A 6-year follow-up study in a community-based population: Is neighbourhood-level social capital associated with the risk of emergence and persistence of psychotic experiences and transition to psychotic disorder?

Background Social capital is thought to represent an environmental factor associated with the risk of psychotic disorder (PD). This study aims to investigate the association between neighbourhood-level social capital and clinical transitions within the spectrum of psychosis. Methods In total, 2175 participants, representative of a community-based population, were assessed twice (6 years apart) to determine their position within an extended psychosis spectrum: no symptoms, subclinical psychotic experiences (PE), clinical PE, PD. A variable representing change between baseline (T1) and follow-up (T2) assessment was constructed. Four dimensions of social capital (informal social control, social disorganisation, social cohesion and trust, cognitive social capital) were assessed at baseline in an independent sample, and the measures were aggregated to the neighbourhood level. Associations between the variable representing psychosis spectrum change from T1 to T2 and the social capital variables were investigated. Results Lower levels of neighbourhood-level social disorganisation, meaning higher levels of social capital, reduced the risk of clinical PE onset (OR 0.300; z = −2.75; p = 0.006), persistence of clinical PE (OR 0.314; z = −2.36; p = 0.018) and also the transition to PD (OR 0.136; z = −2.12; p = 0.034). The other social capital variables were not associated with changes from T1 to T2. Conclusions Neighbourhood-level social disorganisation may be associated with the risk of psychosis expression. Whilst replication of this finding is required, it may point to level of social disorganisation as a public health target moderating population psychosis risk.


Introduction
Social capital may represent an environmental factor influencing the geographical distribution of psychotic disorder (PD; Krabbendam & Van Os, 2005). The dynamic changes within the spectrum of psychosis expression, or extended psychosis phenotype may yield valuable implications. Here, we examine prospective associations between neighbourhood-level social capital and changes within the extended psychosis phenotype over time.
Social capital is generally used to describe the aspects of social networks, relations and trust either at the individual or area level (Whitley & McKenzie, 2005). It can be seen as an umbrella term including informal social control, social cohesion and trust, social disorganisation and cognitive social capital (Moore & Kawachi, 2017). Informal social control is the willingness of the group to enforce social norms for the common good (Kawachi & Berkman, 2015). Social cohesion and trust refers to the extent of connectedness and solidarity among individuals in a society (Kawachi & Berkman, 2015). Social disorganisation is defined as a disturbance in the patterns and mechanisms of human relations (Faris & Dunham, 1939). Cognitive social capital refers to people's perceptions of trust, support and reciprocity (Harpham, Grant, & Thomas, 2002). Higher social capital has been associated with better physical and mental health outcomes (Ehsan, Klaas, Bastianen, & Spini, 2019). In 1939, a higher schizophrenia prevalence was observed in urban neighbourhoods with a higher level of social disorganisation (Faris & Dunham, 1939). Afterwards, it was claimed that one-third of schizophrenia incidence was associated with environmental factors, and social capital has been hypothesised as a possible underlying environmental factor mediating the impact of urbanicity on the non-random geographical distribution of PD (Krabbendam & Van Os, 2005).
A recent review studying the association between social capital and psychosis reported conflicting results as a consequence of heterogeneous measures of social capital (Rotenberg, Anderson, & McKenzie, 2020). An earlier study did not find any association between social capital and schizophrenia incidence but did show that the quantity of inpatient service consumption was higher in neighbourhoods with a higher level of informal social control (Drukker, Krabbendam, Driessen, & van Os, 2006). On the other hand, two studies, including large cohorts, concluded that low levels of social capital were associated with a higher incidence of PD. Kirkbride et al. suggested that the association between social capital and schizophrenia was not linear but U-shaped; incidence rates being higher in neighbourhoods with both low and high levels of social capital but not median levels of social capital . In a non-clinical population, paranoia was associated with the measures of social capital, including decreased trust in others (Freeman et al., 2011). In that study, data regarding symptoms of psychosis and social capital were collected from the same population, raising the possibility of contamination of the social capital data with the individual perceptions of the study population (Drukker & van Os, 2003). In conclusion, there is some evidence that the measures of social capital, including disorganisation and control, are associated with psychosis (Rotenberg et al., 2020).
The present paper is a part of The Izmir Mental Health Survey for Gene-Environment in Psychoses (TürkSch). The first wave of the TürkSch study confirmed the continuity of PE and PD and revealed that 25% of the general population has experiences that phenotypically are in the extended psychosis spectrum (Binbay et al., 2012c). Baseline data showed that as neighbourhood-level social capital (informal social control) and social deprivation increased, the expression of psychosis moved towards the clinical/syndromal end of the spectrum. Furthermore, the association between familial liability of severe mental illness and expression of psychosis was stronger in neighbourhoods with higher levels of deprivation and social control (Binbay et al., 2012b). This finding contradicts the results of studies from Maastricht and London, where the expression of psychosis was more at the clinical/syndromal end in neighbourhoods with lower levels of social capital (Drukker et al., 2006;Kirkbride et al., 2007). The contradiction may be attributed to the higher mean neighbourhood-level social control in Izmir (Dominguez et al., 2010;Lofors & Sundquist, 2007). The finding thus suggests that the effect of social capital on the expression of psychosis is not linear.
As summarised above, the current literature provides conflicting results regarding the association between social capital and the expression of psychosis in cross-sectional designs. As far as we know, this is the first longitudinal study prospectively investigating the role of social capital in occasioning transitions across the severity spectrum of the extended psychosis phenotype. This paper aims to assess whether neighbourhood-level social capital is associated with the risk of psychosis transitions across a spectrum of severity, ranging from no PE, subclinical PE, clinical PE to PD, from baseline (T1) to 6-year follow-up (T2) in a general population sample. More specifically, we assessed whether: (1) baseline characteristics were associated with any of the dynamic transitions from T1 to T2 and specifically with emergence, persistence and transition.
(2) social capital is associated with the risk of any transition from T1 to T2. (3) social capital is associated with the risk of three predefined transitions: emergence of PE, persistence of PE, transition to PD.

Study design and data collection
TürkSch is a prospective, longitudinal study designed to screen and follow up the extended psychosis phenotype and individual, family and neighbourhood-level variables (Binbay et al., 2011). Data were collected in Izmir, which is the third most urbanised area in Turkey (Türkiye İstatistik Kurumu, 2008). TürkSch consisted of a baseline (T1: 2007-2009) and a follow-up assessment (T2: 2013-2015), approximately 6 years apart (mean follow-up time: 6.11 ± 0.94 years). Each assessment consisted of several stages of data collection (T1: stages 1-3; T2: stages 4-5). The present paper used data collected at stages 1, 2 and 4. At stage 1, a baseline sample was randomly selected representing the wider Izmir metropolitan area using a multistage sampling procedure, stratified by urbanicity covering 11 districts and 302 neighbourhoods. The first stage was a cross-sectional study, including assessment of the extended psychosis spectrum and individuallevel sociodemographic variables (T1, stage 1, n = 4011). For stage 2, a second address from the same neighbourhood was contacted for each participant of stage 1. Thus, a different sample from the same area provided neighbourhood-level social capital variables (T1, stage 2, n = 5819). Finally, 6 years after baseline, addresses of all T1 participants were revisited in person. The extended psychosis spectrum psychopathology was collected as well as changes in sociodemographic features, alcohol and cannabis use, and threatening life events from all stage 1 participants who could be reached (T2, stage 4, n = 2185). Details of stages 1-5 have been described elsewhere (Binbay et al., 2011(Binbay et al., , 2012aKırlı et al., 2019). The study was approved by the ethics committee of Ege University; written informed consent was obtained from all participants.

Assessment of the sociodemographic features
The T1 assessment included a sociodemographic questionnaire. At T2, the same questionnaire was used to determine changes over time. Level of education was dichotomised into ⩽8 and >8 years; ethnicity into Turkish and non-Turkish; marital status into married and non-married; employment status into unemployed and not unemployed; alcohol and cannabis use into user and non-user; migration between ages 6 and 15 years into migrated v. not migrated; childhood adversity and traumatic events into none v. at-least once. Childhood adverse life events were the death of any parent, divorce of parents and separation from parents for at least 3 months between the age of 0 and 15 years. Traumatic events were war experience, life-threatening accident, fire, natural disasters, witnessing someone being badly injured or killed, rape, sexual molestation and being physically attacked or assaulted. Socioeconomic status was based on the participant's profession, recoded into four ordinal categories: professional and non-manual high employees; non-manual low employees and skilled workers and technicians; owners of small businesses; manual workers. In order to estimate socioeconomic status at birth, parental socioeconomic status was coded similarly, using the highest position of the participant's mother or father. The data included three urbanicity variables: urbanicity at birth, urbanicity at age 6-15 years and current urbanicity. Urbanicity at birth and in childhood was based on population density measures in the 1990 census, and the classification depended on the level of organised features of streets and buildings. If the participant lived at more than one address, the most urban one was included. Current urbanicity was coded into four categories: rural, low-urban, moderately-urban, highly-urban area. Family history of mental disorder was recoded into four categories (none, other/unknown, common mental disorder, severe mental disorder). Common mental disorder in the family included any diagnosis of depression, anxiety, conversion or somatisation in the absence of severe mental disorder among parents or siblings of the participant. Severe mental disorder in the family was coded positive if any of the participant's first-degree relatives had been diagnosed with PD or bipolar disorder, had died because of suicide or had been admitted to a psychiatric inpatient unit.

Assessment of the social capital of neighbourhoods
Four dimensions of neighbourhood-level social capital were assessed at stage 2: informal social control, social cohesion and trust, social disorganisation, and cognitive social capital. The informal social control scale included eight questions measuring the willingness to intervene in hypothetical neighbourhood-threatening situations such as children misbehaving, using a five-point Likert scale ranging from 'strongly disagree' to 'strongly agree'. The questions were derived from the Sampson collective efficacy scale and adapted for use in the Turkish population (Sampson, Raudenbush, & Earls, 1997). The social cohesion and trust scale measured bonds and trust among neighbourhood residents (Araya et al., 2006;Sampson et al., 1997). The social disorganisation scale consisted of eight questions rating the frequency of certain scenarios occurring in the participant's neighbourhood, such as the presence of graffiti, vandalism, burglary and racist attacks. Items were assessed using a four-point Likert scale ranging from 'not at all common' to 'very common', and the scale was derived from the McCulloch instrument (Buckner, 1988;McCulloch, 2003). The cognitive social capital scale included three questions measuring perceptions of support, reciprocity and sharing among the residents of the neighbourhood (Harpham et al., 2002). For each social capital dimension, sum scores were (negative items were reversed) divided by the number of items and aggregated to the neighbourhood level. Lower scores on the informal social control variable (ISC-variable) indicated higher levels of informal social control, meaning higher social capital. Similarly, lower scores on the social cohesion and trust variable (SCT-variable) indicated higher levels of social cohesion and trust, meaning higher social capital. On the other hand, lower scores on the social disorganisation variable (SocD-variable) indicated higher levels of social disorganisation, meaning lower social capital. Similarly, lower scores on the cognitive social capital variable (Cogn-variable) indicated lower levels of cognitive social capital, meaning lower social capital (Binbay et al., 2012b). All four social capital variables were significantly correlated with each other. The correlation coefficients varied; ISC-variable, SCT-variable and Cogn-variable were moderately to strongly correlated with each other, whereas the SocD-variable was weakly correlated with the others ( Table 1).

Assessment of the extended psychosis spectrum
The procedure used to identify PE and PD has been described in detail elsewhere (Binbay et al., 2011). The assessed time frame was lifetime at T1, whereas it was the last 6 years at T2. Hallucinations, delusions, and associated impairment were assessed using the relevant sections of the Turkish version of the Composite International Diagnostic Interview (CIDI) 2.1 (Kılıç & Göğüş, 1997). The CIDI is a fully structured interview that can be used by both clinicians and trained interviewers . It was designed by the WHO to be implemented in epidemiological studies of mental disorders, and it provides information about frequency, duration and severity of symptoms, psychosocial impairment and help-seeking (Robins et al., 1988). It enables diagnosing various mental disorders in accordance with the criteria of the ICD-10 (World Health Organisation, 1992) and DSM-IV (American Psychiatric Association, 1994). CIDI has excellent inter-rater reliability in almost all sections, with κ values ranging from 0.67 to 0.97 (Wittchen et al., 1991). The κ value for agreement between clinicians for delusions and hallucinations was 0.85 and 0.87, respectively. Sensitivity of the CIDI was higher than its specificity for both delusions (0.93 v. 0.55) and hallucinations (0.86 v. 0.50) (Cooper, Peters, & Andrews, 1998).
Additionally, if the participant had ever been diagnosed with a mental disorder or been treated for a mental health problem, the health professional involved in diagnosis or treatment was contacted to gather relevant information when needed. If a participant had never been diagnosed with a PD but endorsed at least one CIDI psychotic symptom associated with help-seeking or occurring with a frequency of at least once per week, the participant was invited for a clinical evaluation. When the participant did not show up at the hospital visit, clinical interviews were conducted by the psychiatrist either at the participant's residence or via telephone. If clinical reappraisal was insufficient to get a final diagnosis, two further steps were applied with the participant's written consent. First, a family member was called to ask further questions on the illness and treatment of the probable case. Second, the patient register databases of the two university hospitals (Ege and Dokuz Eylül University Hospitals) in Izmir were checked. A best-estimate diagnosis of the CIDI interview was made in a collective evaluation according to the Structured Clinical Interview for DSM-IV Axis I Disorders (SCID-I) (First, 1997).
Finally, an extended psychosis phenotype variable was constructed based on the relevant CIDI items, clinical registries and the SCID-I results. The variable consisted of four categories:

Dependent variable
The dependent variable was constructed using the extended psychosis phenotype variable of all the participants who were interviewed at both T1 and T2 and who were residing in neighbourhoods with social capital data (N = 2175). Figure 1 shows all possible combinations of values at T1 and T2. For the purpose of the analyses, three groups were defined as follows: Emergent group: Those who had no PE at T1 but had subclinical (n = 119) or clinical (n = 84) PE at T2, as well as those who had subclinical PE at T1 and clinical PE at T2 (n = 39), yielding a total of 242.
Persistent group: Those who had either subclinical PE at both T1 and T2 (n = 55) or clinical PE at T1 and T2 (n = 64), yielding a total of 119.
Transition group: Those who had no PE (n = 10) or some subclinical (n = 5) or clinical (n = 13) PE at T1 and PD at T2, yielding a total of 28.

Statistical analyses
All analyses were performed using Stata, version 13.1 (StataCorp, 2013). In the present data, social capital was a characteristic at the neighbourhood level. Thus, data had a multilevel structure with subjects clustered in neighbourhoods. All statistical analyses were, therefore, performed with the equivalent multilevel format of the specific command, thus taking into account the hierarchical structure of the data.
In order to investigate whether neighbourhood-level social capital is associated with dynamic transitions within different levels of the extended psychosis phenotype over the follow-up period (T1 to T2), a multinomial logistic regression analysis was performed, including all 2175 cases. The first model included age, gender and parental socioeconomic status. Unfortunately, the multinomial analysis did not reach convergence and this forced us to remove parental socioeconomic status. The four neighbourhood-level social capital variables were added in separate models to avoid collinearity. Participants without expression of psychosis at any level at both T1 and T2 were used as the reference group. Three groups consisting of ≤5 participants were combined with the group above, resulting in 11 groups to be compared (labelled M1-11) with the reference group (Fig. 1).
Then, emergence, persistence and transition outcomes were analysed separately: The baseline sociodemographic characteristics of the three outcome groups were compared using χ 2 test. Mixed-effects logistic regression analysis was performed with neighbourhood-level social capital as the macro-level independent variable and emergence, persistence and transition as the three dependent variables. First, a model was analysed, including gender, age categories and parental socioeconomic status (model 1). Following this, each of the four neighbourhood-level social capital variables was added to model 1 in separate models.
Additional analyses were performed comparing the social capital data of the participants who moved to another neighbourhood during the 6-year follow-up with the ones who did not move in order to examine if this is also relevant.

Any transitions over time within the extended psychosis phenotype
Multinomial logistic regression analysis revealed that women recovered more frequently from clinical PE (M2; Table 2). Older people were less likely to have emergent subclinical PE (M3). However, if they did, they were less likely to recover (M1). Having persistent clinical PE was less likely for older people (M8), who were also less likely to transition to PD (M10). Participants living in neighbourhoods with higher scores on the SocD-variable, indicating lower levels of social disorganisation and thus higher social capital, were less likely to have emergent clinical PE (M6). Even if they did, their clinical PE were less likely to persist (M8). They were also less likely to transition to PD (M10). The other social capital variables were not associated with the transitions within the extended psychosis spectrum variable.

Emergence, persistence and transition
Baseline sociodemographic characteristics and the level of social capital of the three outcome groups (emergence, persistence and transition) are summarised in Table 3. Lower levels of education and cannabis use were associated with all outcome groups. Family history of mental disorder was associated with persistence and transition. Lower age, not being married and being unemployed were associated with both emergence and transition. Current higher urbanicity, lower socioeconomic status, the presence of traumatic events and lower scores on the SocD-variable (meaning lower social capital) were associated with emergence.
Mixed-effects logistic regression analysis revealed that participants who were living in neighbourhoods with lower scores on the SocD-variable were more likely to be in the emergent group compared with the non-emergent group (Table 4). The other social capital variables were not associated with any of the three outcome groups.

Participants who moved to a different neighbourhood during the follow-up
Five hundred forty-five participants moved to a different neighbourhood between T1 and T2. Baseline social capital levels of those who moved were not significantly different from those who did not move. Five hundred and twenty of those who had moved had T2 social capital data (ISC-variable and SCT-variable). At T2, these 520 participants were living in a neighbourhood with significantly higher scores on the ISC-variable (meaning lower social capital) than the neighbourhood they were living in at the T1 assessment (2.28 ± 0.81; 2.06 ± 0.34; t = 5.61; p < 0.001). Mean scores on SCT-variable were not significantly different.

Discussion
Conducted in a large community-based sample representative of the general population in a highly urbanised area of Turkey, TürkSch yielded comprehensive, high-quality data with a primary focus on the extended psychosis spectrum. In this longitudinal epidemiological survey of 2175 participants, we investigated whether neighbourhood social capital is associated with the risk of transitions within the extended psychosis spectrum over a 6-year follow-up period. The main finding of the primary analysis was that lower levels of social disorganisation decreased the risk of emergence and persistence of clinical PE and also the transition to PD. The finding that lower levels of social disorganisation decreased the risk of emergence was confirmed in our secondary analysis. The results, therefore, suggest that social disorganisation may be associated with transition towards the more severe end of the extended psychosis phenotype.

Multinomial v. mixed-effects logistic regression analysis results
The multinomial regression showed significant associations between social disorganisation and emergent PE, persistent PE and transition to PD, whereas the mixed-effects regression

3978
Ceylan Ergül et al.  showed a significant association between social disorganisation and emergent PE only. In the secondary analyses, the associations between the social disorganisation and the persistence of PE and transition to PD were not significant but in the same direction. The non-significance of the associations with persistence and transition in the secondary analysis may be related to the fact that multiple categories of the outcome measure were combined.

Neighbourhood-level social capital
The results of the multinomial regression analyses showed that participants living in neighbourhoods with lower levels of social disorganisation were less likely to have emergent or persistent clinical PE and less likely to transition to PD. Even though the associations were in the same direction for all 11 groups, only three of them were statistically significant. Therefore, the overall association may not be very strong. The results of the mixed-effects regression analyses showed that the SocD-variable was only significantly associated with the emergence of PE. There are no other longitudinal studies investigating social capital and the extended psychosis spectrum. Therefore, we could not compare our results directly. Within cross-sectional studies, there were only two that used social disorganisation as a measure of social capital (Binbay et al., 2012b;Kirkbride et al., 2008). One of them analysed the baseline data of TürkSch, and the authors showed that in neighbourhoods with more social disorganisation, respondents scored higher on the psychosis spectrum variable. They also reported that the association between the psychosis spectrum and severe mental illness in the family was stronger in disorganised neighbourhoods. However, the effect of social disorganisation was reducible to the other neighbourhood variables (Binbay et al., 2012b). Another study showed that social disorganisation was associated with schizophrenia incidence, but the association was not significant after considering social cohesion and trust . Briefly, our findings are similar to those of previous studies, but contrary to their results, the current analyses revealed social disorganisation as the only social capital dimension significantly associated with the psychosis spectrum.
When we compare our results with those of other studies measuring various aspects of social capital, our finding is broadly consistent with several previous studies showing the protective effects of social capital in psychotic symptoms and disorders (Freeman et al., 2011;Kirkbride et al., 2007;Lofors & Sundquist, 2007;O'Donoghue et al., 2016). Kirkbride et al.'s finding that there is a U-shaped association between social capital and schizophrenia incidence  is different from ours. A possible explanation may be the difference between the two measures of social capital. When analysing social cohesion and trust and schizophrenia, Kirkbride et al. showed a significant association, and we did not. Social cohesion and trust is a dimension of social capital, of which both lower and higher levels may increase the risk of psychosis expression. On the other hand, social disorganisation can be expected to have a one-way effect because the problems defined in the scale are more concrete and less open to interpretation. The causal mechanisms linking social disorganisation and psychosis expression are unclear. It is possible that neighbourhoods with higher levels of social disorganisation may facilitate or enhance factors related with psychosis such as immigration, traumatic events, cannabis use and air pollution (March et al., 2008). Table 3.   Of all four social capital variables, we found significant results for only the SocD-variable; however, we did not find any significant results for the other three social capital variables. There may be several reasons for this pattern of results. First, the ISC-variable, SCT-variable and Cogn-variable were moderately to strongly correlated with each other, whereas the SocD-variable was not. An explanation for this might be that the SocD-variable represents a different dimension of social capital. The ISC-variable, SCT-variable and Cogn-variable may represent a more cognitive aspect of social capital that is called 'collective efficacy' (Sampson, 1997), whereas the SocD-variable may represent more structural aspects such as concentrated disadvantage and population density (McCulloch, 2003). Second, various aspects of social capital may be more important in different contexts. In Izmir, social disorganisation may represent a more robust effect of social capital on psychosis (Saraçog  lu & Demirtaş-Milz, 2014) given very large differences between neighbourhoods in concentrated disadvantage and population density (Sönmez, 2007;Metropolitan Municipality of İzmir, 2007). Finally, social disorganisation may more powerfully represent the social contextual change in Turkey between T1 and T2 because of the socioeconomic impact of the 2008 financial crisis (2010-2014) (Aytaç, Rankin, & İbikoğlu, 2015).

Demographic and background characteristics
In general, the findings are in line with the literature (Linscott & Van Os, 2013;Van Os & Reininghaus, 2016). However, it is noteworthy that we did not find ethnicity to be a significant risk factor for psychosis. A plausible explanation for this might be that the risk factor is not being a minority per se but rather occupying a minority position in society (Van Os, Kenis, & Rutten, 2010). The minority groups investigated in our study do not stand out in relation to the general population in terms of easily distinguishable traits (e.g. skin colour, religious practices).

Methodological issues
Our findings should be considered in light of several limitations. First of all, a limitation is the possibility of selection bias caused by attrition over time. However, the attrition analysis revealed very small effect sizes; details have been published elsewhere (Kırlı et al., 2019). Second, around a quarter of the participants had moved to neighbourhoods with lower levels of social capital between baseline and follow-up. Since it is not known how long it takes for environmental factors to cause changes in mental health, we chose to use the baseline social capital data in our analyses. In addition, the follow-up social capital data were collected from the participants themselves and thus, the follow-up social capital data could be biased by the mental health status of the participants. Since we found no significant difference between the baseline neighbourhood-level social capital data of those who moved and those who did not move, it is unlikely that the mobility of the participants might have affected our results. Third, psychotic patients may tend to neglect or underreport their symptoms. In order to minimise this bias, additional information was gathered through clinical interviews and health records. Fourth, we defined the extended psychosis spectrum categories based on the presence of positive symptoms and the clinical impairment they cause. This definition is arbitrary, and the spectrum could be constructed in different ways. However, the sensitivity of CIDI is high for delusions and hallucinations, and this increased the clinical validity of our spectrum. As a result, our results cannot be generalised to negative or disorganised dimensions of psychosis. Finally, some of the questions used in the social disorganisation scale rely on subjective moral values, such that the results of this investigation can not be generalised to every culture. This is the first longitudinal community-based study investigating whether social capital is associated with the risk of transitions within the extended psychosis spectrum over a 6-year follow-up period. The strengths of our study are having highquality data of a large number of participants at two different time points and the data being collected by house visits by trained interviewers. The longitudinal design allowed us to create categories based on dynamic transitions within the extended psychosis spectrum, and this approach is more useful than working with categories based on cross-sectional data. Working with a general population sample was advantageous because we were more likely to include individuals who might not be captured in a clinical sample. The fact that data were collected randomly throughout a geographical area prevented help-seeking bias, thus revealing the spectrum, including subclinical manifestations. Another strength is that the neighbourhood-level variables were independently collected from a different sample. In addition to this, social capital was measured directly, whereas, in many other studies, proxy measurements such as voter turnout were implemented.

Conclusion
In conclusion, we found that lower levels of social disorganisation decreased the risk of emergence and persistence of clinical PE and also the transition to PD. The evidence about emergence was particularly robust. Our findings implicate that one dimension of social capital may reduce the risk of psychosis expression. Whilst replication of this finding is required, it may point to targeting the levels of social disorganisation as a public health measure associated with population psychosis risk.
Financial support. This work is part of the TürkSch project, which was funded by the Scientific and Technological Council of Turkey 1001 programme (Project numbers: 107S053 and 112S476).
Conflict of interest. None.