Psychological life satisfaction is correlated with a range of health behaviours and outcomes, including smoking, harmful levels of drinking, anxiety, depression and suicide, and it can predict mental and physical health for up to twenty years 1 2 3 . Levels of psychological life satisfaction reflect people’s assessment of their life as it is, compared to life as they wish it were. It is related to (but distinct from) happiness, which refers to transitory levels of affect across time 4 .

Despite the complexity of the concept, psychological life satisfaction is often measured using a single item e.g. 5 . However, there is a general consensus that psychological satisfaction is formed from perceptions of satisfaction with multiple life domains 6 7 8 . A heterogeneous range of domains has been examined and some researchers have sought to develop a comprehensive set 9 . Personal Wellbeing Index (PWI) is notable among these, because it has been iteratively developed over time 10 and, through the International Wellbeing network 11 , it is now being used as the basis of an international effort to develop a standardised cross-cultural measure of life satisfaction. The PWI parsimoniously measures eight domains of satisfaction: achieving in life, future security, standard of living, health, safety, religion/spirituality, relationships and community connectedness 12 13 14 . Cummins and colleagues have shown these domains are inter-correlated, and correlated with overall psychological life satisfaction, while cross-cultural research has validated the scale in several countries including Algeria 15 and China 16 .

Farmers are a population that could particularly benefit from a better understanding of the dynamics of life satisfaction. With the advent of climate change, countries including Australia, China and the United States of America will face periods of prolonged and intense dryness, placing farmer populations at risk for greater incidence of health problems 17 . Farmers are not simply financially vulnerable to the effects of climate variability, remoteness limits their access to health services, and this problem is compounded by a traditional reluctance to acknowledge the health problems; especially mental health problems 18 19 . However, farmers are willing to discuss level of psychological life satisfaction.

Unfortunately, existing research into domain and overall psychological life satisfaction for farmers versus the general community has yielded mixed results 20 21 and ways to address low satisfaction are still under theorised. Recently, researchers have begun to examine ways to improve life satisfaction, rather than purely using it as an effective sentinel measure of health and wellbeing. Thus far, research has examined the relationship between satisfaction and several different factors including participation in multiple social groups 22 , being in a relationship 23 , employment 23 and public welfare spending 24 . However, as it is currently understood, psychological life satisfaction is not very informative about how farmers’ distinct characteristics should be incorporated into health promotion programmes and policy.

The objective of our study was to identify key concepts and relationships that can be used to predict, affect and understand psychological life satisfaction in context, with a view to promoting life satisfaction, and thereby public health. As a population increasingly at risk from climate change, we examined Australian farmers and compared them to a general community sample of Australians using the PWI index. Our approach was to quantify the way that perceived satisfaction with different life domains combines into an overall sense of psychological life satisfaction.

The process by which combined domain satisfaction forms overall psychological satisfaction is typically conceived in terms of simple addition 9 . However, a few studies have argued that some domains may have disproportionate influence on overall satisfaction. Mixed results have been found when asking individuals to explicitly weight the importance of different domains 25 26 27 , but one study has demonstrated that the ratio by which different domains contribute to overall satisfaction can be determined statistically 9 .

The novel approach we took to quantification was to consider how perceptions about satisfaction with life domains may conceptually overlap or co-vary in quantifiable ways prior to combining into psychological life satisfaction. Conceptual overlap in the perception of life domains is suggested by intersecting lines of research on social capital and personal efficacy. Social capital is a broad term capturing the concept of beneficial connection to others through social participation and social cohesion, and it is positively related to life satisfaction 28 . Three PWI domains: community connectedness, relationships and religious participation are all empirically established as components of social capital 29 . Personal efficacy refers to the perceived power to perform challenging tasks and achieve goals, and the PWI domains achieving in life and future security are elements of personal efficacy 30 . Consequently, we hypothesised that perceived domain satisfaction would combine into overarching ‘supra-domains’ that have distinct and separable relationships with psychological life satisfaction.

We did not make a firm prediction about the number of supra-domains that may occur since the role of perceived satisfaction with some PWI domains (standard of living, health and safety) was ambiguous. However we did expect that the eight discrete domains would combine into at least two supra-domains, “efficacy” and “connectedness”. Once modelled, such supra-domains can be tested for unique relationships with each other and with psychological life satisfaction. We therefore tested the hypotheses that:

1. (i) Satisfaction with domains will cohere into psychological life satisfaction. (ii) Some domains will be more influential than others.

2. Before cohering into psychological life satisfaction, domains will first combine into supra-domains, like “efficacy” and “connectedness”.

3. Supra-domains will have unique relationships with each other and with psychological satisfaction.

The data were originally collected to inform the National Review of Drought Policy, undertaken by the Drought Review Branch of the Australian Government Department of Agriculture, Fisheries and Forestry 31 . In keeping with Australian Government regulations, the Australian Bureau of Statistics’ Statistical Clearing House acted as an external ethics committee, reviewed the survey for ethical concerns, and granted approval for its conduct. In June of 2008, under the direction of the Australian Bureau of Rural Sciences, the polling firm Newspoll collected data from two different samples: drought-affected farmers (N=500) and Australian adults (N=1,203; for more detail, see Appendix A in Additional file 1). Variables had 0–2 % missing data with the exception of household income and satisfaction with religion/spirituality (approximately 15-20% missing data in both samples). To avoid loss of power, missing data were imputed using Expectation Maximisation 32 ; results for household income and satisfaction with religion/spirituality were treated with caution.

Descriptive statistics were generated for both samples and regressions adjusted for demographic characteristics compared cross-sample mean differences. Then, separately for farmer and general community samples, we undertook: (i) unadjusted correlations to establish bivariate relationships between satisfaction domains; (ii) multiple regression analyses to assess the level of unique versus shared variance in psychological satisfaction explained by different domains; (iii) exploratory factor analysis with maximum likelihood factor extraction and oblimin rotation to explore the presence of correlated supra-domains; and (iv) multiple regression analysis to replicate mediation pathways identified by structural equation modelling (SEM). Note that multi-group analyses of the SEM models indicated that the factor weightings and/or conceptual structure of life satisfaction was different across the two populations, making separate modelling the appropriate method of analysis.

SEM was performed using asymptotic distribution-free analysis (to allow for non-normally distributed data) and the following fit statistics: Chi-squared test, CFI score and RMSEA score. Note that RMSEA <=.05 was the core criterion used to assess models because it is less sensitive to sample size than the Chi-squared test and less sensitive to parameter number than CFI. SEM incorporates multiple techniques, including one-factor congeneric modelling (to confirm the factor structure of individual latent, or ‘unitary’, constructs), confirmatory factor analysis (to confirm relationships among latent constructs) and structural modelling (to show how multiple latent constructs may be causally related). We used these methods to test: (i) whether satisfaction was a unitary construct; (ii) whether supra-domains suggested by the exploratory factor analysis were unitary constructs; (iii) relationships between supra-domains; (iv) whether socio-economic variables could be helpfully summarised as latent constructs or should be used as separate variables; and then to (v) model multiple opposing pathways between supra-domains and overall psychological satisfaction, adjusting for socio-economic variables. In all modelling, non-significant items and pathways were deleted one-by-one and modification indices were used to add logically possible pathways until the final model reached best possible fit. Model fit was comprehensively re-evaluated after each deletion or modification and the Akaike Information Criterion was monitored to avoid over-fitting. Analyses were conducted using SPSS 17, AMOS (for SEM) and Stata 9IC (to adjust for data clustering in the regression models).

Table  1 shows descriptive statistics. Unlike farmers, about 60% of the general community sample lived in a large metropolitan city. Australians in the general community sample were more likely than farmers to have a university education and farmers were more likely to be male, married and in full-time work. A substantial proportion of the general community sample were not in the workforce (unlike farmers, who were selected by occupation) probably because most were aged 50+ and likely retired. Consequently, the general community sample somewhat over-represents older Australians. Farmers also tended to be mature-aged, which is consistent with the profile of Australian farmers; in 2006, the median age of farmers was 52 and more than two-fifths were over 55 34 . Although household income was similarly distributed across both samples, farmers clustered more tightly within a low-middle income bracket of $30,000–$79,999.

Note. Bolded means for life satisfaction indicate the mean is significantly higher than in the comparison sample in the next column, when controlling for demographic characteristics, p < .05.

^aMean value is higher than Australian norm, p < .05.

^bMean value is lower than Australian norm, p < .05.

The profile of satisfaction in the general community sample was very similar to 2008 Australian norms 35 , however the current community sample was more satisfied with standard of living and achieving in life and was less satisfied with religion/spirituality. The farmer sample differed from Australian norms in a way that was also consistent with the pattern of mean differences between farmers and the general community sample. While adjusted comparisons indicated that many of the mean differences were due to demographic characteristics, it was notable that farmers’ overall satisfaction was significantly lower than the general community sample even when adjusting for confounders (for more detail contact corresponding author).

Unadjusted correlations established that, for farmers and general community Australians, there were significant associations between all combinations of the eight domains, although some relationships were stronger than others (see Additional file 1, Appendix B, Tables B1—B2). However, for both samples, when psychological satisfaction was regressed on the eight domains, several domains failed to explain a significant amount of unique variance (Table  2). There was also a sizeable discrepancy (.15—.41) between the size of domain-overall correlations and domain-overall semi-partial correlations, indicating a high level of covariance between different domain-overall satisfaction relationships. These results gave preliminary support for hypothesis 1; that different domains would cohere into overall psychological life satisfaction, with some domains being more influential.

*p ≤.05, **p ≤.01, ***p ≤.001.

Note. These regressions are unadjusted for demographic characteristics; their purpose is to examine covariance between life satisfaction domains. The semi-partial coefficients (sr) report the unique variance that each predictor shares with overall life satisfaction, adjusting for covariance between different predictors.

To quantify how co-varying domains might cohere into a unitary construct of overall psychological satisfaction, we built one-factor congeneric models (Figure  1). Within both samples, standard of living, achieving in life and future security loaded strongly onto a unitary construct. Supporting hypothesis 1, adequate fit statistics were obtained for the general community sample (χ ²=49.19,p<.001,CFI=.90,RMSEA=.04), and for the farmer sample (χ ²=40.84,p=.004,CFI=.81,RMSEA=.05).

We then used exploratory factor analysis to investigate the presence of supra-domains (see Additional file 1, Appendix B, Table B3). The farmer sample returned a two-factor solution consistent with the hypothesised supra-domains of connectedness and effectiveness. The general community sample initially returned a single factor solution consistent with Figure  1b. However, when a 2-factor solution was requested, the returned factors closely resembled those found within the farmer sample. One-factor congeneric models were next used to confirm supra-domain structure; supra-domains were then correlated to produce the models shown in Figure  2.

For farmers, the connectedness supra-domain comprised satisfaction with relationships, community connectedness, religion/spirituality, safety and health. It was strongly positively correlated with the efficacy supra-domain, which comprised satisfaction with future security, standard of living and achieving in life. Strong model fit statistics were obtained, χ ²=19.14,p=.45,CFI=1.00,RMSEA=.004.

Supra-domain composition for general community Australians was very similar to that for farmers, except that satisfaction with health and with safety loaded onto efficacy, rather than onto connectedness. Again, the supra-domains were strongly positively correlated and the model fit the data well, χ ²=41.77,p<.001, CFI=.92,RMSEA=.04. Notably, these general community fit indices were slightly better for the two-factor model than the unitary model, and the farmer fit indices were markedly better.

To examine relationships between connectedness, efficacy and psychological life satisfaction, using SEM, we concurrently modelled reverse pathways between the supra-domains, and between each supra-domain and the measured overall satisfaction item (i.e. directional arrows were initially tested in both directions). Socioeconomic variables were included to control for their likely relationship with the supra-domains.

The final models for farmers and general community Australians produced identical underlying relationships (Figure  3). For both groups, connectedness predicted efficacy, which predicted overall satisfaction. Other pathways between these three became non-significant when modelled concurrently, despite connectedness having a strong bivariate association with the overall satisfaction item. To confirm the plausibility of the mediation pathway, we used factor weights derived from the SEM to construct weighted composite scores for each supra-domain which we regressed on overall satisfaction to test for the mediation pathways shown in Figure  3. Although the composite scores produced partial (~50%) rather than full mediation, the replicated effects were similarly strong and significant (see Additional file 1, Appendix B, Table B4). It is worth noting that the regressions also showed no evidence of multi-colinearity, supporting the conceptually meaningful nature of the mediation.

For both samples, the final SEM models were less parsimonious than those presented in Figure  2, which resulted in poorer, though still acceptable, model fit (farmers: χ ²=195.06,p<.001,CFI=.88,RMSEA=.05, general community: χ ²=79.37, p<.01,CFI=.87,RMSEA=.03). Consistent with other population health research, in the general community sample, education, paid employment and household income clustered together to form a single construct 36 37 , which positively predicted efficacy supra-domain satisfaction. In contrast, farmers’ efficacy satisfaction was simply predicted by household income. For both groups, being married predicted greater satisfaction with relationships.

Although the relationships reported here are consistent with existing theory, they may be limited to the Australian samples examined; as the results showed, contextual effects do occur. Consequently, establishing the findings’ generaliseability to other countries would require similar research utilising a range of representative samples. It would also be beneficial to examine the modelled relationships longitudinally and test the models’ predictive power across time. These issues are unavoidable limitations of the current work but are also important directions for future cross-contextual research on the structure of life satisfaction and wellbeing.