Psychiatric morbidity and social capital in rural communities of the Greek North Aegean islands
Abstract
Which facets of social capital affect mental health in rural settings? This study explores the association between different aspects of social capital and psychiatric morbidity in rural communities of the Greek North Aegean islands. A large number of individual and community characteristics that may influence psychiatric morbidity are concurrently examined in multilevel models to account for the clustering of individuals within rural settings. The current findings indicate that psychiatric morbidity is, to a large extent, clustered within rural communities. Individuals' perceived divisions in the community, i.e., political party preference, landholdings, low social support networks, and lack of perceived solidarity, are associated with psychiatric morbidity according to theoretical expectation. At the community level, this risk is lower in villages with over 250 residents, where there are youth clubs or a common threat, for instance, property crime. © 2010 Wiley Periodicals, Inc.
INTRODUCTION
Over the last decade there has been a daunting research interest in the investigation of the effects of social capital on health, especially, in the fields of socioepidemiology and public health (Szreter & Woolcock, 2004). This interest signifies an epistemological move from the study of sociodemographic variables to the study of social contextual factors as determinants of health. The concept of social capital arguably provides both a theoretical and a methodological framework for capturing a contextual perspective in this line of investigation (Lochner, Kawachi, & Kennedy, 1999). Although several definitions of social capital exist, most seem to agree that social capital is a multidimensional concept, which comprises norms, relationships, and networks that facilitate cooperation and collective action (Woolcock & Narayan, 2000).
However, the research into the association between social capital and mental health is limited. De Silva (2006) conducted a systematic review of 28 quantitative studies that examined the association between social capital and common mental disorders (CMDs) by placing emphasis on methodology. De Silva's review reveals a diverse collection of studies with varying ways of conceptualizing social capital, levels of measurement (individual or ecological), mental health outcomes, and methodology, including study design, sample size, setting (urban or rural), and statistical techniques. She noted that despite the growing sophistication of data analysis through the use of multilevel modeling, the relationship between social capital and mental health is rather complex and “varies by setting, aspect of social capital, and mental health outcome” (De Silva, 2006, p. 54).
Evidence by McKenzie and Harpham (2006) indicates that different facets of social capital have diverse effects on mental health depending on the socioeconomic population group examined. They concluded that although in some circumstances aspects of social capital may impact mental health, the interplay between wider social problems, such as poverty, gender inequality, unemployment, sociopolitical deprivation, and individual risk factors (e.g., school drop out) are more powerful predictors of negative mental health. Research in social capital suggests that vertical social capital is likely to affect mental health through supporting the linkages of disempowered population groups with the wider structures of resources irrespectively of contexts examined. The importance of bridging social capital and individual factors, such as relative deprivation, has also been evidenced by Whitley and Prince (2005), who reported high levels of CMD's in an urban community in London despite the presence of rich neighbourhood trust and social activity.
In what ways do social capital and mental health interact? Explicit hypotheses about mechanisms linking social capital to mental health have not yet been developed systematically. The long tradition of theoretical and empirical work on the relationship between social ties, social integration, and health conducted throughout the 1970s and 1980s (Cohen & Wills, 1985) may offer insights about the ways that social capital influences mental health. Berkman, Glass, Brissette, and Seeman (2000) proposed a conceptual model that links social network to health by integrating macro-level phenomena, such as wider social and cultural forces, with micro-level psychobiological processes. They argued that social networks function as mediating structures between macro socio-structural conditions and micro-scale forms of behaviours. Social ties and networks are expected to provide opportunities for social support, social influence, social engagement, intimate contact, and access to resources. The mechanisms by which a range of resources impact health include: (a) diffusion of norms that relate to health behaviours, (b) psychological processes including self-efficacy, self-worth, and security, and (c) neurobiological states. The current study is partly motivated by Berkman and colleagues' theoretical proposition of the ways that social capital and mental health may interplay.
This article considers the relationship between perceptions of community life, measures of social capital, and individual mental health in rural communities of the North Aegean Sea, Greece. The region includes three prefectures, Lesvos, Samos, and Chios, and comprises nine small-sized, medium-sized, and large-sized islands, spread in the northeastern part of the Aegean Sea. The choice of rural settings is important for two reasons: First, little research has been directed toward examining the associations between social capital and mental health in such contexts (De Silva, 2006), and, second, a critique to the notion of social capital includes issues of cultural dissonance by rural communities (Forbes & Wainwright, 2001). Therefore, this research provides an opportunity to examine the link between social capital and mental health within sociocultural environments other than urban Anglo-Saxon settings.
Mental health is measured via the revised Clinical Interview Schedule (CIS-R), which is a structured interview about the prevalence of psychiatric disorders (Lewis, Pelosi, Araya, & Dunn, 1992). Social capital measures draw on Putnam's conceptualization (1993) as well as on the model proposed by Lochner et al. (1999). Putnam defines social capital as civic participation and norms of trust and reciprocity. Along the same lines, Lochner et al. (1999) identify cohesion, collective efficacy, psychological sense of community, and community competence as key dimensions of social capital.
-
Which facets of social capital, if any, affect psychiatric morbidity in rural settings, taking into account other individual and community attributes and the clustering of individuals within communities?
To this end, empirical investigation via a multilevel logit model of a unique data set from Greece is employed. The statistical model is appropriate for examining binary health outcomes for nested units of analysis. (For an overview of the uses of multilevel models in health studies, see, for instance, McKeehan, 2000.) A description of the data set and the research instruments employed in this analysis follow. Thereafter, the multilevel logit model is formally given and preliminary associations among psychiatric morbidity and various constructs of social capital are discussed. The main results of this study are presented in the sixth (Discussion) section, while a discussion of the findings in light of the previous literature and how they may inform theory and future research concludes the article.
DATA
The data for this work come from a large-scale, cross-sectional survey on mental health, experiences, and perceptions of community social life (Zissi, Tseloni, Skapinakis, Savvidou, & Chiou, 2010). Community representatives and a randomly selected sample of residents in rural communities of the North Aegean islands participated in the survey, which comprised four modules: two qualitative and two sets of structured interviews. The current work employs quantitative data from the modules of structured interviews, i.e., the Household Questionnaire and the Community Profile.
The Household Questionnaire module was given to representatives of randomly selected households via stratified sampling as follows. First, all small rural communities of the region, whereby the population is under 2,000 inhabitants and at least 33% are farmers, were selected and totalled 89 such areas. Second, a random sample of households from each rural community was selected, with constant selection probability across communities from two mutually exclusive registers, namely, the electoral register and the register of residents without electoral rights. Thus, the total number of selected households is 428. The household representative was, finally, selected on the basis of the adult household member with birth date nearest to the interview date, according to standard survey practice (Hales, Henderson, Collins, & Becher, 2000; Kalton, 1983). Household representatives were face-to-face interviewed by trained male and female social researchers. The fieldwork was undertaken in the summer of 2004, with average interviews lasting 45 minutes. This module includes questions on mental health, social capital, and a wealth of demographic and socioeconomic individual and household attributes, such as age, gender, occupation, schooling, household structure, and affluence, which offer the study's level 1 or individual level covariates.
Additional face-to-face structured interviews with one key informant from each community (n=89)—most often a community councillor, priest, or teacher—were conducted for gaining further insights of the community context in which the participants live. These interviews are based on the World Bank Community Profile (Grootaert & van Bastelaer, 2002), which surveys a wide range of features of local environments (see the Community Profile subsection) and provides the level 2 or community level covariates of this study in clear distinction from the Household Questionnaire module variables, mentioned in the previous paragraph. Thus, individual and community information and the respective data are independent from a statistical viewpoint.
The available data set entails a natural hierarchy of individuals nested within communities with variables available at each level for investigating the association between social capital and mental health. In particular, the study combines individuals' mental health, social capital, demographic and socioeconomic characteristics with the communities' sociodemographic profile, infrastructure, resources, collective organizations, and mobilization.
RESEARCH INSTRUMENTS
Mental Health
Mental health is measured via the CIS-R, a fully structured psychiatric interview, designed to be used by trained lay interviewers (Lewis et al., 1992). The CIS-R was the main instrument used in the national psychiatric morbidity surveys in the United Kingdom (Jenkins et al., 1997) and has been used in several other similar surveys around the world (Araya, Rojas, Fritsch, Acuña, & Lewis, 2001). The CIS-R assesses the presence and severity of 14 different common psychological symptoms: psychosomatic symptoms, fatigue, concentration/forgetfulness, sleep problems, irritability, worry about physical health, depression, depressive ideas, worry, anxiety, phobias, panic, compulsions, and obsessions. Two screening questions in each section are as follows: one inquires about the presence of the symptom during the past month, and the other is a more detailed assessment of the presence, frequency, duration, and severity of the symptom during the past 7 days. Each symptom section is scored from 0 to 4 (except depressive ideas, from 0 to 5), and a score of two or more denotes a clinically significant symptom (Lewis et al., 1992). Additional questions enable the application of the ICD-10 (International Classification of Diseases 10th revision) research diagnostic criteria, using specially developed computerized algorithms. In addition, the distribution of total CIS-R score gives an indication of the severity of symptoms in a dimensional way.
The Greek version of the CIS-R was translated and back-translated, using the procedure recommended by the World Health Organization (http://www.who.int/substance_abuse/research_tools/translation/en/index.html). Male and female interviewers were trained in the use of the CIS-R. The vast majority of respondents reported low or no psychiatric morbidity, rendering the respective distribution highly skewed. A score of 12 (the current sample's upper quartile) or higher was used in this study to denote clinically significant psychiatric morbidity, which was observed in 14.2% of respondents.
Social Capital
The structured interviews to the selected household members and each community representative focus on capturing cognitive and structural social capital dimensions. Structural social capital examines aspects of facilities, services, and organizations, which have been identified as characteristics crucial for the formation and development of social support (Cattell, 2001). The cognitive dimension includes both attitudinal and behavioural items. The Social Capital subsection describes the study's social capital measures at the individual level, while the Community Profile one follows the Demographic and Socioeconomic Characteristics subsection.
Four key aspects of social capital with regards to individuals are examined: (a) social engagement, social networks, and support; (b) collective efficacy; (c) trust and social cohesion; and (d) sense of community. The majority of questions were sourced from the Social Capital Assessment Tool, developed by the World Bank (Grootaert & van Bastelaer, 2002) and adapted by the research team, while additional sources are explicitly mentioned. All social capital measures except the number of perceived differences and friends are dichotomous in this study.
-
The number of close friends that the respondent can rely on or confide in. This is a three-category nominal variable: none (34.9%), one to three (49.5%), and four or more (15.6%) friends.
-
The number of friends to borrow money from, indicating none (34.9%), one to three (44.8%), and four or more (20.3%) friends.
-
Mutual aid or perceived solidarity that refers to helping the community in case of an emergency, i.e., whether the majority of the rural community's residents would join forces to address a common disaster (46.5%)
-
Collective mobilisation or nonformal social participation, i.e., whether the majority would be involved in organizing a festival or fair in the village (45.0%).
-
Willingness to invest both money and time (formal social participation) for a public good, such as the construction of a playground, in the village (61.9%).
-
Individual perceived differences in political party preference (45.3%), landholdings (24.3%) and mentality (21.7%).
-
The total number of perceived differences, which is an aggregate count adding gender, nationality, inter-generational, length of residence, educational level, and wealth to the above differences (mean=1.67, standard deviation=1.90, range=0–9).
The extend of trust in institutions and public services is captured by institutional trust to any of the following bodies: church, police, local authority, members of parliament, government, municipality or regional government, and health services (52.6%).
Sense of community captures satisfaction with and belongingness in the community. In the current study, it is called attachment and refers to being happy or very happy in the village and perceiving it as a large family in which the respondent belongs (50.7%).
Demographic and Socioeconomic Characteristics
A large number of individual, household, and community characteristics, which may be associated with mental health, are examined in this study. These include sociodemographic attributes (age, sex, nationality, household composition, education, employment), indicators of affluence (income, home, car and pick-up trucks ownership), and residential stability of the individual respondents and their households.
Community Profile
The Community Profile (Grootaert & van Bastelaer, 2002) is as follows: (a) general community characteristics (e.g., population size, principal economic activities, availability of employment, quality of housing, quality of roads); (b) principal services (e.g., availability and quality of electrical service, public lighting, drinking water, home telephone service, public telephones, sewage, waste collection, transportation); (c) recreation facilities, labor migration, education structures, environmental issues, community support (e.g., number of organizations existing in the community); and (d) prevalence of collective mobilization to address a local problem with identification of local social problems. Additional subjective perceptions of the community representatives on issues related to quality of life and levels of trust in their community were gauged.
Examples of the contextual attributes and relevant perceptions of this study are as follows: (a) population size; (b) public resources and facilities, such as primary schools (55.1%), Internet access (57.3%), public market (66.3%), parent-teacher associations (42.7%) or other clubs; (c) quality of life (reported as good by 59.6% of community representatives) and other social features, such as (relative (19.1%)) trust (58.4%), and community mobilisation (31.5%); (d) perceptions about the economy, i.e., improved employment prospects (27%); and (e) social problems prevalence, such as crime (16.9%), drug (14.6%), and alcohol abuse (21.3%).
Summary statistics of all individual and community characteristics together with the key community informants' subjective perceptions, which are employed in the current work, are given in Table 1. The number of cases with valid responses across all sample characteristics for the later statistical modelling is 424.
| Individual level (N=424) | % | Mean (min, max) | Standard deviation |
|---|---|---|---|
| Mental health | |||
| Mental Health Index | 4.5 (1, 6) | 1.0 | |
| Poor mental health (MHI≤3.20) | 13.2 | ||
| CIS-R | 5.70 (1, 27) | 5.73 | |
| Psychiatric morbidity (CIS-R≥12) | 14.2 | ||
| Social Capital | |||
| Social cohesion index | 2.7 (1, 5) | 0.9 | |
| Low social cohesion (Index<2.6) | 45.5 | ||
| Perceived differences | |||
| Total number | 1.7 (0, 9) | 1.90 | |
| In political party preference | 45.3 | ||
| In landholdings | 24.3 | ||
| In mentality | 21.7 | ||
| Institutional trust | 52.6 | ||
| Perceived solidarity | 46.5 | ||
| Collective mobilization | 45.0 | ||
| Willingness to invest | 62.3 | ||
| Attachment | 50.9 | ||
| Number of close friends | |||
| None | 34.9 | ||
| One to three | 49.5 | ||
| Four or more | 15.6 | ||
| Number of friends to ask for money | |||
| None | 34.9 | ||
| One to three | 44.8 | ||
| Four or more | 20.3 | ||
| Individual and household characteristics | |||
| Male | 47.9 | ||
| Age | 43.3 (18, 76) | 13.6 | |
| Non-Greek | 0.9 | ||
| Employment status | |||
| Farmer | 28.3 | ||
| Housewife | 22.4 | ||
| Employee | 20.3 | ||
| Small business | 15.3 | ||
| Pensioner | 7.5 | ||
| Other (unemployed or university student) | 6.2 | ||
| Educational level | |||
| Preliminary or lower | 44.3 | ||
| Secondary | 46.9 | ||
| Tertiary | 8.7 | ||
| Number of children | |||
| None | 60.8 | ||
| One | 18.2 | ||
| Two or more | 21.0 | ||
| Number of adults | |||
| One | 4.5 | ||
| Two | 53.0 | ||
| Three or more | 42.5 | ||
| Home owners | 91.7 | ||
| Length of residence | 22.4 (1, 65) | 14.8 | |
| Household income | |||
| Less than 10,000 euros | 40.6 | ||
| 10,000–20,000 euros | 43.4 | ||
| Over 20,000 euros | 14.4 | ||
| Refused to answer | 1.7 | ||
| Number of cars | |||
| None | 33.5 | ||
| One | 54.2 | ||
| Two or more | 12.3 | ||
| Number of trucks | |||
| None | 38.0 | ||
| One | 57.8 | ||
| Two | 3.8 | ||
| Community level (N=89) | |||
| Population Size | |||
| Less than 99 residents | 32.6 | ||
| 100–249 residents | 34.8 | ||
| Over 250 residents | 32.6 | ||
| Primary school | 55.1 | ||
| Nursery | 49.4 | ||
| Improved employment | 27.0 | ||
| Stable employment | 47.2 | ||
| Good quality of life | 59.6 | ||
| Trust | 58.4 | ||
| Perceived higher trust than other communities | 19.1 | ||
| Interest in community's well-being | 29.2 | ||
| Some internet access | 57.3 | ||
| Public market space | 66.3 | ||
| Recreation areas | 58.4 | ||
| Cooperatives | 80.9 | ||
| Parents teachers associations | 42.7 | ||
| Youth organizations/clubs | 12.4 | ||
| Sport clubs | 42.7 | ||
| Culture clubs | 71.9 | ||
| Common action to tackle a problem | 31.5 | ||
| Property crime | 16.9 | ||
| Alcohol abuse | 21.3 | ||
| Drug abuse | 14.6 | ||
- a Note. CIS-R=revised Clinical Interview Schedule; MHR=Mental Health Index.
STATISTICAL MODEL
The multilevel logit specification is employed to examine predictors of psychiatric morbidity, including social capital. The model accounts for the clustering of individuals within communities and estimates any between communities unexplained heterogeneity for binary observed outcomes (Goldstein, 1995; Snijders & Bosker, 1999). Allowing for extra-binomial variance any divergence of the between individuals variation from the logistic distribution is also estimated.
(1)
. In our estimated models below (see Tables 2–4), 
is estimated to test for any extra-binomial variation, which has a multiplicative effect to the standard logistic variance, i.e., 3.29 
.| Contingency tables | Multilevel logit models | |||
|---|---|---|---|---|
| Percentage of respondents | Pearson  |
 |
||
| With psychiatric morbidity and the following: | Total | (p) | (standard error) | |
| Low-social cohesion | 5.9 | 45.5 | 0.21 (0.15) | |
| High-social cohesion | 8.3 | 54.5 | .418 (0.52) | |
| Perceived differences in political party preference | 8.3 | 45.3 | 0.57** (0.28) | |
| None or other | 5.9 | 54.7 | 4.80** (0.03) | |
| Perceived differences in landholdings | 5.0 | 24.3 | 0.67** (0.29) | |
| None or other | 9.2 | 75.7 | 4.36** (0.04) | |
| Perceived differences in mentality | 3.1 | 21.7 | 0.03 (0.32) | |
| None or other | 11.1 | 78.3 | .00 (0.99) | |
| Institutional trust | 7.3 | 52.6 | 0.05 (0.26) | |
| None | 6.8 | 47.4 | .02 (0.88) | |
| Perceived solidarity | 5.0 | 46.5 | −0.50* (0.28) | |
| None | 9.2 | 53.5 | 3.69* (0.06) | |
| Collective mobilization | 5.9 | 45.0 | −0.25 (0.28) | |
| None | 8.3 | 55.0 | .32 (0.57) | |
| Willingness to invest | 9.7 | 62.3 | 0.36 (0.28) | |
| None | 4.5 | 37.7 | 1.10 (0.30) | |
| Attachment | 5.9 | 50.9 | −0.42 (0.27) | |
| Lack of attachment | 8.3 | 49.1 | 2.41 (0.12) | |
Pearson  |
||||
| No close friends | 5.7 | 3.9 | 0.06 (0.28) | |
| One to three close friends | 8.0 | 49.5 | 1 | |
| Four or more close friends | 0.5 | 15.6 | 7.96** (0.02) | −1.66*** (0.64) |
| No friends to borrow money from | 5.0 | 34.9 | −0.18 (0.28) | |
| One to three friends to borrow money from | 7.8 | 44.8 | 1 | |
| Four or more friends to borrow money from | 1.4 | 20.3 | 5.26* (0.07) | −1.00** (0.43) |
| Total | 14.2 | N=424 | ||
| Analysis of variance | ||||
| Psychiatric morbidity | Lack of psychiatric morbidity | F test 1,422 (p) | ||
| Number of perceived differences | Mean=2.20 | Mean=1.58 | 5.50** (0.02) | 0.15*** (0.07) |
- a *.10>p>.05; **.05>p>.01; ***.01≥p.
| Unexplained variance | ||||
|---|---|---|---|---|
| Between | ||||
| Estimated model (assumptions about explanatory variables' categories) | Residual ICC | Explained variance | Communities | Individuals |
| Baseline | 0.27 | 0.00 | 0.27 | 0.73 |
| Model 1a | 0.17 | 0.16 | 0.14 | 0.70 |
| Model 2b | 0.16 | 0.24 | 0.12 | 0.64 |
| Model 3c | 0.11 | 0.30 | 0.07 | 0.63 |
| Model 4d | 0.08 | 0.32 | 0.05 | 0.64 |
- a Note. ICC=intra-class correlation.
- a aFor a housewife with secondary education and household income 10,000–20,000 euros.
- b bFor the (a) individual who additionally reported 1–3 friends to borrow from, perceived solidarity and 1 perceived difference.
- c cFor the (b) individual who lives in a community where property crime was reported by the community representative.
- d dFor the (b) individual who lives in a community of 100–249 residents, with youth clubs, and where common action was reported.
| Baseline | Model 1 | Model 2 | Model 3 | Model 4 | |
|---|---|---|---|---|---|
Fixed parameters  (standard error) |
|||||
| Intercept | −1.73*** (0.17) | −1.81*** (0.98) | −1.91* (1.00) | −1.78*** (1.04) | −2.65** (1.13) |
| Individual and household characteristics | |||||
| Male | −0.71** (0.36) | −0.81** (0.38) | −0.89** (0.39) | −0.95** (0.41) | |
| Age | 0.01 (0.01) | 0.01 (0.02) | 0.01 (0.02) | 0.01 (0.02) | |
| Employment (other) | |||||
| Farmer | −0.87 (0.66) | −0.88 (0.68) | −0.85 (0.70) | −0.85 (0.73) | |
| Housewife | −0.54 (0.66) | −0.61 (0.68) | −0.55 (0.70) | −0.57 (0.73) | |
| Employee | −0.58 (0.64) | −0.70 (0.66) | −0.63 (0.68) | −0.65 (0.71) | |
| Small business | −0.98 (0.71) | −1.15 (0.74) | −1.04 (0.76) | −0.95 (0.79) | |
| Pensioner | −1.58 (1.02) | −1.40 (1.03) | −1.48 (1.07) | −1.61 (1.12) | |
| Education (higher) | |||||
| Primary | 1.63* (0.84) | 1.90** (0.84) | 1.96** (0.87) | 2.15**(0.91) | |
| Secondary | 1.42* (0.79) | 1.69** (0.79) | 1.72** (0.83) | 1.96**(0.86) | |
| Household income (20,000+euros) | |||||
| Less than 10,000 euros | −0.60 (0.45) | −0.63 (0.46) | −0.63 (0.47) | −0.55 (0.47) | |
| 10,000–20,000 euros | −0.97** (0.44) | −0.96** (0.45) | −1.05** (0.46) | −0.96** (0.47) | |
| Individual social capital indicators | |||||
| No. of friends to borrow from (1–3) | |||||
| None | −0.59* (0.34) | −0.65* (0.35) | −0.66** (0.37) | ||
| Four or more | −0.80 (0.50) | −0.84 (0.52) | −0.84 (0.53) | ||
| Number of perceived differences | 0.18** (0.08) | 0.18** (0.08) | 0.19** (0.08) | ||
| Perceived solidarity | −0.54* (0.31) | −0.45 (0.33) | −0.49 (0.34) | ||
| Deviance (degrees of freedom) | 18.54* (11) | 29.44** (15)a | 29.28** (15) | 28.41** (15) | |
| Community characteristics | |||||
| Property crime | −0.93** (0.47) | ||||
| Population Size (over 250 residents) | |||||
| Less than 99 residents | 0.52 (0.51) | ||||
| 100–249 residents | 0.69* (0.38) | ||||
| Youth clubs in the community | −1.08* (0.59) | ||||
| Common action to tackle a problem | 0.41 (0.40) | ||||
| Deviance (degrees of freedom) | 4.01** (1) | 6.64* (4) | |||
Random parameters,  and  |
|||||
Between individuals extra-binomial variance ( ) |
0.83*** (0.06) | 0.98*** (0.07) | 0.97*** (0.07) | 1.01*** (0.08) | 1.08*** (0.08) |
Between communities variance ( ) |
1.00*** (0.38) | 0.64*** (0.37) | 0.61** (0.36) | 0.40 (0.34) | 0.26 (0.33) |
- a *.10>p>.05; **.05>p>.01; ***.01>p. One-tail tests for variance parameters.
- a aEmployment status does not effectively increase the explanatory power of the model. The Deviance of an estimated model without it is 28.36 with 10 degrees of freedom.
The two derivative statistics that are formally introduced in this section are instrumental for disentangling the individual and community influences on psychiatric morbidity. The first is the so-called intra-class correlation (ICC; Snijders & Bosker, 1999), which depicts intra-community correlation. ICC gives the correlation of the probability of psychiatric morbidity between two randomly selected individuals, who reside in the same randomly chosen community (Snijders & Bosker; Goldstein, 1995), and it implies persistent community unexplained heterogeneity. In plain English, it estimates how much psychiatric morbidity is clustered within communities. Formally, the ICC is calculated as follows:
(2)
(Snijders & Bosker, 1999).
(3)
is the variance of the linear predictor for an unobservable variable, which generates the dichotomous outcome of psychiatric morbidity via a threshold process (Snijders & Bosker, 1999), and 
and 
have been defined above. Because most explanatory variables are categorical, the linear predictor and, therefore, its variance, 
, and the resulting 
are functions of the attributes included in each estimated model. Therefore, different predictors would give a slightly different value of the proportion of explained variance by the model.The estimated models below have been obtained using iterative generalized least squares (IGLS) estimation with first-order marginal quasi-likelihood (MQL) approximation via the software package MLwiN 2.0 (Rasbash, Steele, Browne, & Prosser, 2004).
PSYCHIATRIC MORBIDITY AND SOCIAL CAPITAL
The results of preliminary investigations on the relationship between mental health and each social capital variable are given in Table 2. The middle three columns of Table 2 present simple bivariate associations between each social capital aspect and psychiatric morbidity. These are statistically tested via corresponding Pearson's χ2 values at appropriate degrees of freedom, along with their levels of significance (p value) in parentheses and an indication of achieving the 0.10 (*) or 0.05 (**) commonly used p-value thresholds. (See the fourth column of Table 2.) The second column gives the percentage of respondents who reported psychiatric morbidity and the respective social capital characteristic. For instance, 8.3% of respondents reported psychiatric morbidity and high-social cohesion, while 5.9% have high CIS-R scores and low-social cohesion. (See the fourth and fifth rows in the second column of Table 2.) This does not agree with the theoretical suggestion that poor mental health is associated with low-social cohesion and, indeed, the association is not significant as indicated by the low value of the respective Pearson χ2 statistic with one degree of freedom (0.42, first figure in the fourth column of Table 2).
Psychiatric morbidity is significantly associated with perceived differences in political party preference and landholdings, number of perceived differences, perceived solidarity, social networks, and support. (See the fourth column of Table 2.) More people have high-psychiatric morbidity and perceive political party preference differences than not (8.3% and 5.9%, respectively), contrasting the relative group membership in the general population. Indeed, less people perceive these differences than not (45.3% and 54.7%, respectively). The odds of perceiving landholdings differences are significantly higher in conjunction with psychiatric morbidity (5.0/9.2=0.54) than generally (24.3/75.7=0.32). The mean number of perceived differences is significantly higher for people with psychiatric morbidity (2.2) than without (1.6), according to analysis of variance F test (5.5) with 1 and 422 degrees of freedom.
Just over nine percent (9.2%) of respondents do not perceive solidarity in their communities and are psychiatrically morbid against a 5.0% of respondents, who reported both solidarity and high CIS-R. This difference is disproportionate compared with the general population, which is roughly equally divided between those who do (46.5%) and those who do not perceive solidarity (53.5%). Respondents with psychiatric morbidity also reported fewer (than four) friends and less social support, i.e., less than four people to borrow money from, than those with good mental health.
Social capital may be related to partly similar demographic and socioeconomic characteristics as psychiatric morbidity. It may, thus, be endogenous in the later estimated models. Preliminary multilevel logit regressions of psychiatric morbidity, whereby alternative social capital instruments are the only explanatory variable, have been fitted to investigate their respective unconditional effects. The estimated fixed parameters, together with standard errors and an indication of their statistical significance, of each social capital construct on psychiatric morbidity are presented in the last column of Table 2.
Perceived political party preference and landholdings divisions, as well as the number of perceived differences, significantly increase the odds of psychiatric morbidity by 77%—calculated as (exp(0.57)−1)×100—95%—calculated as (exp(0.67)−1)×100—and 16% for each additional perceived difference, respectively. Perceived solidarity is marginally associated with lack of psychiatric morbidity. Social networks and support of four or more friends significantly reduce the odds of psychiatric morbidity by 81% and 63%, respectively. All other social capital aspects, however, seem unrelated to psychiatric morbidity.
RESULTS
Modelling Strategy
Table 3 shows the summary statistics. Table 4 presents the parameters of the empirical models of the association between psychiatric morbidity and social capital, accounting for other individual and community characteristics and the clustering of individuals within communities.
The baseline or empty model, whereby only a random intercept at the individual and community levels is fitted, is given as a benchmark and disentangles the (unexplained) variation of psychiatric morbidity within and between communities. Apart from a baseline, four models are presented. The first model includes only individual demographic and socioeconomic characteristics (Model 1). All demographic and socioeconomic characteristics of Table 2 entered preliminary versions of Model 1, but apart from sex, age, and employment, which are theoretically pivotal for our research questions, only statistically significant predictors have been retained.
Model 2 expands Model 1 via adding individuals' social capital indicators. All social capital indicators, which have been discussed in the Social Capital subsection, entered incrementally a preliminary Model 2, but, again, only the statistically significant ones have been retained. ‘Social networks and support entered alternative models rather than the same one to avoid multicollinearity because they are highly associated’. For the same reason perceived divisions in landholdings, political party preference, and the total number of differences entered separate models. The results, shown here in Tables 3 and 4, employ the number of perceived differences. The models, which use perceived political party or landholdings differences are strikingly similar to the ones presented here. All results that are not shown here are available from the authors.
Community attributes have been incrementally added to Model 2 to give Models 3 and 4. Any level-two characteristics with statistically significant coefficients have been kept in the model even if their significance became eventually marginal. Property crime seems to encompass all community-level variability (Model 3).
The estimated random, 
, and fixed parameters, 
, of respondents' sociodemographic characteristics, social capital, and their community profile over psychiatric morbidity are given in Table 4. Their respective standard errors and an indication of statistical significance are also shown. The latter is based on Wald tests, which are χ2 distributed with one degree of freedom. Deviance statistics test the join significance of each set of explanatory variables, i.e., individual and community. They are multiparameter Wald tests that follow the χ2 theoretical distribution with the appropriate degrees of freedom (Greene, 1997) and an indication of their statistical significance is provided.
How Much Mental Health is Explained Between and Within Communities?
The summary statistics of Table 3 refer to models with increasing complexity as described in the previous section. The (residual) ICC (see Equation 2) is given in the second column of Table 3, while the next one gives the respective proportion of explained variance by each estimated model, 
(see Equation 3).
Two points should be made with respect to 
: First, as mentioned, the proportion of explained variance, 
, depends on the characteristics included in the linear predictor, which bases its calculation. The attributes selected to calculate the linear predictor and 
of psychiatric morbidity are such that their coincidence is plausible. They are given as notations in Table 3. All individual demographic and socioeconomic characteristics, except employment, and population size refer to the sample's mode. Housewife has been used for employment status, as it makes the fictitious female, who bases our predictions, more tangible. Social capital and the remaining community characteristics that are included in the respective Models 2 to 4 have been assumed in calculating the model's explained variance, 
, to take full advantage of significant predictors. The explained variance for any plausible combination of characteristics can be calculated via similar simulations.
The second point that merits some attention is that, in general, R2 values for nonlinear outcomes, such as from multilevel logit models, are “considerably lower than the ordinary least squares [OLS] R2 values obtained for predicting continuous outcomes” (Snijders & Bosker 1999, p. 226). In light of this, the individual and community characteristics of this work explain surprisingly well psychiatric morbidity (0.32).
The last two columns disentangle the proportion of total unexplained variance (
) between the two sources of variation, i.e., between communities and between individuals (
and 
, respectively). The within communities unexplained variability of psychiatric morbidity drops because of accounting for individual characteristics. The between communities unexplained variability is eliminated and essentially fully attributed to the model's community predictors of psychiatric morbidity (see Models 3 and 4).
The previous observations are reflected in the respective ICC values, which, as mentioned, imply persistent variability of psychiatric morbidity across communities. Considering other social sciences results, for instance, in education, the ICC for psychiatric morbidity is surprisingly high, i.e., 0.27 (for the baseline model). It implies that the psychiatric morbidity of two randomly selected individuals from a randomly selected community is correlated by 0.3. Individual characteristics and social capital reduce the residual ICC, while community characteristics seem to fully account for any persistent, unexplained heterogeneity of psychiatric morbidity between communities. (See also the last row in the fourth and fifth column of Table 4.) To sum up, individual and community attributes explain a significant portion (about 30%) of the variance of psychiatric morbidity and all remaining unexplained heterogeneity (64%) is essentially between individuals. Thus, additional individual rather than community factors, which are unmeasured here, such as family history or generally unobserved ones, may shed more light on poor mental health.
Predictors of Psychiatric Morbidity
Table 4 presents the results of psychiatric morbidity over individual demographic, socioeconomic, and social capital measurements as well as community characteristics. Age, employment, nationality, number of children or adults in the household, home ownership, length of residence in the area, number of cars or pick up trucks, most social capital indicators (see also earlier preliminary associations), and community characteristics are unrelated to psychiatric morbidity.
Men have, at most, half the odds of psychiatric morbidity than women (51% reduction, calculated as (exp(−0.71)−1)×100 from Model 1. By contrast, people with primary and secondary education have at least four—calculated as exp(1.63)−1 from Model 1—and three—calculated as exp(1.42)−1 from Model 1—times higher the odds of psychiatric morbidity than those with higher education, respectively. Primary education is confined with age, as older people tend to have lower qualifications. Indeed the mean age of respondents with just elementary schooling is 50.6 years old, while that of people with secondary and higher education degree is 37 and 40 years old, respectively. Having said that, however, a model omitting education did not improve the statistical significance of age. People from medium income households (10,000–20,000 euros) have roughly 62%—calculated as (exp(−0.97)−1)×100 from Model 1—lower odds of psychiatric morbidity than those with high income (more than 20,000 euros).
Perceived solidarity and, surprisingly, “no friends to borrow from” are marginally associated with a lower odds of psychiatric morbidity by 42% and 45%, calculated as (exp(−0.54)−1)×100 and (exp(−0.59)−1)×100 from Model 2, respectively. The latter, however, should be interpreted with caution, especially because the unconditional association showed that having four or more friends to borrow from predicts a significant reduction of psychiatric morbidity by 63%, while no social support had a negative but nonsignificant parameter (see Table 2). One possible explanation is that social support is highly associated with the sociodemographic characteristics, which are included in the final models, especially age, educational, and income levels. This issue is revisited in the Discussion section.
Each additional perceived difference between community members significantly increases the odds of psychiatric morbidity by roughly 20%. Similarly, perceived divisions with regards to political party preference and landholdings continue to significantly raise the likelihood of psychiatric morbidity when other individual and community attributes are accounted for. Property crime and youth clubs in the community, as reported by the community representative, are associated with lower odds of psychiatric morbidity (60% and 66%, see respective Models 3 and 4, Table 4), while living in a small community (100 to 249 residents) increases these odds by 99%, with marginal statistical significance (see Model 4, Table 4).
All individual characteristics that are included in Table 4 are jointly important predictors of psychiatric morbidity, while property crime is more so than population size and youth clubs together (see respective deviance statistics). The estimated extra-binomial variation of psychiatric morbidity, 
, confirms that its distribution is well approximated by the logit specification. The between communities variation is fully accounted for by the community characteristics in Models 3 and 4 of Table 4.
DISCUSSION
To our knowledge, this is the first study on social capital and mental health that employs a fully structured, well-validated clinical interview (CIS-R) as a mental health outcome within a rural setting. A large number of rural communities (n=89) was examined from five islands, ensuring variation in scores between individuals and communities. Sampling all the small rural settlements of a whole region for data collection and conducting a detailed psychiatric interview are strengths of this study. Another contribution is the use of standard definitions of social capital in conjunction with measures of its different aspects, including social participation, trust, social cohesion, beliefs of collective efficacy, and sense of belonging. This study disentangles the individual and community influences on psychiatric morbidity through multilevel modelling. Psychiatric morbidity is, to a large extent, clustered within rural communities.
The present findings succeeded in providing evidence about the roles that some aspects of social capital may play in mental health. Perceived divisions between community members with regards to political party preference and landholdings significantly raise the likelihood of psychiatric morbidity, even when other individual and community attributes are accounted for. Each additional perceived difference between community members significantly increases the odds of psychiatric morbidity by roughly 20%. Perceived differences with regards to political party preferences may reflect differentiated access to power resources, both material and symbolic. Social anthropological research has revealed that political preferences constitute an indication of how rural people relate to central administration and decision-making structures (Papataxiarchis, 1991). The way rural people relate to central authorities is indicative of how social positions are constructed within the specific sociocultural context. This suggests that some community members are more privileged than others, highlighting the need to examine issues of social position and roles in these small communities. The findings support the “psycho-social” theoretical perspective in the area of health inequalities, which argues that perceptions of relative deprivation or low-social status engenders psychological distress, as expressed in feelings of low self-esteem, which affect the breakdown of social cohesion (Wilkinson, 1996).
Social cohesion is conceptualized as the degree of trust, sense of familiarity, shared values, and bonding relations between individuals within a community (Carpiano, 2006). A critique, however, on the notion of social cohesion as an umbrella term that covers a range of social processes seems to gain some support by our results. Indeed, it was expected that strong perceptions of trust and social cohesion would correlate inversely with psychiatric morbidity. This research, however, shows that trust and social cohesion are unrelated to mental health, unlike previous evidence of significant associations among trust, social cohesion, and GHQ (General Health Questionnaire) scores in urban settings (Araya et al., 2006). The lack of association may be because of the scale's inability to capture enough complexity or meaning in rural communities. Another interpretation is that close-knit networks, generalized trust, and shared values do not seem to integrate status differences. It is also speculated that strong ties through kinship networks within these communities may signal heavy obligations. Therefore, social cohesion appears to be a multifacet phenomenon that deserves further research.
Social capital in this study measured structural aspects by asking respondents to indicate participation of household members in voluntary or local organizations, extent of help received from friends for various needs, e.g. borrow money, and willingness among neighbours to help in hypothetical situations. The item concerning group membership was found to have no meaning to our respondents. Within the study's cultural setting, group membership is more implicit and informal in nature. It is clear that group practices of Greek villages vary from those described by Putnam (1993), and, therefore, the current study failed to reveal associations between structural social capital and mental health when other factors are accounted for because of cultural factors: borrowing money may seem degrading and lack its original theoretical meaning. The evidence that wider social support (four or more friends to borrow from) unconditionally reduces psychiatric morbidity risk suggests that its role may be conditioned by sociodemographic characteristics (McKenzie & Harpham, 2006).
Questions about perceived solidarity and collective efficacy consider community members' beliefs in their ability to act collectively to address a common issue. These beliefs were indicated by whether the majority of the rural community's residents would join forces to address a common disaster, would be involved in organizing a festival or a fair in the village, and would be willing to invest both money and time. The present research found that perceived solidarity is associated with lower odds of psychiatric morbidity, albeit at marginal statistical significance. This suggests that feelings of community competence can have many positive effects on mental health such as a sense of security, perceptions of control, and hope (McCulloch, 2001).
Sense of belonging is arguably an important dimension of social capital and a number of theoreticians propose that it is the glue that holds communities together (Sarason, 1974). The current study however did not evidence any relationship with mental health. Our study is limited in that only two items were used to capture this dimension.
Community attributes play a significant role to levels of psychiatric morbidity. The unexpected negative association between property crime and psychiatric morbidity implies that property crime may foster sense of community. A number of researchers have suggested that there is a curvilinear relationship between local problems and sense of community (Anderson & Milligan, 2006). A moderate degree of fear of crime may serve as a catalyst for the members of a community to come together to work on resolving threats (Chavis & Wandersman, 1990).
To sum up, the results offer powerful evidence that perceived social distinctions in a rural context may damage individuals' psychological well-being. The internal dynamics, however, between psychological processes, which link perceived social divisions, social statues, and psychiatric morbidity, are not fully understood. This study reinforces the need for measures of social capital that capture the complexity of the concept and empirical analyses that model mental health and social capital jointly.
