Participants were drawn from the Nathan Kline Institute-Rockland study (NKI-Rockland) a large (>1,000 individuals) community-ascertained sample representative of the broader United States population (age 6–85) based on the 2010 census (Nooner et al., 2012). From the total cohort, 226 adult participants (age 18–65) were identified who completed measures of paranoid thinking, empathy, and emotion recognition (described below) within a 5-day period (M_age = 38.64, SD_age = 15.91; 58.4% self-identified as female, 41.6% male; 64.6% White, 20.8% Black/African American, 14.6% Asian/Native Hawaiian/American Indian/Other Race).

The NKI Rockland Sample was collected in a multi-phase National Institute of Mental Health (NIMH) funded study. Phase one collected psychiatric, behavioral, and cognitive data from 250 convenience sampled individuals from 4 to 89 years old. The second phase aimed to collect data from 1,000 participants from 6 to 85 years old who reflect the demographics of the 2009 United States census (Nooner et al., 2012). For each study visit, participants fill out self-report questionnaires (including the PDI-21 and IRI) at home within 28 days of the in-person study day. The baseline visit included 2 days of study related tasks including neuropsychological testing, neuroimaging, and behavioral tasks. A full overview of participant protocols can be found here: http://fcon_1000.projects.nitrc.org/indi/enhanced/ALGFullEndUserProtocol.pdf. The first and second phase were approved by the Institutional Review Board (Nathan Kline Institute Phase I #226781 and Phase II #239708; Montclair State University Phase I #000983A and Phase II #000983B) and participants were paid up to $200 for the baseline 2-day study or $250 if that baseline included an MRI.

Mean, standard deviation, range, and Cronbach’s alpha are presented in Table 1.

Relationships between paranoia, emotion recognition, and empathy, were first examined in zero-order correlations using SPSS v.25.0 (IBM Corp. Released, 2017). Next, a priori hypotheses were tested using structural equation modeling (SEM), conducted in R (lavaan R 4.0.2 package; Rosseel, 2012). In our a priori SEM model, each IRI subscale (exogenous variable) predicted paranoia (endogenous). Paths were also specified from each IRI subscale to overall emotion recognition ability and from emotion recognition ability to paranoia. To test our hypothesis that there would be a relationship between empathy and paranoia through emotion recognition, we estimated an indirect path from empathy (IRI subscales) to emotion recognition ability to paranoia. Our primary model used overall emotion recognition ability. Follow-up sensitivity analyses were conducted to examine whether specific types of emotion recognition (e.g., fear) were related to empathy and paranoia. All analyses included age, sex, and race as covariates. Standard errors were calculated based on 1000 bootstrapped samples. We evaluated model fit using CFI, RMSEA, and SRMR based on Hu and Bentler (1999) criteria.

Correlations between all variables are presented in Table 2. The Fantasy subscale was significantly positively correlated with paranoia (p < 0.001) and fear recognition (p < 0.001). Emotion recognition was significantly negatively correlated with Empathic Concern (p = 0.037). Paranoia and emotion recognition were not correlated with any other measure.

The main hypothesized model (Figure 1) had poor fit (CFI = 0.46, RMSEA = 0.25, SRMR = 0.08) so results should be interpreted with caution. Fantasy was significantly positively associated with paranoia (p < 0.001) (Figure 2A) and Empathic Concern was significantly negatively associated with overall emotion recognition ability (p = 0.002) (Figure 2B). Contrary to expectation, there was not a significant indirect path between empathy subscales, emotion recognition, and paranoia.

Given the relatively poor model fit of the a priori theorized model, we ran several post hoc sensitivity analyses to determine whether modeling relationships with specific emotions would better fit our data (Supplementary Figures 1–4). First, we examined the paths between Fantasy and paranoia through emotion recognition for each emotion separately (e.g., fear). When considering ability to recognize specific emotions, we found that Fantasy was positively associated with fear emotion recognition (p = 0.008) and paranoia (p < 0.001), but no significant indirect path (Figure 3A) emerged. Of note, this model was oversaturated (CFI = 1.00, RMSEA = 0.00, SRMR = 0.00) indicating that it did not fit the observed data. Next, we tested an exploratory model in which we alternated the order of the endogenous and exogenous variables – fear recognition (exogenous) predicting paranoia (endogenous) through Fantasy. This model was also oversaturated (CFI = 1.0, RMSEA = 0.00, SRMR = 0.00) and showed similar results: fear emotion recognition was positively associated with Fantasy (p = 0.003) and Fantasy was positively associated with paranoia (p < 0.001) (Figure 3B) with no significant indirect path. Please see Table 3 for all SEM parameters.

In this general population dataset, empathic fantasy was positively associated with paranoia, suggesting that individuals who experience more paranoia report a tendency toward empathically relating to fictional characters. Previous studies in schizophrenia that have investigated empathy using the IRI have reported varying levels of disruption in empathic fantasy (Rankin et al., 2005; Sucksmith et al., 2013; Bonfils et al., 2017), with some research demonstrating deficits (Fujiwara et al., 2008; Hooker et al., 2011) while others have shown greater or similar Fantasy scores in schizophrenia as compared to healthy controls (Sparks et al., 2010; McCormick et al., 2012; Matsumoto et al., 2015). Meta-analysis reported slightly reduced Fantasy scores in schizophrenia, with a small effect size (Hedge’s g = 0.19).

Although the IRI has been thoroughly investigated in patients with schizophrenia, there are fewer studies looking at the IRI and sub-clinical paranoia. In one study, Buck et al. (2017) found that distorted mind perception (perceiving that someone/something possesses a thinking, feeling mind) partially explained the association between paranoia and the perspective taking and empathic concern subscales but found no direct effects. In another study, Lee (2017) found that patients with paranoid personality disorder had diminished perspective taking and increased personal distress, but no differences in Fantasy. Interestingly, although Buck et al. (2017) did not specifically investigate the Fantasy subscale, they do report that paranoia is positively associated with a tendency to perceive mind in dead people, trees, robots and Superman. Mind perception of a character like Superman reflects the person’s sense that the fictional character has memories, agency, and ability to feel fear, pleasure and hunger. These findings in a larger sample may bear on our data, suggesting that those with greater tendency toward paranoia experience more empathy toward fictional characters.

Also relevant to our findings is a previous report of a significant relationship between Fantasy scores and delusion severity in individuals with schizophrenia (Sparks et al., 2010). Although Fantasy scores did not differ between patients and controls in this prior analysis, greater Fantasy was strongly related to more severe delusions. In addition, in a study looking at empathy and theory of mind in first-degree relatives of individuals with schizophrenia (i.e., those at elevated genetic risk), it was only the Fantasy subscale of the IRI (not theory of mind or other measures of empathy) that was associated with a history of subclinical delusional ideation, such that greater delusional ideation was related to greater Fantasy scores (Montag et al., 2012).

Our findings therefore add to a growing literature that connects delusional ideation and the Fantasy subscale of the IRI. The relationship between these two measures is somewhat confounding, in part because the Fantasy subscale is, itself, difficult to interpret (Nomura and Akai, 2012). The Fantasy scale measures the individual’s tendency to use their imagination in order to take the perspective of a fictional character. Unlike the other IRI subscales, Fantasy is stable across adolescence, suggesting it may function slightly differently than other facets of empathy, which change over time (Davis and Franzoi, 1991). In factor analysis, Fantasy loads with Perspective Taking on a factor of cognitive empathy (Shamay-Tsoory et al., 2007), indicating it may reflect more advanced capabilities similar to theory of mind. In fact, scores on the Fantasy scale have been previously related to verbal measures and intellectual ability (Mayer and Geher, 1996; Montag et al., 2012). Paranoia is a facet of delusional ideation focused on a belief that one is under threat from others (Freeman, 2016). Delusions, including paranoia, are a self-conscious experience that are thought to be born from an unstable boundary between the self, the world, and others (Sass and Parnas, 2003). While speculative, relationships between Fantasy and paranoia may be an unexpected glimpse into this boundary disturbance that can occur in individuals on the delusion spectrum (Feyaerts et al., 2021). Its relationship with paranoia could also be a byproduct of the safety that fictional characters can provide. More paranoid individuals may feel more connected with fictional characters, because they cannot cause them imminent harm and may themselves be persecuted.

Another contribution of the current study is our investigation into role that emotion recognition plays with empathy and paranoia. Both emotion recognition and empathy are critical for healthy social interactions, and impaired social functioning contributes to paranoia (Fiske and Taylor, 2013; Chen, 2014; Wickham et al., 2014). Emotion recognition is a more “basic” ability that supports interaction with the environment, underlying the more complex process of experiencing empathy (Duesenberg et al., 2016). Due to empathy’s dependence on recognizing facial emotion, we expected to find an association between the IRI subscales and emotion recognition. After post hoc analyses, the only subscales we found to be associated with emotion recognition was the Fantasy subscale and the Empathic Concern subscale. Previous studies linking empathy and emotion recognition have found that emotion recognition capabilities correlate with higher dispositional empathy (Davis and Kraus, 1997; Hall et al., 2009). Soto et al. (1998) found that people who are empathically accurate (better at rating how a stranger feels from moment to moment) are better at identifying positive and negative emotions. Papers that have looked specifically at the IRI and ER-40 measures generally found increased emotion recognition in relationship to higher empathy (Martin et al., 1996; Gery et al., 2009).

In our study, emotion recognition was associated with both Fantasy and Empathic Concern, however, there was not an indirect relationship between these facets of empathy and paranoia through emotion recognition. Interestingly, Fantasy was specifically associated with recognition of the emotion of fear, such that better fear recognition was related to greater empathic fantasy. Expressions of fear are distress cues that can drive feelings of empathy (Panksepp and Panksepp, 2013), and better fear recognition is related to more pro-social behavior (Marsh et al., 2007). Relationships with Fantasy and fear recognition are novel, as prior work linking fear recognition and empathy have looked only at Empathic Concern (Besel and Yuille, 2010). Unlike our findings, however, which revealed a negative relationship, prior work has demonstrated a positive relationship between Empathic Concern on the IRI and emotion recognition ability (Jiang et al., 2014). This finding was strongest for emotion recognition after a brief exposure (50 ms), which is much shorter than the current task parameters. This implies that sensitivity to the relationship between emotion recognition and Empathic Concern may depend on quick, automatic processing. Overall, the findings between empathy and emotion recognition in the current study are intriguing but require replication and further investigation.

The lack of a significant association between emotion recognition and paranoia is also somewhat surprising. Past research completed in the general population has shown that those with higher subclinical paranoia had lower overall emotion recognition ability (Combs and Penn, 2004; Klein et al., 2018), and misidentified neutral expressions for anger (Pinkham et al., 2011), although a recent study found that positive schizotypy was not associated with emotion recognition (Dawes et al., 2021). This relationship has also been studied extensively in patients with schizophrenia. In a study comparing schizophrenia participants and subclinical populations with varying levels of paranoia, schizophrenia participants had worse emotion recognition ability than those with low or moderate subclinical paranoia, but similar emotion recognition as subclinical participants with high paranoia (Combs et al., 2006). Lack of replication in the current sample may be due to our use of the PDI-21 to measure paranoia. Prior studies have used the Paranoia Scale – a 20-item self-report focused entirely on paranoia (Combs and Penn, 2004; Klein et al., 2018), allowing for more power to detect relationships within a subclinical population.

This investigation approached the relationship between empathy, emotion recognition and paranoia through a sophisticated statistic method that allowed us to test the relative contribution of different types of empathy on paranoia, as well as model our hypothesized pathways. However, this study was limited by relatively low power for a structural equation model, which may have affected the specificity and fit of each model (Anderson and Gerbing, 1984). Another limitation to this study is the nature of the Peters Delusion Inventory and Interpersonal Reactivity Index. Self-report questionnaires are often influenced by response biases like social desirability. As the Peters Delusion Inventory asks questions about thoughts that may be stigmatizing (e.g., Do you ever feel as if there is a conspiracy against you? Do you ever think people can communicate telepathically?), a participant may report lower levels of distress, preoccupation, and conviction. Furthermore, the IRI may be best interpreted as a participant’s thoughts about their empathic abilities and not their actual abilities. Self-reported cognitive empathy abilities have been found to be unrelated to actual empathic accuracy during a brief interaction with another person (Ickes et al., 2000; Zaki and Ochsner, 2009). A more interactive or dynamic task may be necessary to tap into the social cognitive domain of empathy (Haut et al., 2019). Finally, Batson et al. (2002) has questioned whether empathy can be measured validly by self-reports and the internal consistency of the fantasy subscale as a measure of empathy has been previously questioned (Baldner and McGinley, 2014) and therefore should be interpreted carefully. One main note is that the fantasy subscale, unlike the other three subscales, was created from two sources [Stotland’s (1969) Fantasy-Empathy scale and Davis (1983)]. Baron-Cohen and Wheelwright (2004) proposed that fantasy was not characterized by affective-cognitive dimensions of empathy and suggested that the scale assesses imagination, not empathy itself (Lawrence et al., 2004). In the schizophrenia literature, results of differences in fantasy are mixed (Horan et al., 2015; Bonfils et al., 2017). Therefore, while we are intrigued by the fairly consistent relationships between fantasy and delusional ideation across the literature, we acknowledge that its interpretation as a facet of empathy requires further validation.

This study was one of the first to identify relationships between a cognitive facet of empathy and paranoia, as well as fear recognition. The role of fear recognition in the relationship between Fantasy and paranoia should be explored further. If replicated, future studies should examine the directionality of these relationships, for instance testing whether increased Fantasy is present prior to paranoia onset. This would require longitudinal or temporal data that could truly capture directionality. This study provides future support for considering symptoms as a continuum and highlights the importance of studying the social-cognitive aspects of experiences like paranoia.