbourne Human Research Ethics Committees.

Demographics
Participants completed the National Adult Reading Test (NART) 75 , and a demographics questionnaire. Patients additionally reported current medication, illness duration, and were assessed using the Positive and Negative Syndrome Scale (PANSS) 76 . PANSS interviews were conducted by a graduate student with extensive experience and all ratings were reviewed by an experienced PANSS rater and psychiatrist (Sundram).

Emotion discrimination
Detailed explanation of task development can be found in Darke and colleagues 52 and are available for download from http://go.unimelb.edu. au/e3t6. Stimuli used were 2000 ms videos of faces changing from neutral expressions to either disgust or fear, adapted from the MMI-Facial Expression Database 77,78 and the Facial Expressions and Emotion Database (FEED) 79 . Fear and disgust were chosen because these expressions are not easily confused with one another in healthy controls (unlike anger and disgust, or fear and surprise 80 ) and, unlike positive expressions, are more likely to elicit emotion-recognition impairments in clinical populations, such as in schizophrenia 81 and bipolar disorder 21 .
Faces were edited to remove non-face cues (e.g., hair or glasses) and were presented centrally in greyscale against a black background. Stimuli were 5x4cm viewed at a distance of approximately 50 cm (5.7×4.6°of visual angle). To vary the intensity of emotion, peak expression frames from each video were "morphed" together with neutral frames to create new stimuli using Fantamorph 5 82 . Original videos consisted of six unique individuals (3 male, 3 female) each showing one expression of disgust and one of fear. These 12 videos were morphed to create five levels of expression intensity (33%, 50%, 67%, 83%, and 100%), totalling 60 stimuli (examples shown in Supplementary Figs. 1-3). In each trial one expression was shown, then followed by a second face of a different individual showing either the same or different expression (Fig. 1a). Pairs of expressions were always shown at the same intensity level. Participants were instructed to state aloud whether each pair of faces showed the "same" or "different" emotion.

Emotion labelling
Stimuli used were the same as those in Emotion Discrimination. Each expression was shown for 2000 ms. Half of trials were "disgust" and half were "fear". Participants were instructed to state aloud whether each face more closely resembled "fear" or "disgust".
H. Darke et al.

Identity discrimination
Stimuli used were videos of faces showing non-emotive facial movements, such as opening the mouth or raising the eyebrows. Animations were created using the same methods described for Emotion Discrimination, except that video of different individuals (of the same sex) were morphed together to vary the degree of similarity between faces. Six pairs of unique individuals (3 male, 3 female) were used. Each pair was morphed to create six new animations ranging from one identity to the other at 20% increments, totalling 36 stimuli.

Sex labelling
Stimuli used were identical to Identity Discrimination above, with the exception that each identity was morphed with an opposite-sex identity instead of a same-sex identity. Six sets of 6 face animations were created, ranging from male to female. Half of the trials were "male" (i.e.,: 60%, 80%, or 100% male) and half were "female" (0%, 20%, and 40% male). Participants were instructed to state aloud whether each face more closely resembled "male" or "female".

Car discrimination
Stimuli used were 1000 ms videos of 3D car models rotating from a side view to a 45°view. 3D meshes were obtained online via a free 3D modelling website 83 then edited and animated using 3Ds Max Design 84 . Twelve unique cars were animated and paired with similar looking models. For each trial, one car video was shown for 1000 ms, followed by a 500 ms blank screen, then a second car video. Participants verbally responded whether each pair of stimuli were "same" or "different".

General procedure
Participants completed the five computerised tasks in one of four counterbalanced orders. Prior to each task, participants completed practice trials with feedback. For the three Discrimination tasks, participants were instructed to say whether each pair of stimuli (either faces, cars or emotions) were the same or different (the ratio of same/ different trials was 50:50). Identity Discrimination and Car Discrimination consisted of 120 trials each (see Fig. 1a, c, and e). As Emotion Discrimination required longer presentation times, this task was reduced to 100 trials to limit participant fatigue.
For the two Labelling tasks, participants were instructed to state whether each face more closely resembled "fear" or "disgust" (Emotion), and "male" or "female" (Sex), respectively ( Fig. 1b and d). Sex Recognition consisted of 72 trials. After piloting, Emotion Recognition was reduced to 60 trials to reduce testing time. The experimenter logged all verbal responses using a keyboard to reduce any impact of impulsive or impaired motor response mapping.
Testing took~2 h to complete, and participants were permitted as many breaks as desired. Computerised tasks were completed on a laptop computer (60 Hz, 16 inch screen size) at a comfortable viewing distance in a quiet distraction-free environment.
Results were analysed using the software package SPSS version 20. To limit response bias, percentage correct was converted to d' scores using formulae recommended by MacMillan and Creelman 85 . For Sex Labelling and Emotion Recognition this was calculated as: d' = z(Hit rate)z(False alarms). For the three Discrimination tasks this value was then converted to a modified d' 85 . To avoid dividing by zero, Hit Rate and False Alarms were adjusted according to Corwin 86 . A measure of response bias, c, was also calculated using the formula: c = −0.5 [z(Hit rate)+z(False alarms)] 85 . Task performance was compared across the five groups using Repeated-Measures ANOVA and MANOVA (where the assumption of sphericity was violated). All post-hoc comparisons were Bonferroni-corrected to control for multiple comparisons. For Pearson and Spearman correlation, bootstrapping was used to calculate bias-corrected and accelerated (BCa) confidence intervals using 1000 resamples 87 .

Reporting summary
Further information on research design is available in the Nature Research Reporting Summary linked to this article.

DATA AVAILABILITY
The raw data that support the findings of this study are available from the corresponding author upon request.