New Gambling Task Reveals an Atypical Relation Between Social Anxiety and Social Avoidance in Schizophrenia

Abstract

A core symptom of schizophrenia is a deficit in social cognition. Furthermore, many patients suffer from relevant social anxiety. How social anxiety relates to social avoidance in schizophrenia is, however, still unclear. To address this question empirically, 30 participants with schizophrenia were compared to individually matched controls in a recently developed experimental task. In each trial, participants chose between a monetary gamble with a human partner providing social feedback and a defined monetary amount that varied across trials. The tendency to avoid the interaction under acceptance of monetary loss served as a measure of social avoidance. Patients showed higher levels of social anxiety assessed using the Liebowitz Social Anxiety Scale (LSAS) than controls, took longer to decide whether to engage in the social interaction and rated social feedback as less intense. Despite making rational decisions, patients surprisingly showed no social avoidance. We assume that the lack of experimentally induced social avoidance in contrast to the well-known social avoidance behavior in everyday life, also reflected by elevated self-rated LSAS scores, was due to a partial compensation of delayed social information processing in the absence of the time constraints in our task. This additional time may have allowed rational decisions and masked potential social avoidance behavior. We conclude that social anxiety in schizophrenia may have different characteristics than in individuals without schizophrenia, namely avoidance of negative feedback caused by prolonged social processing instead of fear of supposed negative feedback. Treatment strategies for social anxiety in schizophrenia may benefit from considering these differences.

Introduction

Schizophrenia is among the most disabling psychiatric disorders¹ and leads to substantial morbidity and mortality.^2,3 Besides psychotic symptoms, cognitive impairment is of high relevance, as it predicts patients’ functional outcome.⁴ One important domain of cognitive impairment is social cognition,⁵ of which several subdomains are affected in schizophrenia, eg, social perception, emotion processing, theory of mind, and emotion regulation.⁶ In the time course of the disease, social cognitive deficits are amongst the first to emerge and can already be found in individuals at high risk for psychosis.^7–9 It was suggested that social deficits mediate between cognitive deficits in the strict sense and functional outcome^10,11 and help to explain low recovery rates.¹² Due to their importance for full recovery, effective approaches to improve social cognitive skills are currently under active development.¹³ Both with respect to diagnosis and therapy, it is of crucial importance to improve the understanding of the cognitive processes underlying these social dysfunctions.

In the last decades, it has become apparent that social anxiety is a major problem in schizophrenia.¹⁴ Between approximately 15% and 30% of schizophrenia patients additionally fulfil criteria for a diagnosis of social anxiety disorder.^15–17 Despite this high prevalence, social anxiety in schizophrenia remains frequently neglected and undiagnosed.¹⁸ Addressing this issue is important because comorbid social anxiety is associated with a higher general disability,^19,20 higher lifetime rate of suicide attempts and of successful suicides, more frequent substance abuse, lower social adjustment and lower overall quality of life.^16,21 Furthermore, social anxiety has been identified as a relevant factor contributing to the conversion to schizophrenia in patients at risk.²² An established instrument to measure social anxiety symptoms is the Liebowitz Social Anxiety Scale (LSAS),^23,24 shown to be adequate and reliable to assess social anxiety symptoms in psychosis.^16,25

Here, we attempted, in a preliminary study, to relate real-life social anxiety symptoms measured using the LSAS to deficits in social cognitive functions in schizophrenia by quantifying social avoidance using a recently developed risky economic decision task.²⁶ In contrast to standard clinical measures such as interviews and questionnaires, experimental approaches help to quantify specific dimensions of observable behavior in a fine-grained manner as proposed under the Research-Domain-Criteria Initiative,²⁷ and are not influenced by the accuracy of self-reflection, by retrospective bias or the social desirability of certain responses.²⁸ Our approach allows to quantify an individual’s value assigned to the engagement in a simple social situation. This value is obtained by the indifference point between choosing to engage in the social situation as opposed to obtain a monetary amount. As social anxiety may be associated with a general risk avoidance tendency that is not specific to the social domain, our task uses a combination of conditions to control for risk avoidance. In a previous study, this task revealed a link between social avoidance and social anxiety in healthy individuals: the more socially anxious participants were, the more money they accepted losing in order to avoid engaging in a social interaction.²⁶ Given the association between schizophrenia and social anxiety,¹⁷ and between social anxiety and social avoidance,²⁶ we expected the value of social avoidance to be higher in patients than in controls.

Materials and Methods

Thirty individuals satisfying DSM-V criteria for schizophrenia (DSM-V 295.90) as determined by medical records and consultation with treating psychiatrists (age [mean ± SD] 33.6 ± 10.3 y, 8 females) and 30 control subjects (age 34.1 ± 10.9 y, 8 females) individually matched for age (±2 y) and sex, were recruited to take part in this study. One patient did not terminate the task and dropped out of the study. We determined the number of participants using a power analysis, assuming an effect size of 0.65, which corresponds to a cost of social avoidance of 0.008€/LSAS point observed in the study first reporting this task²⁶ and an estimated difference of about 20 LSAS points between patients and controls (based on studies using LSAS in schizophrenia^15–17), an alpha error of P = .05 and a power of 80%. Participants’ age ranged from 19 to 60 years. Patients were in- and outpatients recruited from the Department of Psychiatry at the University Hospital of Cologne. Healthy volunteers were recruited among the general population. Details of the sample are provided in table 1.

Table 1.

Demographic Details and Assessment Instrument Scores of Patients (n = 30) and Control Subjects (n = 30)

Measure	Patients	Controls
Age (y)	33.6 ± 10.3	34.1 ± 10.9
Sex (N females; N males)	8; 22	8; 22
LSAS overall score	51.27 ± 30.71	14.63 ± 12.07
LSAS fear subscale	26.00 ± 16.24	7.6 ± 5.82
LSAS avoidance subscale	25.66 ± 15.51	7.03 ± 7.01
Disease duration (mo)	106.9 ± 90.9	N.A.
Medication (mg Chlorpromazine/day equivalent)	678.8 ± 416.1	N.A.
PANSS total score	50.6 ± 11.8	N.A.
PANSS positive symptoms subscale	11.73 ± 3.62	N.A.
PANSS negative symptoms subscale	14.13 ± 5.87	N.A.
PANSS general symptoms subscale	24.76 ± 5.55	N.A.
LSAS avoidance subscale	51.27. ± 30.71	N.A.
CDSS	0.07 ± 0.25	N.A.
CGI	4.07 ± 0.98	N.A.
SOFAS	56.53 ± 13.81	N.A.

Note: CDSS, Calgary Depression Rating Scale for Schizophrenia; CGI, Clinical Global Impression-scale; LSAS, Liebowitz Social Anxiety Scale; PANSS, Positive and Negative Syndrome Scale for Schizophrenia; SOFAS, Social and Occupational Functioning Assessment Scale; N.A.: not applicable. Mean ± SD are shown except if otherwise stated.

The duration of illness was between 0 (first episode psychosis) and 25 years. Mean duration of disease was 8.9 ± 7.6 (median: 6.5) years, but this duration was widespread: Fifteen patients ranged between 0 and 6 years, 8 from 6 to 16 years, and 7 from 16 to 25 years. Patients were excluded if presenting evidence of any comorbid Axis-I-diagnosis, acute suicidality measured by Item 8 of the Calgary Depression Rating Scale for Schizophrenia (CDSS),²⁹ significant medical illness including past or present neurological disorder or acute intoxication. While we did not assess IQ explicitly, patients were recruited by experienced psychiatrists who took care to avoid, including patients with clinical impression of mild or higher degrees of mental retardation (ICD-10: F70-F73). None of the participants was hospitalized or treated coercively by order of the responsible local authorities. Healthy volunteers denied having any past or current psychiatric diagnoses. All participants had normal or corrected-to-normal vision and a sufficient proficiency of the German language to understand the aims of the study, the task instructions, and to provide written informed consent. Ethical approval was provided by the University of Cologne Ethics Committee (Ethics approval number 18–265). Patients’ symptom severity and classification were assessed using the Positive and Negative Syndrome Scale for schizophrenia (PANSS).³⁰ All patients were under antipsychotic medication; type and dose were left to the treating physician. Fourteen patients were treated in monotherapy, 14 received combination therapy of 2 preparations, and 2 patients received combination therapy of 3 preparations. Additionally, 7 patients were treated with antidepressants, 3 with anxiolytics, one with mood stabilizers, and one with low potency neuroleptics. Chlorpromazine equivalent of antipsychotic medication dosage was calculated according to published conversion tables.^31–33

Experimental Task

We evaluated whether the value of engaging in a social situation varied between patients and controls using a recently developed social decision-making task.²⁶ On each trial of the task (figures 1A and 1B), participants chose between 2 options: a risky monetary gamble with 2 potential known outcomes (0 or 3 Euros), or a safe option consisting of a certain amount of money (values between 0 and 3 Euros, varied in 50 cent steps). The gamble consisted of a game of dice against 6 different virtual human opponents (figure 1C), and a computer as control. Human and computer opponents were used in separate blocks of 126 trials. Importantly, the gamble included a display of the opponent’s reaction to the outcome (figure 1D): The human partner’s reaction consisted of a positive facial expression (ie, admiration) or negative facial expression (ie, condescension) depending on whether the participant won or did not win (respectively), presented as a 1-second duration video sequence. The computer’s reactions were abstract symbols (figure 1B). The choice frequencies of the monetary amounts offered in the safe option allowed us to identify the individual value of engaging in the game against both the human and the computer opponent; we took the difference between these values as a quantitative measure for the participants’ social avoidance. The individual value of the gambles against the human and computer partners were estimated by fitting a cumulative Gaussian function with lapse rate³⁴ to the choice frequencies for each amount offered in the safe option, and reading off the calculated amount of money to be offered in the safe option that would lead to 50% of safe and risky choices (Certainty Equivalent, CE). Fitting a cumulative Gaussian function with lapse rate is widely accepted in psychophysics and considers participant responses to follow gradual changes as a function of an independent variable, combined with a proportion of button press errors (the lapse rate). The approach yielded 2 additional estimates: (1) an estimate of the variability of the participants’ response pattern (the width of the fitted Gaussian, with larger widths corresponding to flatter curves and more variable responses), and (2) an estimate of the amount of decisions that are likely to be stimulus-independent errors, so-called “lapses,” ³⁴ which are choices deviating from the rational, expected choices (for a safe option of 0 Euros choose “risky,” and for a safe option of 3 Euros choose “safe,” respectively).

Figure 1.

Experimental task. In each trial of the task, participants chose between a risky and a safe option, which yielded different outcomes, including positive or negative feedback. There were 2 conditions: (A) a social condition with a human partner and (B) a control condition with a computer opponent. (C) The 6 human gambling partners with neutral facial expression. (D) Illustration of the facial expressions of one of the partners: neutral, admirative, and condescending (left to right; the latter 2 images are stills from the video stimuli). Image reproduced from Schultz et al. 2019, available under the terms of the CC-BY 4.0 license (https://creativecommons.org/licenses/by/4.0/).

Clinical Assessment Instruments

In addition to the PANSS and Item 8 of the CDSS questionnaire mentioned above, we used the following assessment procedures. Social anxiety traits were assessed using a standard questionnaire, the LSAS. We used the German language edition of the questionnaire.²⁴ To better characterize our patient sample, we evaluated their clinical state by the Clinical Global Impression-scale (CGI)³⁵ and their social functioning (Social and Occupational Functioning Assessment Scale [SOFAS]).³⁶

Statistics

Data were analyzed using 2-tailed, 2-sample t-tests or 2-way, mixed between/within design, repeated-measures ANOVAs. Threshold for significance was P = .05.

Results

Social Anxiety Score

Patients’ LSAS were about 3.7 times higher than those of controls (mean overall score: controls 14.6; patients 51.3; t(58) = 6.08, P < .001, Cohen’s d = 1.57; table 1). Patients’ scores were much more widely distributed than those of controls, with 33% of patients showing values between 30 and 60, and 40% above 60 (figure 2A). In order to evaluate whether these differences were due to fear or avoidance scores, we analyzed the fear and avoidance subscale data using an ANOVA, which allowed us to assess the main effects of group (F(1,58) = 38.49, P < .001, = 0.40), type of subscale (F(1,58) = 0.23, P = .63, = 0.004) and the interaction between these factors (F(1,57) = 0.02, P = .90, ${\mathrm{\eta }}_{\mathrm{p}}^2$ < 0.01). These results show that differences between groups were independent of the subscale.

Figure 2.

Results. (A) Distributions of scores on the Liebowitz Social Anxiety Scale. (B) and (C) Patients’ and controls’ decisions in social and nonsocial conditions. Group-averaged data were fitted with cumulative Gaussian functions with lapse rate. (D) Response variability, quantified by the widths of the cumulative Gaussian functions fitted to each participant’s data. (E) Lapse rates estimated during the individual model fits. (F) Subjective value of the gambles with human and computer partners (estimated as the mid-point of the fitted curves = Certainty Equivalents). (G) and (H) Response times for the decision, separately displayed for both participant groups. (I) Participants’ ratings of the subjective valence of positive and negative feedback provided by the human and computer opponents. Mean and standard error of the mean across participants are shown in all panels. SCZ: patients with schizophrenia.

Experimental Task

We first evaluated whether patients and controls made similarly rational decisions using a series of tests. The data of one patient were discarded due to technical dysfunctions during data acquisition. Group-averaged data revealed that both groups’ likelihood of choosing the safe option increased as a function of the amount offered in this option, against both the human and the computer partners (figures 2B and 2C). In particular, participants of both groups chose the safe option when the amount offered in this option was higher than the expected value of the risky option (expected value = outcome * probability, ie, 3 Euro * 0.5 + 0 Euro * 0.5 = 1.5 Euro), and chose the risky option when the safe option amount was lower than the gamble’s expected value. This indicates that overall the decisions of both groups were based on rational arguments and displayed neither risk-seeking nor risk avoidance tendencies. Next, we analyzed values obtained by fitting each participant’s choice proportions with cumulative Gaussian functions (see Methods). We found that patients showed a less clear decision pattern than controls (reflected in the shallower slope of the fitted curves in figure 2B compared to figure 2C): SDs of the cumulative Gaussian function fitted to each participants’ choice frequencies were larger (indicating shallower slopes) for patients than controls, irrespective of the nature of the partner (figure 2D; main effect of group: F(1,57) = 10.98, P = .002, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.16; main effect of human vs computer partner: F(1,57) = 0.75, P = .39, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.01; interaction: F(1,57) = 0.17, P = .68,${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.003). Further, patients made more unlikely choices, irrespective of the gamble partner (Captured by the “lapses” parameter in the fitted function; figure 2E; main effect of group: F(1,57) = 5.86, P = .019, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.09; main effect of partner: F(1,57) = 0.63, P = 0.43, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.01; interaction: F(1,57) = 0.08, P = 0.78, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.001).

Based on the decision curves fitted to each participants’ choice frequencies, we calculated the certainty equivalent (CE, ie, the mid-point of the curve) of the gamble against each partner (CE Human; CE Computer), which represents a measure of the individual value of these gambles. Interestingly and to our surprise, we found no variation in the individual value as a function of the partners, participant groups or interactions between these factors (figure 2F; main effect of partner: F(1,57) = 0.66, P = .42, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.01; main effect of group: F(1,57) = 0.12, P = .74, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.002; interaction: F(1,57) = 0.23, P = .64, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.004). Crucially, the difference between the estimated values of the 2 gambles (CE Human minus CE Computer), our experimental measure of social avoidance controlled for variations in risk preference, did not vary between patients and controls (t(57) = 0.48, P = .637, Cohen’s d = 0.12). Thus, we found no evidence for larger social avoidance in patients than controls, contrary to our expectation based on the observed differences in LSAS scores.

The time needed by the participants to decide between the safe and the risky option was significantly longer for human than computer partners (figures 2G and 2H; F(1,57) = 16.0, P < .001, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.03), and this difference was larger in patients than in controls (interaction between partner and participant group: F(1,57) = 10.0, P = .003, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.01). There was no significant difference between groups (F(1,57) = 0.13, P = .72, ${\mathrm{\eta }}_{\mathrm{p}}^2$ < 0.01).

After the decision trials, participants were asked to rate the subjectively perceived valence of the stimuli used to provide feedback. These data confirmed that positive feedback was generally considered more positive than negative feedback (figure 2I; F(1,57) = 202.7, P = 2.01e^-20, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.78), that this difference was larger for human than computer partners (interaction between partners and valence: F(1,57) = 49.0, P = 3.22e^-9, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.46), and interestingly, that both effects were smaller in patients than controls (interaction between participant group and valence: F(1,57) = 4.07, P = .048, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.07; interaction between participant group, valence and partner: F(1,57) = 8.5, P = .005, ${\mathrm{\eta }}_{\mathrm{p}}^2$ = 0.13).

Because disease duration and medication (table 1) varied widely across patients, we ran post hoc analyses to assess whether disease duration (in months), dose of antipsychotic medication (in mg chlorpromazine equivalent), antidepressant use (yes or no) and anxiolytic use (yes or no) had an influence on our dependent variables (CE, response curve slope, lapse rate, valence, response times, LSAS). Tests were implemented by entering these additional variables as covariates in our analyses or running additional correlations or independent samples t-tests. We found that a higher dose of antipsychotic medication was associated with more positive average valence ratings (Pearson’s correlation: r = 0.47, P = .01), and that the patients receiving anxiolytic medication had higher LSAS values than those that did not (fear: t(27) = 2.1, P = .043, Cohen’s d = 1.3; avoidance: t(27) = 2.1, P = .045, Cohen’s d = 1.28). No other test reached significance. These additional findings were obtained with underpowered tests and should thus be interpreted with great caution.

Discussion

This preliminary study shows that patients with schizophrenia do not demonstrate social avoidance on a behavioral level in an experimental social decision-making task, despite displaying high self-rated levels of social anxiety in both the fear and avoidance subscales of the LSAS. These findings are incompatible with our initial hypothesis that social anxiety symptoms are accompanied with experimentally measured avoidance in patients with schizophrenia. In a previous study on healthy participants with the same experimental paradigm, social avoidance was positively associated with social anxiety in healthy participants.²⁶ A similar link between social anxiety and an indirect experimental social avoidance measure has been found in an approach-avoidance task with emotional faces.³⁷ The absence of social avoidance behavior despite the presence of social anxiety symptoms in our current results suggests that different mechanisms underlie social anxiety in schizophrenia compared to controls.

Two findings suggest specific differences in processing trials with human partners in patients compared to controls. The first finding is that patients took longer to decide whether to engage in social interaction. The second finding is that schizophrenia patients rated social feedback as less intense than controls: The positive feedback was rated as less positive, and the negative feedback as less negative. These differences were not found in the computer condition. Deciding whether to interact with a person represented by an image of their face requires the judgments of social attributes of that face, including attractiveness, distinctiveness, trustworthiness, or approachability.^38,39 Individuals with schizophrenia show a reduced ability to recognize and interpret affective information in faces.^40–46 An explanation linking both findings could be that patients’ difficulties in extracting decision-relevant social information from faces, including delayed facial affect processing,⁴⁷ may prolong their social decision-making process.

We would like to propose the following hypothetical link between these delays, patients’ social avoidance expressed in elevated self-rated LSAS score, and the lack of avoidance in our experimental task. Delayed social signal processing and difficulties in interpreting social signals lead to delayed responses to social signals of interaction partners, resulting in challenges in real-life interactions and avoidance of future interactions. In sum, they avoid social interactions because of their difficulties in dealing with them. This is different from social anxiety disorder, where people try to avoid social interactions in part due to negatively biased interpretations of social feedback.^48,49 In our experiment, there were no time constraints on the interpretation of the stimuli or on the decision. Longer processing times may have allowed patients to partially compensate their processing difficulties that are here expressed as longer delays, thereby avoiding the time pressure of real-life social interactions. Our patients were not overwhelmed by the decision and made overall rational decisions (see next paragraph). Without the problems they encounter in real life, there was no reason to avoid the social gamble in the experiment, and thus we did not find any sign of social avoidance. An elegant test of this hypothesis would be to repeat the experiment under time constraints, which should lead patients to avoid the interaction.

Given the known range of cognitive impairments in schizophrenia,^50–52 it was important to show that patients could adequately understand and perform our task. Patients followed a systematic, rational approach in their responses, although their response variability was higher than controls (figure 2D) and they made more unlikely choices (described as lapses in the Results; figure 2E). Overall, however, their response pattern was clearly rational: patients, as well as controls, tended to choose the safe option as long as its expected value was higher than that of the gamble, and preferred the gamble otherwise. This suggests that the unexpected absence of social avoidance observed in the task is unlikely to have been due to a deficit in cognitive capacities or task comprehension.

We found that 33% of our schizophrenia patient sample had an LSAS value between 30 and 60, and 40% above 60. A value of 30 is the recommended cutoff value for classifying patients with social anxiety disorder and a cutoff of 60 to classify generalized social anxiety disorder.^53,54 Thus, 73% had relevant symptomatology. These values appear high with regard to the literature: eg, a recent meta-analysis reported a 21% prevalence of social anxiety disorder in patients with psychosis.¹⁷ One could assume that antidepressant and anxiolytic medication could have had an influence on LSAS values, in particular on the avoidance subscale. While scores did vary with antidepressant medication, LSAS did vary with anxiolytic medication. These associations should be investigated in further studies with larger patient samples.

What might contribute to the explanation of this discrepancy between reported rates of social anxiety disorder and our high LSAS values is that (1) LSAS scores above these cutoffs are not directly equivalent to a diagnosis of social anxiety disorder, which requires consideration of the overall clinical presentation without psychotic symptoms and further assessments, (2) the LSAS is considered to be a very sensitive measure of social anxiety symptoms in patients with psychosis,⁵⁵ and (3) the studies included in the meta-analysis cited above used partly different measures than the LSAS. LSAS values were about 4 times higher in patients compared to controls and interestingly, showed a relatively broad distribution across the sample (figure 2A). Social anxiety symptoms of varying severity were prevalent in nearly three quarters of the patients in our sample, suggesting that social anxiety symptoms in schizophrenia patients represent a spectrum and are not either very low (no comorbid social anxiety) or very high (comorbid social anxiety present). From this perspective, social anxiety would be intrinsic to the psychosis diathesis.^56,57 We believe that the high frequency of occurrence of the symptoms and the widespread distribution of symptom severity support this interpretation.

In light of potential differences in the etiology of social avoidance in schizophrenia and people without schizophrenia, common and habitual treatment strategies for social anxiety, as eg, exposure therapy, may not be highly effective in schizophrenia. There are no operational guidelines for the treatment of social anxiety symptoms in schizophrenia. Nonetheless, as social anxiety is highly prevalent in schizophrenia patients, in clinical practice, it is necessary to carefully monitor for social anxiety symptoms, because these symptoms affect patients’ everyday functioning and psychoeducation can help to reduce the impact of those symptoms. If evidence for different mechanisms of social anxiety and avoidance in schizophrenia are confirmed in subsequent studies, adapted psychosocial treatment options need to be developed to adequately address social anxiety in schizophrenia. If the reduced speed of processing and the reduced sensitivity for social signals in schizophrenia are related to social anxiety, then improvements in social cognition may have a beneficial effect on social anxiety symptoms. Treatment approaches that directly target key determinants of social functioning gave evidence of improvement in primary target domains, eg, facial affect recognition performance.^58–60 Whether these functional improvements also improve social anxiety symptoms has not been investigated so far and should be addressed in clinical trials.

Limitations

Several limitations of this study need to be acknowledged. First, our patient sample was small, consisted of in- and outpatients, and differed from the controls in more factors than just the diagnosis of schizophrenia. While basic parameters such as age and sex were matched across groups, we did not match the number of years of education and did not assess IQ using questionnaires. However, importantly our measure of social avoidance is not very susceptible to variations in task performance: while less systematic decisions lead to “flatter” curves (evidenced by larger widths of the cumulative Gaussian functions fitted to the data, as seen in the patient group), our measure of social avoidance is the difference between the mid-point of the fitted curves, which is independent of the steepness of the curve. Therefore, cognitive capacity is unlikely to explain the lack of social avoidance in our patient group. Second, we could have evaluated the presence of social anxiety disorder. This would have allowed to relate LSAS to clinical diagnoses and facilitate comparison with previous studies. Lastly, comparing patients and controls in a real-world (ie, out-of-lab) experimental measure of social avoidance may allow to better explain the difference between our experimental and questionnaire measures of social avoidance. Indeed, our laboratory experiment was necessarily somewhat artificial: pictures and videos of facial expressions, as well as the clearly defined outcomes of the interaction, are not identical to the social stimuli and the consequences of social interactions encountered in the real world. Further, as discussed earlier, the timing of the decisions was also artificial: while they were fully under participant control in our task, they depend on both interaction partners in real life.

Conclusion

Social anxiety is a major problem in schizophrenia. This study shows a dissociation between the self-experienced social anxiety and experimentally assessed social avoidance behavior in a newly developed experimental task. Taken together, our findings let us to propose that the social anxiety symptomatology is a subtle, but constitutive feature of patients with schizophrenia. In an experimental context, this may not become fully visible if patients have enough time to consider their responses in highly structured interaction but may become manifest in an everyday context in which fast decisions are required in complex social environments.