Introduction

There is a well-recognised geographic distribution of the social, demographic, economic and environmental factors that may influence health status [40, 45]. There is also evidence of geographic disparity in the burden of ill-health between metropolitan and rural regions in many countries [2, 3, 49]. However, our understanding of the basis for health disadvantage has been obscured by the various ways in which “rural” regions are defined in the international research [45]. The influence of “rurality” in previous ecologic studies could be re-conceptualised as a simple proxy for the set of specific geographically dispersed health determinants that prevail in non-metropolitan areas [22, 32], including: socio-economic disadvantage (greater in many rural areas); greater exposure and vulnerability to environmental adversity (e.g., severe drought) and its socio-economic sequelae; vulnerability to change and related impacts on community infrastructure; poorer access to health and social services; and geographic isolation (and, more specifically, social isolation) [17, 40, 45].

Improving our understanding of the role, these factors play in the relationship between rurality and health, and identifying other health determinants that may be uniquely rural in nature (such as environmental adversity), can help in the development of programs to overcome the health disparity associated with rural life. The need for fine grained analysis of locality in relation to health is supported by a large European study [26], which suggests that locality size, with age and gender, can reveal important differences in rates of mental disorder across regions. Australian research has highlighted the influence of gender, socio-economic status and migration to rural areas in understanding rural suicide [31, 48] and the adverse impact of population decline on mental health in some rural areas [17]. These findings support the view that epidemiologic research to investigate mental health needs in diverse localities should incorporate more detailed understanding of the interaction of individual level determinants of health with the social and demographic characteristics in those regions [21]. Furthermore, most research regarding the influence of community/neighbourhood characteristics on health has focussed on urban environments—less attention has been paid to investigating how aspects of location or place in rural regions (such as economic, social, environmental characteristics) influence mental health [17]. Weich et al. [50] investigated the impact of similar contextual characteristics on mental health in a large UK study, but found no significant effect of these on regional variation in psychological symptoms. Their findings demonstrated the importance of household level influences (e.g., household income and characteristics) on variance in levels of mental health. Similarly, the findings from a large Australian study [10] confirm the contribution of household level factors to mental health. Although these studies were not specifically addressing rural populations, they reinforce the importance of investigating a range of contextual factors in improving our understanding of variations in mental health.

The potential vulnerability of people in rural areas to environmental adversity is starkly illustrated in the social and economic consequences of protracted drought in rural Australia. Greater reliance in rural areas on primary industry and agricultural productivity increases this vulnerability [45]. The limited research to date has highlighted the impact of severe drought, such as loss of employment, population and infrastructure decline, and related social change [41, 46]. These life stressors may be an important influence on health outcomes, particularly mental health outcomes, in rural communities. However, it is also essential that studies examining more comprehensive sets of rural factors do so against the backdrop of the traditional individual-level determinants of mental health and well-being that have consistently been identified in other studies, such as socio-demographic, dispositional, life events and social support factors [36].

The current study

The Australian Rural Mental Health Study (ARMHS) is a longitudinal population-based study established to examine the determinants of mental health in rural and remote communities, with specific reference to the interaction of individual, family/household and community level factors. This paper reports on a cross-sectional analysis of the baseline sample examining the contribution of individual dispositional factors, community characteristics and particular features of the rural environment to mental health and general well being. It focuses particularly on the role of geographically distributed characteristics. We hypothesised that factors known to exhibit significant variability across rural communities, such as the extent of remoteness and environmental adversity, would have a significant independent effect on the measures of mental health and well-being, after accounting for the role of individual dispositional characteristics and other factors that may moderate the level of health and well being (e.g., social support, social connectedness, perception of community characteristics). The findings reported here describe data from a subset of instruments used in the baseline ARMHS surveys (see [23], for further details).

Methods

Recruitment

Sixty local government areas (LGAs) were identified from three rural health service regions of New South Wales (NSW) using the rural, remote and metropolitan areas (RRMA) classification. These areas cover approximately 70% of the geographic region of non-metropolitan NSW. The sampling frame excluded metropolitan zones (i.e., capital city and other urban centres with populations >100,000); the remainder comprises rural and remote zones based on the population size. Over-sampling of the remote and very remote regions of NSW in the current study ensured sufficient sample sizes from these regions (see below).

The baseline sample comprised household residents aged 18 years or older living in private dwellings in rural communities in NSW recruited between 2006 and 2009. Individuals were randomly selected from the Australian electoral roll (AER) and linked telephone numbers were obtained using a national telephone directory. The survey excluded special dwellings (such as hospitals, nursing homes, prisons, hotels and hostels) and overseas visitors usually resident outside Australia. An introduction letter and brochure were provided, giving background to the survey, and information about associated interview processes. For households where residents could not be contacted in the first instance, another letter was sent and five contacts attempted before a household was classified as a non-contact. The project was approved by the Human Research Ethics Committees of the participating institutions.

Exclusion criteria

Participants aged 65 years or over were screened using the modified telephone interview for cognitive status (TICS-M) [14] and those with a TICS-M total score <17 were excluded. Non-english speaking members of a household, those with significant hearing impairment that impeded consent and/or interview, and those with no identifiable telephone contact number (after directory and electronic database search) were also excluded.

Instruments

Current psychological distress and well-being

The Kessler 10 + LM (K-10) [5, 25] was used to assess psychological distress (ten items) and related disability (e.g., days out of role) during the last month (four items). In addition, single item self-report measures were devised to assess general functioning over the same period on several domains: ability to perform everyday duties and tasks, overall physical health, overall mental health, satisfaction with relationships in general, and overall satisfaction with life. Responses were scored on 5-point Likert scales from ‘poor’ to ‘excellent’ for functioning, health and relationships, and from ‘not at all satisfied’ to ‘extremely satisfied’ for satisfaction with life. The K-10 and general functioning/health items were also combined to provide a single measure of well being (see below in “Data analysis”).

Individual factors

These were measured using the following instruments

  1. 1.

    Personality—the 12-item short form EPI measure of neuroticism (EPI-12; [16]) was used, from which a 7-item subset was identified (i.e., being easily hurt, a nervous person, a worrier, being highly strung, suffering from nerves, worrying too long, and often guilty) that conceptually reflected predispositional or trait characteristics, and may be usefully delineated from current distress items. This reflects previous work by the investigators [11] and others [15, 35] regarding the confounding effect of state factors (such as current distress) on the scores obtained from measures of “neuroticism”, and the potential overlap of some item content from the EPI and instruments such as K-10. A confirmatory factor analysis supported this approach, with the 7 selected items contributing to a distinct single factor (eigenvalue, 2.86) accounting for 41% of the variance in EPI-12 scores;

  2. 2.

    Recent adverse life events/difficulties—an inventory of events/difficulties in the preceding 12 months was adapted from the life events inventory [8];

  3. 3.

    Social support—a self report measure of availability of social supports was used (derived from the interview schedule for social interaction; [20]); and

  4. 4.

    Sense of community index [12]—this 12 item true/false self-report instrument assesses individual perceptions of place or district, across a number of key domains of social capital: ‘shared emotional connection’, ‘community membership’, ‘community influence’, and ‘community reinforcement of needs’; sample items: “I think my district is a good place for me to live”; “I feel at home in this district”; “few people know me in this district”; “people in this district don’t get along with each other”.

Individual level contextual factors (i.e., personal exposure to and perception of rural characteristics)

These measures included:

  1. 1.

    Infrastructure and services accessibility—four items were specifically designed to reflect common concerns in rural communities about infrastructure, including population change (e.g., access to health care or other services, concerns regarding fuel prices, and people moving in or out of the district). Each item was scored on 5-point Likert scales ranging from ‘not at all’ to ‘a lot’;

  2. 2.

    Perception of drought severity—a single item screening question regarding level of ‘stress’ caused by drought, using a similar Likert scale; and

  3. 3.

    Duration of participants’ exposure to the specific rural environment and community—which was assessed through the number of years residing in the current rural district.

District/neighbourhood level contextual factors

District-level variables were obtained from existing databases linked to individual-level data by postcode. These data sets were extracted from the Australian Bureau of Statistics 2006 Census data (measures of population change, socioeconomic status) at either local government area (LGA) or postcode levels.

  1. 1.

    The Index of Relative Socio-Economic Disadvantage (IRSD) [6]—which is a standardised score based on collation of household level census data; higher IRSD scores indicate less disadvantage. It was chosen for its capacity to provide a postcode level average score representing key dimensions of disadvantage (including income, education, employment and household vehicle access). Remoteness is not included in this measure.

  2. 2.

    District population change—representing the percentage change (from 2002 to 2006) within the estimated resident population of the LGA, based on the birth and death registrations and net migration data.

  3. 3.

    Drought severity—exposure variables were calculated for our study regions by Ivan Hanigan at the National Centre for Epidemiology and Population Health of The Australian National University for the 12 months preceding data collection (using a PostgreSQL database [http://www.postgresql.org] with the PostGIS spatial extension [http://postgis.refractions.net]). The Australian Bureau of Meteorology’s gridded monthly rainfall data 1890–2008, at a resolution of 0.25° latitude–longitude [33], were used to calculate a drought index based on six-monthly percentiles for each grid cell’s rainfall record [44]. These grid values were then averaged within our spatial units with no weighting by population density as this is not relevant for drought exposures [19]. Further details are available from the corresponding author.

  4. 4.

    Remoteness

    1. a.

      The Accessibility/Remoteness Index of Australia Plus (ARIA+)—which is considered to be Australia’s most authoritative measure of remoteness, developed by the GISCA (National Centre for Social Applications of Geographic Information Systems: www.gisca.adelaide.edu.au). ARIA describes remoteness from goods and services for any part of Australia. It relies on road distance as a surrogate for remoteness and on the population size of a service centre as a surrogate for the availability of services (p 6; [1]). The index is based on the distance to five categories of service centre. Socio-economic factors and locality population size are not included in the measure. The index values for any populated locality range from 0–12 (with a value of 0 representing the highest level of access to goods and services, and a value of 12 the highest level of remoteness).

    2. b.

      The Australian Standard Geographic Classification (ASGC)—which allocates classes of remoteness to localities based on ARIA index values: major cities, inner regional, outer regional, remote and very remote. This study selected people residing in non-metropolitan areas (i.e., outside major cities).

Data analysis

Data entry, cleaning, aggregation and analysis techniques primarily involved SPSS (version 17.0; SPSS, Chicago, IL, USA) and SAS (SAS V9.1; SAS Institute Inc., Cary, NC, USA) statistical software.

Seven scores were included in an aggregate measure of current well-being: two K-10 + LM based scores (i.e., overall K-10 score, and days out of role in the past month); and five Likert ratings (overall physical health and mental health, ability to perform everyday duties and tasks, satisfaction with relationships, and overall satisfaction with life). Based on a principal components analysis of these scores (n = 2,620), the current well-being measure demonstrated a robust single factorial structure (eigenvalue = 3.92, accounting for 56% of the variance), with factor loadings >±0.45 for each component measure. A current well-being index score was calculated for each participant by standardising their raw scores for these measures (using the relevant baseline grand means and SDs), reversing the K-10 and days out of role standardised scores, and then averaging these scores. Participants with five or more of the seven measures were assigned a current well-being index score, with higher scores indicating greater well-being. The correlation between baseline current well-being and K-10 scores was r = −0.78.

For the major analyses, hierarchical multiple regressions were used, with a predetermined order of entry for the predictor variables, which reflected a mixture of factors, including the presumed order of influence of these variables (e.g., chronological) and their immediacy (e.g., individual vs district-level variables). Planned orthogonal contrasts were used to characterise differences within several of the categorical variables (e.g., employment status and marital status). Aspects of reported personal exposure to rural factors were included in the hierarchical models before related district level characteristics (e.g., personal concerns about the drought versus drought severity indices). As a partial control for the number of statistical tests, the minimum threshold for statistical significance was set at p < 0.01.

Results

Response rates and sample biases

Sample characteristics and biases have been described previously [23] and are only summarised here. Because of the longitudinal nature of the study, priority during recruitment was on establishing a cohort that could be maintained over an extended period rather than on maximising the initial response rate. Letters of invitation were sent to 13,251 individuals, of whom 3,570 (26.9%) were unable to be contacted (n = 3,152); deemed ineligible (n = 343, for reasons such as: residency in excluded dwellings, non-English speaking, hearing impairment, cognitive impairment); or deceased (n = 75). Of the remaining 9,681 individuals, 5,565 (57.5%) were contacted and declined; and 1,477 (15.3%) gave initial verbal agreement by phone but did not reply to further contact. Of those contacted and eligible, 2,639 (representing 1,879 households) agreed to participate and completed an initial survey (participation rate of 27%). Participation rates varied across ASGC categories (χ 2(3)  = 18.20, p < 0.001), with a marginally higher rate in remote regions (31.1%).

As remote and very remote regions were over-sampled, our examination of sampling and participation biases focussed on fluctuations within ASGC categories. There was no difference in gender distributions between those sampled and the ABS 2006 Census population data (χ 2(1)  = 0.30, p = 0.58). Within ASGC categories, there were fewer people in the 18–47-year-old categories compared with population data (inner regional, χ 2(4)  = 118.25, p < 0.001; outer regional, χ 2(4)  = 121.63, p < 0.001; remote, χ 2(4)  = 145.55, p < 0.001), except in very remote regions. Among those eligible, no significant participation bias was evident for gender, either among those who were unable to be contacted, declined to participate, or failed to return surveys. Among those deemed ineligible for reasons already outlined, there were significantly fewer males (χ 2(1)  = 19.28, p < 0.001). Age category was also associated with contact rates (χ 2(12)  = 32.22, p < 0.001) and survey return rates (χ 2(12)  = 35.77, p < 0.001). In short, younger people in general, the oldest age group within inner regional areas, and younger people in very remote regions were among the most difficult to contact. Within the inner regional and very remote regions, younger people were also more likely to refuse to participate, or to agree but not return a survey.

The major regression analyses were based on 2,462 participants (i.e., those with complete data for the selected independent and dependent variables, see Table 3). For consistency, participant characteristics are reported for this subsample.

Sample and regional characteristics

Table 1 outlines the demographic and psycho-social characteristics of the sample, while Table 2 reports district-level characteristics by ASGC category. Significant differences were detected across remoteness categories for age, with participants from very remote regions tending to be younger (p < 0.001). Highest rates of employment, lowest rates of retirement, and lowest education levels were present in the very remote category (p < 0.001). Gender differences across categories did not reach statistical significance. Lowest well-being and highest total K-10 scores and “caseness” rates were also detected among participants in the very remote category (p < 0.001).

Table 1 Participant characteristics by ASGC (remoteness) category
Full size table
Table 2 District characteristics by ASGC (remoteness) category
Full size table

As shown in Table 2, key regional characteristics of interest demonstrated significant differences across ASGC categories. Predictably, ARIA index scores differed significantly in accordance with the criteria for ASGC/ARIA categorization of LGAs. IRSD scores were lowest in remote and highest in inner regional categories (p < 0.001), indicating greater disadvantage in remote regions; drought severity demonstrated significant variation by ASGC category with highest drought severity in the very remote category (p < 0.001). Inner and outer regional areas experienced minor population growth (0.50–0.87%), whereas in remote and very remote regions population decline was noted (−0.70 to −0.98%).

Determinants of mental health and well-being

Conceptually, related variables were grouped into five sets or domains, reflecting the study’s theoretical interest in the role of predispositional, environmental and contextual factors with a focus on rural-related characteristics: set 1, basic demographic factors (age and gender); set 2, pre-dispositional factors (trait neuroticism, marital status, education); set 3, recent difficulties and support (adverse events/difficulties and employment status, perception of individual social support, sense of community); set 4, personal rural exposure (infrastructure and service accessibility, drought-related concern or worry, and duration of residence in rural district); and set 5, rural contextual factors (secondary data sources regarding district characteristics: remoteness category, population change over time, regional index of socio-economic disadvantage, and regional drought severity over the preceding 12 months).

Table 3 reports the univariate associations (i.e., Pearson correlations) between these variables and the well-being score and total K-10 scores. Significant univariate associations were detected between approximately 60% of the predictor variables and both outcome scores, with the strongest associations (r > ±0.30) detected for neuroticism scores, recent adverse events/difficulties, and personal social support (p < 0.001)—indicating that higher well-being (and lower K-10 scores) were associated with lower neuroticism, fewer adverse events, and higher social support. The associations between sense of community index scores and rural service and support accessibility were also statistically significant. Of interest is that worry about drought, years in the district and remoteness demonstrated only modest associations with well-being scores. Contextual factors such as population change, regional socioeconomic ranking, and 1-year drought data were not associated with well-being or K-10 scores.

Table 3 Predictors of current well-being and Kessler 10 (K-10) scores—results from five-step hierarchical regression analyses (n = 2,462)
Full size table

A five-step hierarchical regression analysis (see Table 3) was conducted using these five sets of variables, effectively giving higher priority to individual demographic and predispositional factors, followed by recent difficulties and support (as proposed moderating factors), then personal and geographic level rural contextual factors—thereby, enabling investigation of the impact of rurality measures on well-being after accounting for other predisposing and moderating factors. This model accounted for between 41% and 45% of total variance in well-being and K-10 scores, respectively. With respect to the total well-being index scores, the predispositional factors accounted for 22% of the variance (chiefly neuroticism, and to a lesser extent marital status). Recent difficulties and support accounted for a further 18% of the variance (chiefly adverse events and difficulties, employment status, social support and neighbourhood/community support). Rural exposure and remoteness factors made a minor contribution in this model (e.g., perceptions of infrastructure and services, standardised regression weight = −0.11). Not surprisingly K-10 scores followed a similar pattern, although there was a minor but significant effect for age (standardised regression weight = −0.15), with younger participants reporting higher distress on the K-10.

We also conducted a parallel series of regression analyses excluding neuroticism, although the items most likely to reflect current distress had already been dropped from the neuroticism measure. Apart from overall reductions in explained variance (well-being: to 32%; K-10: to 29%), the pattern of significant predictors was unchanged.

In summary, the factors chiefly associated with well-being and its component K-10 scores were predispositional (neuroticism), recent adversity, and personal social support and social capital measures. Personal rural exposure also had a significant additive effect on the primary outcome measures. Postulated rural factors such as drought worry failed to demonstrate a significant effect in these models and years of living in a rural area did not influence well-being or K-10 scores in either direction.

In view of the significant contribution of recent adverse events and difficulties, these associations were examined in more detail (see Table 4). Almost a third (29.5%) of participants reported no events during the previous 12 months, 49.1% reported 1–2 life events, and 20.4% reported 3 or more events. It is of interest to note the extent of difference across these categories, with substantial lowering of well-being scores and elevation of mean K-10 scores at what appears to be a threshold level of three or more life events. As shown in the lower-half of Table 4, those events associated with the highest distress scores reflected problems in personal social networks (e.g., arguments, financial crisis, or loss of job) or serious personal accidents. There was no significant difference in the total number of life events/difficulties across ASGC categories (see Table 1).

Table 4 Profile of adverse life events during the last 12 months (n = 2,462)
Full size table

Discussion

Research within rural communities provides an important opportunity to investigate individual and community factors contributing to mental health and well-being, and the influence of place or locality [17]. The ARMHS project will provide comprehensive longitudinal data from rural and remote communities, with a particular focus on individual, social and environmental factors pertinent to non-urban Australian settings. The baseline (cross-sectional) phase of the study, reported here, aimed to investigate the influence of a range of factors within rural districts reported at an individual level and also examined through district-level data. Rural characteristics that were assessed included specific rural adversity (e.g., perceived adverse impacts of drought), perceived accessibility of health and other services, community connectedness and community population changes. District-level contextual variables assessed severity of drought, population change, area-based indices of remoteness and socioeconomic disadvantage.

Minimal support was found for our original hypothesis regarding the impact of remoteness and environmental adversity on mental health and well-being, but the study demonstrated an important moderating effect of a set of personal characteristics and individual level contextual factors. Within the univariate analyses (see Table 2), an expected regional variation was detected (across ASGC categories) in a range of postulated district-level influences on mental health (e.g., socio-economic disadvantage, population change, drought severity), and regional variability was also detected in levels of psychological distress and well-being (see Table 1), with the poorest health in very remote regions. Nevertheless, district level impacts on mental health were not evident in the major regression analyses (see Table 3).

Overall, the study findings suggest that despite these regional differences, the most significant contributors to well-being in this sample were those relating to personal predispositional factors, life events and aspects of social support, rather than the district or locality factors that we measured. These findings are consistent with those from other large scale studies such as the UK National Mental Health Study, which demonstrated that geographic location was far less relevant to mental health outcomes than personal and experiential factors (marital status, life events, social support) [36], and consistent with international research indicating more complex interactions between location, socio-demographic factors and mental health than can be addressed in simple rural–urban comparisons (ESEMED study) [26]. The effects of locality that were significant were reflected in findings regarding sense of community (indicative of individual perception of community), suggesting a significant influence of social capital factors on well-being. Neuroticism accounted for the greatest proportion of variance in measures of distress and well-being (over 20–30%). There is a large body of research regarding neuroticism as a measure of individual propensity to poorer mental health, including poorer outcomes in response to adversity (e.g., [24]). There has also been considerable debate about the state and trait characteristics of neuroticism [15, 35]. In this study, we delineated a set of items that most closely identified longer term emotional responses and characteristics in an abbreviated measure, building on previous research undertaken by the investigators [11, 29], arguably strengthening the capacity to capture trait characteristics that will be assessed in more detail in subsequent phases of this study. Nevertheless, trait neuroticism may lead to a reporting bias that influences propensity to endorse difficulties or adverse events, or may, through greater interpersonal sensitivity, contribute to an increased likelihood of reporting problems within social networks (e.g., perception of poorer social support or household difficulties/arguments).

An independent association between life events/difficulties and mental health and well-being was detected in this study. It is possible that the measure of life events/difficulties represented a proxy measure for factors reflecting geographic differences, but the absence of an association between well-being and the district-level variables used in this study would argue against this. The life events with greatest impact on the outcomes measured in this study related to financial strain and interpersonal difficulties. This may reflect the impact of common forms of adversity in rural regions, and provide further evidence of the potential importance of personal economic and interpersonal factors. More detailed study of the geographic distribution of life events is necessary to identify potential “rural” experiences embedded in these data (e.g., financial impact of drought).

The increasing recognition of the significant influence of community characteristics on health is reflected in a growing literature examining the role of social capital in health outcomes [37]. This concept encapsulates both individual and community-level characteristics, such as perceived trust, shared values, security, participation, and community resources, amenities and services [4]. Attention has been given to such factors, and the role they might play in understanding geographic health variations [13, 51]. This includes subjective ratings of quality of communities [18], social support [9, 47], social integration and attachment [39], access to services [52], confidence in getting help from neighbours [28], and key socioeconomic factors such as income inequality [51]. After taking account of individual-level characteristics, community-level socio-economic characteristics have been associated with depressive symptoms and higher rates of major depression in urban settings [42]. Lower rates of depression among rural residents in a large Canadian study were significantly associated with a stronger sense of community belonging and higher social support [38]. These studies highlight the importance of research on individual, household and community level characteristics as a critical context that may moderate individual mental health over time. It is also important to consider differences between the impact of districts on people (contextual factors) and “compositional factors” (i.e., the impact that people make on a district) [27], requiring more fine-grained investigation in multilevel modelling, particularly when considering supportive behaviours and connections within communities.

The potential importance of location was demonstrated by the small but significant contribution of perceived barriers to services and supports, and the important role detected for perceived community support. More detailed pathways investigating how these factors interact with rural events such as drought are needed. The univariate findings identified stress/worry related to drought rather than drought effects per se (or at least those evident in the geographic data used). It is possible that longer term drought severity data (including long term versus shorter term drought exposure, or drought in those regions previously less affected by drought, as compared with those “accustomed” to drought impact), and attention to other environmental events (such as severe storms and floods that also affect communities in this study) will be necessary. It is possible that worries about drought and general levels of distress coexist, with drought being a dominant theme among those currently anxious or depressed, and the worries being a symptom of that distress rather than a trigger for such distress. These cross-sectional data are limited in their capacity to clarify the causal pathways, but longitudinal data such as that being collected on this sample at the time of writing, could inform this issue. While there was a significant overlap between the current distress (K-10) and well-being measures used here, more comprehensive assessments of overall functioning are likely to be indicative of longer term aspects of well-being.

There are a number of limitations to this study. Sampling bias is unavoidable with any recruitment strategy. The primary aims of this study concern the patterns of association and determinants of mental health over time, rather than determining cross-sectional prevalence rates per se, hence lessening the potential impact of any response bias [30]. Although the response rate is low, it matches other population-based surveys in rural communities [7, 21] and those using telephone-based recruitment [34]. Procedures recommended to maximise the response to telephone recruitment were incorporated in the study methods [34] and a strategy of timely local community promotion, with the survey conducted through local rurally based research units, was also part of the strategy to enhance local support for the study. An urban sample was not included as the study’s focus was on rural characteristics and the distribution and type of risk factors across rural areas. Whether these models are replicated in urban areas will be examined in future analyses linked to other comparable Australian urban data sets (e.g., [43]).

Despite these limitations, the current ARMHS project has a number of strengths. The investigation includes a broad range of individual and district level variables, including well-established individual-level determinants of mental health, individual perception of community characteristics and contextual factors reflecting important aspects of rural adversity (such as drought severity). As such, it is one of the few research studies to assess this range of individual and contextual factors within non-urban populations.

The findings suggest that the personal predispositional factors and individual perception of social context (including factors reflecting aspects of social capital) are the major determinants of well-being. There are inherent limitations in the reliance on cross-sectional data and the influence of current distress on perceptions of support and community characteristics. Nevertheless, the findings provide qualified support for the importance of focusing on enhancing community connectedness and personal support, and reducing isolation, as a means of promoting resilience and adaptation in these regions and the important potential moderating effect of these factors when considering the impact of environmental adversity in rural areas.