Paranoia and conspiracy: group cohesion increases harmful intent attribution in the Trust Game

Introduction

Although paranoia is the most common positive symptom of psychosis (Freeman, 2007), it is also present to varying degrees in the general population. Paranoia can range from mildly exaggerated concerns about how others see us, to frank and disabling paranoid delusions of conspiratorial harm (Freeman & Garety, 2014; Taylor, Freeman & Ronald, 2016; Elahi et al., 2017). Estimations suggest a third of the general population frequently experience paranoid thoughts (Freeman et al., 2005). Paranoid ideation follows an exponential distribution where increasingly severe thoughts are increasingly rare (Bebbington et al., 2013).

One well-established component of paranoia is an alteration to social threat perception. Patients with persecutory delusions have better memory for threat-related words (Bentall, Kaney & Bowen-jones, 1995) and actively paranoid patients diagnosed with schizophrenia over-perceive anger in neutral faces (Pinkham et al., 2011). Highly paranoid members of the general population over-estimate chances of future victimisation (Jack & Egan, 2016) and are more likely to perceive harmful intent in ambiguous social exchanges (Raihani & Bell, 2017a; Raihani & Bell, 2017b).

Paranoia also frequently involves problems with accurately judging the intentions of groups, rather than individuals (Raihani & Bell, 2018). In clinical studies, concerns about conspiracy are a well-established component of paranoia that have been documented from early in the history of psychiatry (Harper, 1994) and form part of current definitions (Oyebode, 2008). Here, conspiracy perception refers to concerns about being persecuted by a group of others who are coordinated in attempts to harm the individual, but who do not correspond to any collection of people with these aims—something Cameron (1959) conceptualised as the ‘paranoid pseudocommunity’. This is distinct from belief in public conspiracy theories more broadly, which describe conspiratorial explanations of important historical events that are not centred on the believer (Douglas, Sutton & Cichocka, 2017). Indeed, recent cross-cultural research highlights that that while paranoia and conspiracy thinking (belief in public conspiracy theories) are associated, they are divergent constructs (Imhoff & Lamberty, 2018).

Rather than viewing paranoia only as a symptom of a mental disorder, recent research has suggested that it might be more accurate to view paranoia across the full spectrum of severity as part of a normally-functioning human psychology that evolved in the context of detecting and avoiding social threat (Green & Phillips, 2004; Raihani & Bell, 2017a; Raihani & Bell, 2017b). Studies that have examined social threat processing as a broader psychological mechanism suggest it is likely to be supported by a specialised, evolved mechanism that is attuned to threats from both individuals and groups. The Coalitional Index Model (Boyer, Firat & van Leeuwen, 2015) proposes that individuals can judge cues of social threat and support and can integrate them to encode their own vulnerability in any given environment. Under this hypothesis, activation of the ‘coalitional safety index’ triggers a set of physiological, cognitive and behavioural responses attuned to the perceived degree of social threat. Threat inputs to this mechanism include coalitional identity, as supported by evidence that out-group cues increase anxiety and fear responses (e.g., Navarrete et al., 2009; Hart et al., 2000).

Such a psychological mechanism to detect social threat should be sensitive to the cohesiveness of rival groups. More cohesive groups are more able to act towards a common goal. Therefore, cohesive opponent groups should be perceived as more threatening than similar-sized but non-cohesive groups (Boyer, Firat & van Leeuwen, 2015). Indeed, group entitativity (its perceived unity) increases negative cognitive and behavioural responses towards the group (Campbell, 1958; Hamilton & Sherman, 1996). High-entitativity groups are also perceived as more morally suspicious (Newheiser, Sawaoka & Dovidio, 2012), conspiratorial (Grzesiak-Feldman & Suszek, 2008), more capable of negative retaliation (Dasgupta, Banaji & Abelson, 1999) and evoke increased negative stereotyping (Spencer-Rodgers, Hamilton & Sherman, 2007).

One clear implication of these studies is that individuals will exhibit increased threat responses when in the presence of cohesive compared to non-cohesive groups. However, despite the clear overlap with paranoia both conceptually and clinically, the effects of group cohesiveness on threat perception have never been tested in this domain. Research is needed in this area to further our understanding of social experience of paranoia, and how mechanisms of social cognition function in paranoia.

Various hypotheses regarding paranoia and group cohesion emerge. The first is that paranoia may involve an exaggerated reaction to group cohesion (a social over-sensitivity bias) so that highly paranoid people will increase their attributions of harmful intent to cohesive compared to non-cohesive groups, over and above the degree to which low-paranoia individuals do. Another possible pattern is that paranoia may involve a misperception of group cohesiveness such that all groups are perceived to be cohesive. In this case highly paranoid people will attribute equally-high levels of harmful intent to cohesive and non-cohesive groups alike, unlike low-paranoia individuals who will attribute greater levels of harmful intent to cohesive compared to non-cohesive groups. Finally, it could be that paranoia simply involves a higher baseline of harmful intent attribution—a general tendency to over-attribute harmful intent in social situations when compared to low-paranoia individuals—but show the same pattern of reactivity to group cohesiveness as low-paranoia individuals.

Here, we report a pre-registered study testing these predictions. We recruited a dual online and offline sample, including a large online sample covering the population distribution of paranoid ideation, and a lab-based panels of participants who formed cohesive and non-cohesive pairs to respond to decisions made by the online participants. Previous studies have demonstrated that large-scale game theory approaches are effective in capturing live paranoid attributions and testing how paranoid attributions are modified by experimentally-induced social threat (Raihani & Bell, 2017a; Raihani & Bell, 2017b; Saalfeld et al., 2018). For example, modified Dictator Games have revealed that individuals spanning the full paranoia spectrum rate unfair Dictators as intending more harm than fair Dictators, signifying that fairness is used as a threat cue irrespective of an individual’s level of paranoid ideation (Raihani & Bell, 2017a; Raihani & Bell, 2017b). Similarly, interacting with higher status or out-group partners triggers exaggerated attributions of harmful intent in otherwise ambiguous interactions (Saalfeld et al., 2018).

In the current study, participants played as proposers in an adapted Trust Game in one of two conditions, where they interacted (i) against a pair of cohesive responders or (ii) against a same-sized group of non-cohesive responders. Participants were given an initial sum of money ($0.50) and were asked to indicate what proportion they would like to send ($0.25 or $0.05) to a pair of responders. Participants kept any of the money they did not send to the responders. Participants were told that any amount they sent would be quadrupled by the experimenter, and that the responders could decide whether to be fair and return half of the increased amount to the participant or to be unfair and return nothing to the participant. Participants were randomly allocated to either fair or unfair responder pairs.

The cohesiveness of the responder pair was characterised by whether responders could communicate with one another or not. In the cohesive condition, responders had to agree on their decision, whereas in the non-cohesive condition responders came to their decision independently.

Participants chose how much to send to the responders and were then presented with the responders’ decision. After finding out how much the responders returned to them, participants rated the extent to which they believed the responders’ decisions to be driven by (i) self-interest and (ii) harmful intent, using two separate slider scales. The true motives underpinning responder decisions in this game are ambiguous and could plausibly reflect either self-interest (desire to keep more money) or harmful intent (desire to exploit the participant). Previous work indicates that paranoia positively predicts harmful intent attributions but not self-interest attributions in similarly ambiguous settings (Raihani & Bell, 2017a; Raihani & Bell, 2017b; Saalfeld et al., 2018).

Materials and Methods

This study was approved by the UCL Ethics board, under project 3720/001. Participation was voluntary and informed consent was obtained from all participants prior to their participation.

Results

The mean (±SEM) paranoia score was 50.7 ± 0.67, reflecting the combined scores of social reference (27.1 ± 0.35) and persecution (23.5 ± 0.36) subscales, respectively. Paranoia scores ranged from 32 to 158, with 54 participants (4.63%) being over the clinical mean of 101.9 reported in Green et al. (2008).

As predicted, participants attributed more harmful intent to cohesive compared to non-cohesive opponents (estimate: 0.39, CI [0.15–0.64]; Fig. 1, Table 1) but paranoia did not interact with opponent cohesiveness to exaggerate this effect (estimate: −0.06, CI [−0.35–0.24], Table 1). All predictions regarding the validity of the paradigm were supported. Participants attributed more harmful intent to unfair rather than fair opponents (estimate: −1.19, CI [−1.43–−0.95], Table 1) and more self-interest to unfair rather than fair opponents (estimate: −3.37, CI [−3.67–−3.06], Table 2). In addition, paranoia independently and positively predicted harmful intent attribution (estimate: 0.77, CI [0.55–1.00]; Table 1, Fig. 2) and there was no effect of paranoia (Fig. 2) or opponent cohesiveness on attributions of self-interest (Table 2, Fig. 2).

Discussion

We investigated the effects of pre-existing paranoia on harmful intent attribution when interacting with cohesive and non-cohesive groups. Participants played as proposers in an adapted Trust Game against a pair of cohesive or non-cohesive responders. Although the intentions of responders are ambiguous in this task, cohesive responder groups were perceived as more malevolent than non-cohesive responder groups, as measured by harmful intent attributions made by proposers. Paranoia was associated with a general tendency to make stronger harmful intent attributions in all situations but highly paranoid individuals showed the same increase in harmful intent attribution when interacting with cohesive groups as individuals with lower paranoia. This suggests that pre-existing paranoia reflects a generally heightened tendency for threat perception but an intact sensitivity to group cohesion. Therefore, there was no evidence of a qualitatively nor quantitatively different response to the social context that may reflect a dysregulation in perceiving group interactions.

This finding converges with studies that found no interaction between paranoia or psychosis and social threat on behavioural and physiological responses (Veling et al., 2016; Counotte et al., 2017; Saalfeld et al., 2018). Together, these studies suggest that paranoia is associated with a lower threshold for detecting social threat; but the scaling of response to changes in social threat is similar in all individuals, irrespective of where they lie on the paranoia spectrum. This goes against theories that suggest paranoia involves an increased reactivity to social threat (Yiend et al., 2017; Savulich et al., 2015; Pot-Kolder et al., 2017); and against a generalised insensitivity to social information (Fett et al., 2012; Fett et al., 2016; Gromann et al., 2013).

Our results support the hypothesis that group cohesiveness acts as a social threat cue, since paranoid attributions increased in response to greater group cohesiveness. This coincides with evidence that tightly-formed groups are seen as more legitimate subject of conspiracy theories (Grzesiak-Feldman & Suszek, 2008). In light of the Coalitional Index Model (Boyer, Firat & van Leeuwen, 2015), this would imply that group cohesiveness is an input into the Coalitional Safety Index. Changes in cohesiveness of the responder group affected proposers’ attributions of harmful intent but not self-interest: the responder group was perceived as no more self-interested in conditions of high compared to low cohesiveness, even though cohesive responders were perceived as more malevolent than non-cohesive responders. This can be viewed as a conceptual replication of data showing other social threat cues, group affiliation and social status, independently alter attributions of harmful intent attribution, but not self-interest (Saalfeld et al., 2018).

Despite capturing paranoid ideation scores common to patients with psychosis, it is unclear to what extent these results would characterise patients with paranoid delusions. The mean paranoia score in our sample was similar to that of the non-clinical sample in Green et al.’s (2008) original study, with a similar percentage of participants scoring above the clinical mean. However, we did not record whether any participant had a clinical diagnosis and assessing the presence of delusions is best completed with a structured clinical interview, rather than self-report measures. It would be surprising if in such a large sample no participants with delusions were present but we have also noted that severe presentations of delusional paranoia are likely to involve specific changes to social perception that are not present in non-delusional presentations.

The online nature of the study permitted a large sample size, as well as a more representative sample than typical samples from typical university undergraduate and community participant pools (Berinsky, Huber & Lenz, 2012). However, MTurk participants still show important demographic differences to the general population (Huff & Tingley, 2015) and in terms of mental health show higher levels of social anxiety although they are not more likely to report clinically relevant emotional dysregulation than the wider population (Shapiro, Chandler & Mueller, 2013). Furthermore, given all our participants were based in the US it is not clear how well these effects might generalise more widely.

Conclusions

Our results replicate and extend existing research in paranoia. Firstly, they suggest that perceptions of opponent group cohesion moderate live paranoid ideation, where more cohesive groups evoke greater perception of threat. Secondly, these results support evidence that those scoring high in paranoia in the general population have a greater tendency to perceive conspiracy, but these attributions of malevolent intent show intact scaling in response to changes in social threat.

Supplemental Information