Negative Symptoms Independently Predict Theory of Mind Impairment in Clinically Stable Schizophrenia Authors: Burcu Eser, Esma Akpınar Aslan, Sedat Batmaz, Sare Aydın Published: 2026 ABSTRACT Objectivetheory of mind (ToM) impairment is a well-established feature of schizophrenia; however, the relative contribution of negative symptoms and disorganization remains unclear, particularly in clinically stable outpatients. Moreover, few studies have modeled symptom domains simultaneously while accounting for illness chronicity and antipsychotic exposure.Methodin this cross-sectional case–control study, 52 clinically stable outpatients with schizophrenia and 52 healthy controls completed the Reading the Mind in the Eyes Test (RMET). Symptom dimensions were derived from the Positive and Negative Syndrome Scale (PANSS) using a five-factor framework. Hierarchical regression analyses examined independent predictors of RMET performance within the schizophrenia group. Sensitivity analyses included illness duration, olanzapine-equivalent dose, and alternative disorganization operationalizations. Receiver operating characteristic (ROC) analysis evaluated discriminative capacity.Resultspatients demonstrated significantly lower RMET scores than controls (d = −1.42, p < .001). Within the schizophrenia group, negative symptoms (β = −.33, p = .015) independently predicted RMET performance after controlling for demographic variables and disorganization. This association remained robust after adjusting for illness duration and medication dose. Disorganization showed modest bivariate associations but did not retain independent predictive value in multivariable models. RMET demonstrated good discriminative accuracy (AUC = 0.84), with high specificity at the optimal cut-off.ConclusionsToM impairment in clinically stable schizophrenia is substantial and appears more strongly linked to negative symptom severity than to disorganization. These findings support routine assessment of social cognition in patients with prominent negative symptoms and highlight the dimensional interplay between motivational deficits and social-cognitive dysfunction. Introduction Theory of mind (ToM), the capacity to attribute mental states such as beliefs, intentions, and emotions to others, represents a central domain of social cognition. Over the past two decades, ToM impairment has emerged as one of the most consistently replicated findings in schizophrenia research. A seminal meta-analysis by Sprong et al. (2007) demonstrated robust ToM deficits across illness stages, including first-episode and remitted patients. Subsequent quantitative syntheses confirmed large effect sizes relative to healthy controls (Bora, Yücel, & Pantelis, 2009). In a broader meta-analysis of social cognition domains, Savla et al. (2013) reported that ToM deficits were among the largest impairments observed in schizophrenia and were strongly associated with functional disability. Importantly, ToM impairment appears only partially explained by general neurocognitive deficits. While cognitive functioning contributes to performance variability, meta-analytic evidence indicates that ToM remains impaired even after accounting for IQ differences (Bora et al., 2009; Harrington et al., 2005). These findings suggest that social inferential processing represents a partially independent domain of dysfunction rather than merely a downstream effect of global cognitive decline. From a clinical standpoint, ToM deficits manifest as misinterpretation of social cues, reduced sensitivity to contextual meaning, and difficulties in perspective-taking. These impairments are strongly linked to real-world functional outcomes (Fett et al., 2011; Schmidt et al., 2011), underscoring their clinical significance beyond symptom severity. The relationship between negative symptoms and ToM has attracted increasing attention. Negative symptoms, particularly avolition, anhedonia, alogia, and blunted affect, are among the strongest predictors of poor functional outcome and remain resistant to pharmacological treatment (Kirkpatrick et al., 2006; Correll & Schooler, 2020). A meta-analysis by Ventura et al. (2013) examined the association between negative symptoms and social cognition and found a moderate but consistent negative relationship across studies. This association remained significant even after accounting for general neurocognitive performance. Similarly, Savla et al. (2013) reported that negative symptoms were more strongly associated with social cognition deficits than positive symptoms. Several mechanisms have been proposed. One possibility is motivational: Reduced social engagement may limit opportunities for social learning and calibration of mental state inference. Another possibility is that negative symptoms and ToM deficits share underlying neurocognitive or neurodevelopmental substrates. Notably, Bora et al. (2009) observed that ToM deficits persist in remitted states, suggesting that their association with negative symptoms is not simply an artifact of acute psychosis. Despite these findings, relatively few studies have tested whether negative symptoms remain independent predictors of ToM performance after controlling for other symptom dimensions, particularly disorganization. Disorganization, reflecting disturbances in thought structure, executive control, and cognitive integration, is another symptom domain plausibly linked to ToM impairment. Disorganized thinking may interfere with maintaining task rules, integrating contextual cues, and sustaining attention—all processes relevant to ToM tasks (Barch & Ceaser, 2012; Minor & Lysaker, 2014). Early work suggested that formal thought disorder was associated with poorer ToM performance (Brüne, 2005). Empirical studies have reported correlations between disorganization scores and ToM measures, particularly in tasks requiring narrative integration or belief reasoning. However, findings have been less consistent than those observed for negative symptoms. Ventura et al. (2013) noted that associations between social cognition and symptom dimensions often weaken when multiple symptom domains are entered simultaneously into regression models. This raises the possibility that observed correlations between disorganization and ToM may reflect shared variance with negative symptoms or broader cognitive deficits. Negative symptoms and disorganization are conceptually distinct but clinically overlapping domains. Factor-analytic studies of PANSS data demonstrate moderate correlations between these dimensions (van der Gaag et al., 2006). In practice, reduced spontaneous speech (alogia) may be misinterpreted as disorganization, and conversely, impoverished output may co-occur with cognitive fragmentation. This overlap complicates interpretation of ToM findings. If both domains correlate with ToM, it becomes unclear whether social cognitive deficits are primarily motivational-expressive in nature (negative symptoms) or reflect executive/cognitive integration problems (disorganization). Without modeling both dimensions simultaneously, conclusions risk being oversimplified. The NIMH workshop on social cognition emphasized the need for multivariate modeling approaches to disentangle symptom–cognition relationships (Green et al., 2008). Yet many studies have relied on bivariate correlations, leaving the independence of these associations unresolved. In the meta-analysis by Bora et al. (2009), symptom severity showed modest associations with ToM performance overall. Ventura et al. (2013) reported that negative symptoms had stronger and more consistent associations with social cognition than positive symptoms. However, few studies have simultaneously controlled for disorganization and demographic variables in regression frameworks, leaving uncertainty about independent effects. Theory of mind has been assessed in schizophrenia research using a range of experimental paradigms, each tapping slightly different aspects of mental state reasoning. Early studies frequently relied on false-belief tasks, including first- and second-order belief reasoning paradigms adapted from developmental psychology, which evaluate the ability to represent another person’s mistaken belief. More linguistically demanding instruments, such as the Strange Stories Test (Happé, 1994), assess understanding of non-literal language, irony, deception, and socially embedded intentions. The Faux Pas Test requires participants to detect socially inappropriate remarks within narrative contexts and to explain the underlying mental states of the characters involved. The Hinting Task, widely used in schizophrenia research, focuses on indirect speech comprehension and pragmatic inference. Finally, the Reading the Mind in the Eyes Test (RMET) assesses the ability to infer complex mental states from photographs depicting only the eye region of faces, thereby emphasizing rapid mental state attribution based on subtle perceptual cues rather than extended narrative processing. Meta-analytic evidence indicates that performance deficits are observed across task types (Sprong et al., 2007; Bora et al., 2009). However, tasks differ in cognitive load. Narrative-based tasks (e.g., Strange Stories) demand sustained attention and verbal working memory, potentially confounding ToM with executive functioning. The Faux Pas Test requires integration of contextual information over extended passages, increasing memory demands. The Hinting Task, though widely used, can be influenced by pragmatic language ability. RMET occupies a somewhat distinct position. It presents minimal contextual information and focuses on rapid mental state inference from visual cues. This reduces memory load relative to narrative tasks but introduces reliance on vocabulary and perceptual discrimination (Oakley et al., 2016). RMET has become one of the most frequently used ToM measures in schizophrenia research, partly because of its brevity and ease of administration (Bora et al., 2009). Studies using RMET consistently demonstrate impaired performance in schizophrenia relative to controls (Bora et al., 2009; Savla et al., 2013). Effect sizes are typically large. Some studies report associations between RMET and negative symptoms, whereas relationships with positive symptoms are weaker or inconsistent. RMET has been adapted and used in Turkish samples, including both clinical and non-clinical populations. Turkish validation studies have reported acceptable reliability and expected normative distributions in healthy adults (Yıldırım et al., 2011). Clinical studies conducted in Turkish schizophrenia samples similarly report lower RMET performance compared to controls, consistent with international findings. However, systematic examination of the independent roles of negative symptoms and disorganization in Turkish outpatient samples remains limited. Although theory of mind impairment in schizophrenia is well established, the field still lacks clarity regarding which symptom dimensions most meaningfully account for individual variability in ToM performance. Many studies have relied on bivariate correlations or have examined negative symptoms and disorganization separately, without modeling their overlapping variance within the same analytic framework. As a result, it remains uncertain whether ToM deficits are primarily driven by motivational-expressive pathology, by cognitive disorganization, or by their shared variance. This ambiguity is particularly relevant in clinically stable outpatient samples, where acute psychotic symptoms are minimized and enduring symptom dimensions become more salient. A more precise understanding of the independent contributions of negative symptoms and disorganization to ToM impairment would not only refine theoretical models of social cognition in schizophrenia but also inform clinical assessment and intervention priorities. The present study was designed to address this gap directly by examining ToM performance in clinically stable outpatients and simultaneously modeling negative symptoms and disorganization within a hierarchical regression framework. This gap is clinically relevant, as cultural context may influence social cue interpretation and mental state vocabulary. Examining RMET performance and its symptom correlates within Turkish samples contributes to cross-cultural validation of the broader social cognition literature. Taken together, the existing literature indicates that theory of mind impairment is a robust and well-replicated finding in schizophrenia. Negative symptoms have demonstrated relatively consistent associations with social cognition deficits, whereas the contribution of disorganization appears more variable across studies. At the same time, the conceptual and statistical overlap between negative symptoms and disorganization complicates interpretation, particularly in studies relying solely on bivariate associations. Although the Reading the Mind in the Eyes Test is among the most frequently used instruments in this field, relatively few investigations have modeled symptom dimensions simultaneously in order to clarify their independent contributions to ToM performance. The present study was designed to address these gaps by examining ToM performance in clinically stable outpatients with schizophrenia and by testing the independent effects of negative symptoms and disorganization within a hierarchical regression framework. The present study had three primary objectives. First, we sought to compare theory of mind performance, as measured by the RMET, between clinically stable outpatients with schizophrenia and healthy controls. Second, we aimed to determine whether negative symptom severity independently predicts RMET performance within the schizophrenia group after controlling for disorganization and relevant demographic variables. Third, we examined the discriminative capacity of RMET in differentiating patients from controls using receiver operating characteristic (ROC) analysis. We hypothesized that individuals with schizophrenia would demonstrate significantly lower RMET scores than healthy controls, that negative symptoms would remain an independent predictor of ToM performance even after accounting for disorganization and demographic factors, and that RMET would show good discriminatory performance in ROC analysis. Methods Study design and ethical approval This cross-sectional case–control study was conducted in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines (von Elm et al., 2007). The study protocol was approved by the Clinical Research Ethics Committee of Gaziosmanpaşa University Faculty of Medicine (Approval date: 06.02.2018 #: 17-KAEK-194), and all participants provided written informed consent prior to participation. All procedures complied with the Declaration of Helsinki and its subsequent amendments. Participants Fifty-two schizophrenia outpatients aged 18–65 years who were being followed at a university-affiliated psychiatry outpatient clinic with a diagnosis of schizophrenia were included. Diagnosis was confirmed using the Structured Clinical Interview for DSM-5 Disorders – Clinician Version (SCID-5-CV) (First et al., 2016), administered by trained psychiatrists. Clinical stability was operationally defined as: (i) Absence of hospitalization within the previous three months, (ii) No acute psychotic exacerbation within the previous four weeks, (iii) No major changes in antipsychotic medication dose during the previous four weeks, and (iv) PANSS total score not exceeding moderate severity levels at the time of assessment. Exclusion criteria included: (i) Intellectual disability, (ii) Neurological disorders affecting cognitive functioning, (iii) Current substance use disorder, (iv) Comorbid major psychiatric disorders, and (v) Uncorrected visual impairment. Information regarding illness duration, number of hospitalizations, and olanzapine-equivalent antipsychotic dose was obtained from medical records and clinical interview. Fifty-two healthy controls aged 18–65 years were recruited from the community. Absence of current or lifetime psychiatric disorder was confirmed using a brief clinical diagnostic interview. Controls were screened for neurological disorders, substance use disorders, and major medical conditions affecting cognition. Exclusion criteria paralleled those of the patient group. Measures Theory of mind Theory of mind was assessed using the RMET (Baron-Cohen et al., 2001). The Turkish validated version was used (Yildirim et al., 2011). The instrument consists of 32 photographs depicting only the eye region of faces. For each item, participants select one of four mental state descriptors. Total scores range from 0 to 32, with higher scores indicating better mental state inference ability. In the present sample, internal consistency was acceptable (Cronbach’s α = 0.822). Symptom severity Symptom severity was evaluated using the Positive and Negative Syndrome Scale (PANSS; Kay et al., 1987). Ratings were performed by trained psychiatrists who were blinded to RMET performance. Negative and disorganization dimensions were derived according to the validated five-factor structure proposed by van der Gaag et al. (2006). The negative dimension included blunted affect, emotional withdrawal, passive/apathetic social withdrawal, lack of spontaneity, and motor retardation. The disorganization dimension included conceptual disorganization, difficulty in abstract thinking, poor attention, and related cognitive-disorganization items. In the present sample, internal consistencies of the PANSS dimensions were acceptable (Cronbach’s α ranging from 0.675 [positive symptoms] to 0.869 [negative symptoms]). The Turkish version was used in this study (Kostakoglu et al., 1999). Demographic and clinical variables Age and sex were recorded. Education was categorized into three levels: low (<5 years), medium (5-12 years), and high (>12 years). Education category was entered into regression analyses using dummy coding (reference category = low). Information regarding illness duration (months since diagnosis), number of lifetime hospitalizations, and current olanzapine-equivalent antipsychotic dose (mg/day) was obtained from medical records and clinical interview. Olanzapine-equivalent dose was calculated according to established dose-conversion guidelines (Leucht et al., 2016). These variables were included in secondary sensitivity analyses to examine whether the association between negative symptoms and RMET performance remained significant after accounting for illness chronicity and medication exposure. Procedure Participants attended a single face-to-face assessment session in the psychiatry outpatient clinic. After diagnostic confirmation with SCID-5-CV (Elbir et al., 2019), participants completed the RMET. RMET instructions were read verbatim. No time limit was imposed, though participants were encouraged to respond without prolonged deliberation. The interviewer recorded the demographic and clinical data, and administered the PANSS. The whole interview lasted around 30-60 minutes. Statistical analysis All statistical analyses were conducted using Jamovi (Version 2.7) (The jamovi Project, 2025). All tests were two-tailed, with statistical significance set at p < .05. Preliminary analyses Descriptive statistics were calculated for demographic and clinical variables. Continuous variables were expressed as mean ± standard deviation. Categorical variables were presented as frequencies and percentages. Group comparisons were performed using Welch’s independent samples t-tests for continuous variables, and Chi-square tests for categorical variables. Effect sizes were reported as Cohen’s d with 95% confidence intervals. Correlational analysis Within the schizophrenia group, Pearson correlation coefficients were calculated to examine associations between RMET performance, negative symptoms, and disorganization. Hierarchical regression analysis Hierarchical multiple linear regression analysis was performed to identify independent predictors of RMET total scores within the schizophrenia group. Three sequential models were specified: Model 1: Age, sex, education category, Model 2: Model 1 + disorganization, and Model 3: Model 2 + negative symptoms. Standardized beta coefficients (β), R2, and ΔR2were reported. Multicollinearity was evaluated using variance inflation factors (VIF), with values below 5 considered acceptable. Assumptions of linearity, homoscedasticity, and normality of residuals were examined through diagnostic plots. Sensitivity analyses To evaluate the robustness of the primary regression findings, additional hierarchical models were conducted within the schizophrenia group. Illness duration and olanzapine-equivalent dose were entered sequentially into the final regression model. Changes in explained variance (ΔR2) and regression coefficients were examined to determine whether clinical variables attenuated the association between negative symptoms and RMET performance. ROC analysis ROC analysis was conducted to evaluate the discriminative ability of RMET in differentiating patients with schizophrenia from healthy controls. Area under the curve (AUC) values and 95% confidence intervals were calculated using nonparametric methods. The optimal cut-off score was determined using the Youden index, and corresponding sensitivity and specificity values were reported. In addition to sensitivity and specificity, positive and negative likelihood ratios (LR+ and LR−) were calculated to provide clinically interpretable indices of diagnostic utility. Because predictive values depend on disorder prevalence, likelihood ratios were preferred as prevalence-independent measures of classification performance. The optimal cut-off value was determined using the Youden index; however, it was interpreted as sample-specific and exploratory rather than as a universal diagnostic threshold. Sample size considerations An a priori power estimation was conducted to determine the minimum detectable effect size for group differences in RMET performance. Based on previous meta-analytic findings reporting large effect sizes for theory of mind impairment in schizophrenia (Bora et al., 2009; Savla et al., 2013), a conservative medium-to-large effect size (d = 0.70) was assumed for planning purposes. With α = .05 (two-tailed), a total sample of 104 participants (52 per group) provides approximately 80% statistical power to detect an effect of this magnitude in independent samples comparisons. For the hierarchical regression analyses conducted within the schizophrenia group (n = 52), the sample size was sufficient to detect medium effect sizes (f2 ≈ 0.15) with up to five predictors while maintaining a power of approximately 0.80 at α = .05. Given that the final regression model included four predictors, the ratio of cases to predictors exceeded the commonly recommended minimum of 10:1, supporting model stability. These considerations suggest that the study was adequately powered to detect clinically meaningful effects in both between-group and within-group analyses. Results Preliminary analyses and sample characteristics The final sample consisted of 52 outpatients with schizophrenia and 52 healthy controls. Groups did not differ significantly in age, t(101.81) = 0.32, p = .753, or education category distribution, χ2(2) = 0.58, p = .748 (table 1). Sex distribution was identical across groups, χ2(1) = 0.00, p = 1.00. Descriptive statistics for RMET scores and demographic variables are presented in table 1. Table 1.Demographic characteristics and RMET performanceVariableSchizophrenia (n=52)Controls (n=52)Statistic p Effect Size (95% CI)Age (years), M ± SD37.33 ± 12.0636.60 ± 11.55t(101.81) = 0.32 .753—Education (Low/Medium/High), n2 /20 /301 /18 /33χ2 (2) = 0.58 =.748—Sex (Female/Male), n25 /2725 /27χ2 (1) = 0.00 =1.00—RMET total score, M ± SD13.23 ± 4.9020.35 ± 5.13t(101.79) = -7.24 < .001d = −1.42 [-185, −0.99] Demographic characteristics and RMET performance Group differences in theory of mind Patients with schizophrenia demonstrated significantly lower RMET scores compared to healthy controls, t(101.79) = -7.24, p < .001. The magnitude of this difference was large (d = −1.42, 95% CI [−1.85, −0.99]), indicating substantial impairment in mental state inference. Correlational analyses within the schizophrenia group Within the schizophrenia group, RMET performance was significantly negatively correlated with negative symptom severity, r = −.45, p = .001, and with disorganization, r = −.37, p = .007. The correlation between negative symptoms and disorganization was moderate (r = .43, p = .001) (table 2). Table 2.Correlation Matrix Within the Schizophrenia Group (n = 52)Variable 1 231. RMET—2. Negative symptoms−.45*—3. Disorganization−.37*.43*— Note. *p < .01. Correlation Matrix Within the Schizophrenia Group (n = 52) Note. p < .01. Hierarchical regression analysis A hierarchical multiple regression analysis was conducted to identify independent predictors of RMET performance within the schizophrenia group. Model 1 (age, sex, education category) was significant, F(4, 47) = 4.28, p = .005, explaining 26.7% of the variance (R2 = .27). Adding disorganization in Model 2 significantly improved the model, F(5, 46) = 4.78, p = .001, with total explained variance increasing to 34.2% (R2 = .34; ΔR2 = .075). In the final model (Model 3), inclusion of negative symptoms resulted in a further significant increase in explained variance, F(6, 45) = 5.54, p < .001, yielding R2 = .43 (ΔR2 = .083). In the fully adjusted model, negative symptoms remained a significant independent predictor of RMET performance (B = −0.34, SE = 0.13, β = −.33, t(45) = −2.54, p = .015), whereas disorganization was no longer statistically significant (B = −0.63, SE = 0.57, p = .269). Variance inflation factors ranged from 1.12 to 1.39, indicating no evidence of multicollinearity. Detailed regression coefficients are presented in table 3. Table 3.Hierarchical Regression Predicting RMET Scores (Schizophrenia Group, n = 52)PredictorBSE95% CIβ t p Age−0.010.05[−0.11, 0.09]−.02−0.11.914Sex−1.511.18[−3.82, 0.80]−.16−1.28.207Education (Medium)5.313.22[−1.00, 11.62].231.65.106Education (High)8.603.29[2.15, 15.05].362.62.012Disorganization−0.630.57[−1.75, 0.49]−.17−1.12.269Negative symptoms−0.340.13[−0.59, −0.09]−.33−2.54.015Note. Model 1: R2 = .27. Model 2: R2 = .34, ΔR2 = .08. Model 3: R2 = .43, ΔR2 = .08. Hierarchical Regression Predicting RMET Scores (Schizophrenia Group, n = 52) Note. Model 1: R2 = .27. Model 2: R2 = .34, ΔR2 = .08. Model 3: R2 = .43, ΔR2 = .08. Clinical characteristics of the schizophrenia sample Within the schizophrenia group, the mean illness duration was 129.06 months (SD = 98.52). The mean number of lifetime hospitalizations was 2.22 (SD = 2.62). The mean olanzapine-equivalent dose was 15.00 mg/day (SD = 8.55). Sensitivity analyses To examine whether illness chronicity or medication exposure accounted for the association between negative symptoms and RMET performance, two additional hierarchical regression models were conducted within the schizophrenia group. In the first sensitivity model, illness duration (months since diagnosis) was added to the final regression model. The overall model remained statistically significant, F(7, 44) = 4.78, p < .001, explaining 43.4% of the variance in RMET scores (R2 = .434). The addition of illness duration did not produce a meaningful increase in explained variance (ΔR2 = .009). Illness duration itself was not a significant predictor (B = −0.002, SE = 0.003, t(44) = −0.71, p = .483). Negative symptoms remained a statistically significant independent predictor of RMET performance (B = −0.38, SE = 0.14, t(44) = −2.66, p = .011). In the second sensitivity model, olanzapine-equivalent dose (mg/day) was added to the model including illness duration. The model remained significant, F(8, 43) = 4.33, p < .001, with total explained variance of 44.1% (R2 = .441). The inclusion of medication dose resulted in a minimal increase in variance explained (ΔR2 = .007). Olanzapine-equivalent dose was not a significant predictor (B = −0.02, SE = 0.05, t(43) = −0.39, p = .701). Negative symptoms continued to show a statistically significant association with RMET performance (B = −0.39, SE = 0.15, t(43) = −2.63, p = .012). ROC analysis ROC analysis demonstrated that RMET significantly discriminated between patients with schizophrenia and healthy controls. The AUC was 0.838 (95% CI [0.756, 0.910]), indicating good overall classification accuracy (figure 1). The optimal cut-off score of 14 yielded a sensitivity of 65.4% and a specificity of 86.5%. The corresponding positive likelihood ratio (LR+) was 4.84, and the negative likelihood ratio (LR−) was 0.40. These values indicate that very low RMET scores substantially increase the probability of belonging to the schizophrenia group, whereas higher scores do not fully exclude the presence of illness. ROC Curve for RMET Performance Discussion Overview of principal findings This study examined ToM performance in clinically stable outpatients with schizophrenia and tested whether negative symptoms independently predict ToM performance beyond disorganization and demographic variables. Three main findings emerged. First, patients demonstrated markedly lower RMET performance compared to healthy controls, with a large effect size (d = −1.42). Second, within the schizophrenia group, both negative symptoms and disorganization were correlated with RMET scores; however, when modeled simultaneously, only negative symptoms remained an independent predictor. Third, RMET showed good discriminative ability in distinguishing patients from controls (AUC = 0.838). Together, these findings support our primary hypotheses and help clarify the relative contribution of symptom dimensions to social-cognitive impairment. Magnitude of ToM impairment The large between-group effect size observed in this study is consistent with prior meta-analytic findings. Bora, Yücel, and Pantelis (2009), in a comprehensive meta-analysis, reported large impairments in ToM across illness stages, including chronic and remitted samples. Similarly, Savla et al. (2013) found that ToM deficits were among the most robust social-cognitive impairments in schizophrenia. The magnitude observed here (d = −1.42) falls squarely within the upper range of previously reported effects, reinforcing the notion that ToM impairment is not merely a subtle cognitive inefficiency but a clinically substantial deficit. Importantly, our sample consisted of clinically stable outpatients rather than acutely psychotic individuals. This strengthens the interpretation that ToM impairment is not exclusively driven by acute symptom exacerbation. Sprong et al. (2007) similarly reported persistent ToM deficits in remitted patients, suggesting that social-cognitive impairment may represent a relatively stable characteristic of the disorder. One possible difference from some earlier studies lies in the magnitude of impairment. In certain samples, effect sizes are moderate rather than large. Differences in illness chronicity, medication exposure, or educational background may account for this variability. Our sample’s relative clinical stability may paradoxically amplify the visibility of enduring social-cognitive deficits, as acute psychotic noise is minimized. The persistence of marked ToM impairment in a clinically stable outpatient sample also contributes to the longstanding debate regarding whether social-cognitive deficits in schizophrenia reflect state-related symptom fluctuations or more enduring trait-like vulnerabilities. Prior work has suggested that ToM impairments are observable even outside acute psychotic episodes and may represent relatively stable characteristics of the disorder (Pinkham et al., 2003; Sprong et al., 2007). The present findings, obtained in a clinically stable cohort, lend further support to the view that ToM dysfunction is not solely a byproduct of acute symptom exacerbation but may reflect a more persistent dimension of schizophrenia. Negative and disorganization symptoms as predictors of ToM Within the schizophrenia group, RMET performance correlated significantly with both negative symptoms and disorganization. However, hierarchical regression revealed that negative symptoms remained an independent predictor after controlling for disorganization and demographic variables, whereas disorganization did not. This pattern aligns with findings from Ventura, Wood, and Hellemann (2013), who reported that negative symptoms show more consistent associations with social cognition than positive symptoms. Savla et al. (2013) similarly noted stronger links between negative symptoms and social-cognitive deficits relative to other symptom domains. Our modeling approach extends this literature by demonstrating that the association between negative symptoms and ToM is not reducible to overlapping variance with disorganization. In clinical practice, alogia and blunted affect may be conflated with thought disorganization; however, these findings suggest that motivational-expressive pathology carries distinct explanatory weight for mental state inference deficits. Mechanistically, several interpretations are plausible. Reduced social engagement associated with avolition and asociality may limit opportunities for practicing complex interpersonal inference. Alternatively, shared neurodevelopmental substrates may underlie both negative symptoms and ToM deficits. Bora et al. (2009) argued that ToM impairments are not fully explained by general cognitive deficits, suggesting partially independent pathways. The finding that disorganization lost significance in the fully adjusted model is noteworthy. While disorganization correlated with RMET performance at the bivariate level, its effect did not survive when negative symptoms were included. This supports concerns raised in the NIMH workshop on social cognition (Green et al., 2008) regarding the need for multivariate modeling to disentangle symptom contributions. Our data suggest that some previously reported associations between disorganization and ToM may reflect shared variance with negative symptoms. It is important not to overinterpret the non-significant regression coefficient for disorganization. The moderate correlation between negative symptoms and disorganization (r = .43) indicates meaningful overlap. Factor-analytic work on PANSS has consistently shown partial covariance between these domains (van der Gaag et al., 2006). Therefore, the absence of an independent disorganization effect does not imply irrelevance; rather, it suggests that when motivational-expressive deficits are accounted for, disorganization adds limited incremental explanatory value for RMET performance. This distinction may depend on task characteristics. Narrative-heavy ToM tasks (e.g., Strange Stories, Faux Pas) require sustained contextual integration and may show stronger links with disorganization. RMET, by contrast, relies on rapid inference from minimal cues. The relative cognitive demands of RMET may make it more sensitive to motivational or attentional engagement rather than extended executive control. Discriminative capacity of RMET Although RMET was not designed as a diagnostic instrument, the present findings (AUC = 0.838) suggest meaningful discriminative utility in this clinical context. The optimal cut-off score of 14 demonstrated high specificity (86.5%) and moderate sensitivity (65.4%), yielding a positive likelihood ratio of 4.84. This pattern suggests that markedly low RMET performance substantially increases the likelihood of schizophrenia within a comparable outpatient setting. However, the moderate sensitivity and LR− of 0.40 indicate that intact or near-intact RMET performance does not reliably exclude the diagnosis. Importantly, no universally accepted RMET cut-off exists for schizophrenia. Most prior studies have focused on group differences rather than diagnostic thresholds, and reported large pooled effect sizes (Bora et al., 2009; Savla et al., 2013). Therefore, the present cutoff should be interpreted as sample-specific rather than definitive. Nevertheless, the observed AUC aligns with the large effect sizes reported in meta-analytic studies, supporting the robustness of social-cognitive separation between patients and controls. Comparison with previously reported RMET norms further contextualizes these findings. In the Turkish validation study of the 32-item RMET, healthy adults demonstrated mean scores of approximately 23–24 correct responses (Yıldırım et al., 2011). In the present study, the control group mean (20.35) was somewhat lower, possibly reflecting sample-specific demographic differences. More importantly, the patient mean (13.23) represents a substantial deviation from normative expectations. Studies involving clinically stable or remitted psychotic samples have generally reported patient RMET means in the mid-to-upper range relative to total possible scores (e.g., ~19 correct responses on the 36-item version) (Ekinci & Ekinci, 2021). Although direct comparison is limited by differences in test length, proportionally lower performance in the present sample suggests pronounced social-cognitive impairment even in clinically stable outpatients. This may reflect the relatively high negative symptom burden observed in the current cohort. Taken together, the cut-off of 14 appears to represent performance well below normative expectations, corresponding to clinically meaningful impairment rather than subtle inefficiency. Robustness of findings: illness duration and medication effects An important strength of the present study is the demonstration that the association between negative symptoms and RMET performance remained stable after controlling for illness duration and olanzapine-equivalent dose. Chronicity and medication exposure are frequently cited as potential confounders in social cognition research. However, the minimal change in explained variance and the persistence of the negative symptom effect suggest that the observed relationship may reflect a more intrinsic link between motivational-expressive pathology and mental state inference, rather than a secondary consequence of long illness duration or higher antipsychotic dosage. This robustness strengthens the interpretation that negative symptoms represent a meaningful clinical dimension associated with social-cognitive impairment. Further this robustness might be clinically meaningful by suggesting that social-cognitive impairment in patients with prominent negative symptoms should not be dismissed as merely a downstream consequence of long-standing illness or high-dose treatment. Instead, it likely reflects a core dimension of the disorder that warrants targeted clinical attention. Clinical implications These findings have several practical implications. First, the independent association between negative symptoms and ToM performance suggests that social-cognitive assessment should be considered particularly in patients with prominent negative symptom profiles. Meta-analytic evidence indicates that structured social cognition training programs can produce moderate improvements in social-cognitive performance in schizophrenia (Horan et al., 2008; Kurtz & Richardson, 2012). The present findings suggest that such interventions may be particularly relevant for patients with prominent negative symptom profiles. Second, disentangling negative symptoms from disorganization clarifies treatment targets. If ToM impairment were primarily driven by disorganization, executive or cognitive remediation might suffice. Our findings instead suggest that motivational-expressive domains warrant parallel attention. Third, by modeling symptom domains simultaneously, this study addresses a methodological gap emphasized in prior literature (Green et al., 2008). This multivariate clarification contributes incremental precision to theoretical models of social cognition in schizophrenia. Finally, the use of a Turkish outpatient sample extends cross-cultural validation of ToM findings. Cultural context may further influence the interpretation of subtle emotional cues and mental state vocabulary. Cross-cultural research suggests that theory of mind performance can vary as a function of linguistic, socialization, and emotional display norms (Wang & Su, 2013). Demonstrating similar patterns of association between negative symptoms and ToM impairment in a Turkish outpatient sample strengthens the cross-cultural generalizability of social cognition models in schizophrenia. Strengths and limitations Several features strengthen the present study. The sample consisted of clinically stable outpatients, reducing confounding from acute psychosis. Education was treated as an ordinal construct and modeled appropriately, minimizing statistical distortion. Yet, the relatively strong association between higher education and RMET performance warrants cautious interpretation. Education may proxy not only formal schooling but also literacy, vocabulary breadth, and exposure to complex social contexts. Future research should consider more granular indices of verbal intelligence and socioeconomic background to disentangle educational and cognitive influences. The hierarchical regression design allowed simultaneous modeling of symptom domains, directly addressing the overlap between negative symptoms and disorganization. Furthermore, ROC analysis provided clinically interpretable information regarding classification accuracy. Importantly, by demonstrating that negative symptoms retain explanatory power when disorganization is included, the study clarifies a source of conceptual ambiguity in the literature. This contributes to a more nuanced understanding of social cognition in schizophrenia. However, the proposed RMET cut-off should be considered exploratory and sample-specific. External validation in independent cohorts is necessary before clinical thresholds can be generalized. Several additional limitations warrant consideration. The cross-sectional design precludes causal inference. Longitudinal studies are needed to determine whether changes in negative symptoms precede or follow changes in social-cognitive functioning. The sample size, while adequately powered for medium effects, limits detection of small associations. We did not include a comprehensive neurocognitive battery. Thus, we cannot definitively rule out partial mediation by executive functioning or processing speed. Future studies incorporating detailed neurocognitive assessment would allow more precise modeling of cognitive versus motivational contributions. Medication effects were not formally modeled, and antipsychotic dose may influence both motivational symptoms and cognitive performance. Additionally, RMET captures a specific aspect of ToM and does not encompass broader pragmatic or contextual reasoning. Despite its widespread use, the Reading the Mind in the Eyes Test (RMET) has been the subject of ongoing debate regarding its construct validity as a measure of Theory of Mind (ToM). Several authors have argued that the RMET may predominantly assess emotion recognition rather than higher-order mental state attribution, as it requires participants to infer mental states from static images of the eye region in the absence of contextual and situational cues (Peterson & Miller, 2012; Olderbak et al., 2015). Furthermore, RMET performance appears to be influenced by verbal ability and general cognitive functioning, as the task relies on the comprehension of relatively complex and nuanced mental state descriptors (Oakley et al., 2016). Another important limitation of the RMET is that it provides a unidimensional accuracy score, which precludes differentiation between qualitatively distinct mentalizing errors. In particular, the RMET does not allow for the distinction between hypomentalizing (reduced attribution of mental states) and hypermentalizing (over-attribution of mental states), the latter of which has been increasingly emphasized in schizophrenia and related disorders (Abu-Akel & Shamay-Tsoory, 2011). As a result, lower RMET scores may reflect either diminished mental state attribution or a tendency toward inaccurate over-interpretation of others’ mental states. Taken together, these limitations should be considered when interpreting the present findings. Future research employing more comprehensive and multidimensional measures of ToM may provide a more nuanced understanding of mentalizing impairments in schizophrenia. In addition, studies incorporating multiple ToM paradigms may help clarify whether the independence of negative symptoms generalizes across different task types. Conclusion In a clinically stable outpatient sample, theory of mind impairment was substantial and independently associated with negative symptom severity. Although disorganization correlated with ToM performance, its effect did not remain significant when modeled alongside negative symptoms. These findings refine existing models of social cognition in schizophrenia by clarifying the relative contribution of symptom domains. They underscore the importance of integrating social-cognitive assessment into evaluation of negative symptomatology and highlight the value of multivariate modeling approaches in disentangling overlapping clinical constructs.