Physical health in a Canadian Old Order Mennonite community

Introduction: This article explores physical health and its determinants in two rural populations in Waterloo, Canada: Old Order Mennonites (OOMs) and non-OOM farmers. OOMs were selected because their distinct lifestyle might offer health benefits, and cultural homogeneity and isolation might more clearly expose the determinants shaping their health. Comparing the two Waterloo groups reduces the effect of contextual features impacting both, such as local economic conditions. The study considers a comprehensive list of determinants in order to evaluate their relative importance in shaping physical health. This information enables policy action to focus on the determinants having the greatest impact. Methods: A survey was used to obtain information from both groups on health status and health determinants. The survey was distributed in spring-summer 2010. All members of both groups were invited to complete the survey anonymously. The physical component summary (PCS) score of the SF-12 survey was used to measure physical health status. Age-gender breakdowns of PCS scores for both groups were compared, and differences evaluated using statistical significance and the interpretation cut-off recommended by SF-12 developers. Multiple (ordinary least squares) regression was used to identify key determinants shaping health. In the regressions, PCS scores represented the (continuous) dependent variable and the determinants of health were the independent variables. Results: Non-OOMs were found to experience better physical health than OOMs, with mean PCS scores of 49.24 for non-OOMs versus 47.39 for OOMs. The difference in PCS scores (1.85) was statistically significant (p=.002) and above the interpretation cutoff. While PCS score differences were significant for both genders, differences among the women were larger. OOM men and women may face health risks due to low incomes, offspring out-migrations and health service usage. OOM women may face


Introduction
Considerable research has been undertaken exploring the health impact of social and physical environments. These characteristics, or health determinants, have been prominent in Canadian policy discourse since the 1970s. The Public Health Agency of Canada (PHAC) currently recognizes 12 such determinants: (i) income and social status; (ii) social support networks; (iii) education and literacy; (iv) employment/working conditions; (v) social environments; (vi) physical environments; (vii) personal health practices and coping skills; (viii) healthy child development; (ix) biology and genetic endowment; (x) health services; (xi) gender; and (xii) culture 1 . There are strong parallels between Canada's heath determinants and those of other developed nations. The WHO Commission on Social Determinants of Health (CSDH) recognizes a similar list, contextualized within a framework indicating interactions within and across determinants (Fig1). As such, intermediary determinants directly influence health, but are, in turn, shaped by broader factors representing socio-economic and political contexts 2,3 .
It remains unclear whether there are unique determinants underpinning rural health or whether a generic, more broadly applicable set of determinants is distributed differently in rural settings 4 . This uncertainty reflects the lack of rural health research, but may also result from the standard approach of comparing rural and urban populations where the many social/physical environmental differences make it difficult to pinpoint the most influential health determinants. Significant socio-demographic diversity and health status variation exists even within rural settings, suggesting that it is important to restrict the geographical unit so that internal diversity is unmasked and results can be translated into action 4 . An action lens has been present in health determinants discussions, with research aimed at practical, economically efficient solutions remaining a priority. As such, information on the relative importance of the determinants can be beneficial. However, most determinants research focuses on a subset of determinants, thus their relative importance is largely unknown 5 . This study addresses these research problems by comparing physical health status and its determinants in two farming populations that live in the same location. This approach reduces the number of factors responsible for health differences by eliminating many shared contextual determinants common to both groups. By focusing on two rural populations, this study may also offer unique insights into the health determinants of rural communities. A comprehensive list of determinants is included in the analysis, so that their relative importance can be assessed and policy actions can be designed that focus on those having the greatest health impact.
An additional feature of this study is its focus on a unique rural population -the Old Order Mennonites (OOMs) of Waterloo, Ontario (Canada). OOMs are farmers and key features of their lifestyle include no smoking, low/no alcohol consumption, high religiosity (Christian), strong family and community support, high levels of social interaction, and minimal reliance on technology 6 . Health benefits have been linked to many features of their lifestyle. Moreover, the OOMs lifestyle has remained relatively stable and culturally isolated for generations. Studying isolated populations like OOMs is advantageous because distinct lifestyle practices may expose health benefits or risks (determinants) less easily identified in larger populations 7,8 . We hypothesize better physical health in OOMs compared to non-OOM farmers because of the health benefits of aspects of their lifestyle, such as high levels of religiosity, social capital, social support and sense of community.

OOM and rural health, social determinants of health links
The OOM community is closed with negligible in-migration, increasing the likelihood of population bottlenecks combined with genetic drift, inbreeding, and thus genetic diseases 9 . Genetic studies of the OOM Waterloo lineage have identified a number of physical health disorders that are relatively rare due to a surprisingly high genetic diversity 9 . Furthermore, Fretz 10 found a broad-based discouragement of close marriages and no evidence of higher rates of mental illness in Waterloo OOMs than in the general population. The work, although dated and lacking statistical validity, is nonetheless consistent with broader genetic research on OOM mental health 10 .
Studies examining the health-lifestyle linkage have found that Waterloo OOM and Old Order Amish (OOA) children demonstrate higher physical fitness levels than non-OOM/OOA urban and rural children [11][12][13] . Most other (nongenetic) health information comes from US studies of OOA, and indicates differences largely in favour of Old Orders for death rate and life span 14 , women's mental and reproductive health 15 , risk of cardiovascular disease 16 , certain cancers [17][18][19] and Type 2 diabetes 20 .
The two populations in this study live in the same region but occupy different social environments. The OOMs deliberately separate themselves from the larger society, yet within their community they are highly supportive, cohesive, trusting and spiritual 10 . These are features thought to contribute to better health. Beyond age, one of the most consistent findings is between poor health and low socioeconomic status (SES), one that holds across many populations, diseases, and SES indicators 21, 22 . While OOMs have comparatively low incomes, they cannot be considered low SES in the typical sense, since wealth is embedded in property and other agricultural assets. Nor does their nonparticipation in provincial health insurance limit access to health care services, since the OOM mutual aid program assists households unable to pay their medical bills.
Health behaviours such as smoking and physical inactivity are linked to poor health or intermediate outcomes (eg obesity), coping significantly impacts health, and life-course studies highlight the importance of healthy childhood development in fostering a lifelong advantage 5,[23][24][25] . Little is known about coping in OOMs, and health behaviours appear mixed with no/low smoking and alcohol consumption offset by a potential lack of concern about diet and physical fitness (especially among OOM women). OOM families are known for being strong and stable 10 , providing a solid foundation for healthy childhood development.
Social support -a strong factor in OOM communities -is considered an important determinant of physical health, with studies finding higher mortality or morbidity rates among people with fewest social connections [26][27][28][29] . Evidence for societal-level social support is inconsistent, with some studies finding poorer health in areas of high social disadvantage 30,31 , and others finding better health and/or resource access 32,33 . Social capital studies also explore the impact of social environments, with the main elements including social networks, participation, reciprocity and trust 34 . A systematic review found consistent evidence linking social capital with physical health, especially self-rated health 35 . Some studies suggest that societal-level social capital effects may be weak once individual-level factors are accounted for [36][37][38] .
Religiosity is one of the most prominent cultural features of OOM. Most studies focus on Jewish and Christian faiths 39 , with considerable evidence linking religion with physical health benefits 40-42 . However, reviews cite limitations including difficulties in measuring religiosity, small 'convenience' samples, treating correlation as causation, separating religious effects from non-sacred ones like social support, and inappropriate control groups [43][44][45] . Spirituality is also increasingly recognized as important in health research 46 . Evidence suggests that spirituality is more difficult to measure than religiosity because it is comparatively abstract and internal 47 . This means research examining spirituality should employ measures other than church attendance, and explore whether highly spiritual people (who may infrequently attend church) experience health benefits.
Rurality is a feature of both study groups. There have been a number of recent rural health studies within developed countries. The comprehensive review of Smith et al 4 .
concluded that much variation exists in both urban-rural and intra-rural health differentials within and between countries.
In Canada, life expectancy decreases as rurality increases, but is significant only in men 48 . This suggests higher mortality rates among rural Canadians, although studies on US, UK and Dutch populations report lower rates in rural residents 33 . Asthma and certain respiratory diseases are lower in rural Canada, yet certain cancers, cardiovascular disease and obesity are higher 48 . This variability is common to most developed countries 4 . Farming locations and areas with high pesticide exposure may be at increased risk for cancer 49 . However, CIHI 48 and Smith et al 4 . conclude that rurality per se does not translate into health disadvantage, but instead is a proxy for geographically dispersed determinants including personal behaviour and socio-economic factors.

Research setting, design
Both groups involved in the study reside in the Wellesley, Woolwich and Wilmot Townships of Waterloo, Ontario (Fig2). The Waterloo Region ranks second in Ontario in agricultural production 50 , and the majority of the members in both groups are farmers. The two groups are compared with respect to physical health status, and the Social Determinants of Health (SDOH) for each group are compared in an effort to explain health differences. Since both groups are mainly farmers living in the same location, determinants such as occupation and physical environment are unlikely to explain health differences.
A cross-sectional survey captured data on physical health status and the SDOH. Early in the study design the challenges of accessing the closed OOM community had to be addressed. Consequently, the article's first author spent 1½ years meeting with OOMs or people knowledgeable about them, and regularly observing OOMs in everyday interactions such as shopping, working in local shops and farming. This built trust within the community and acceptance of the project's utility.
OOM study participants were recruited through the churches. The senior OOM Bishop prepared a support letter to accompany the survey package, and arranged for the deacons to hand deliver the survey packages to all adults after the spring 2010 church services. Anonymity was assured by providing OOMs with a selfaddressed, postage-paid envelope for mailing back the completed survey. 1200 OOM surveys (60% response) were received, and 1171 were sufficiently completed for use in the analyses. The OOM sample was reduced to 850 in the following analyses, by eliminating those under the minimum age (28) of the non-OOMs. This was done in an effort to age-standardize the two groups.
Municipal tax rolls were used to identify non-OOM farmers. Directories of Mennonite and Amish groups were used to eliminate members of these groups from tax roll farmers, to avoid control group contamination. The survey package was mailed to all remaining tax roll farmers, with approximately 800 non-Mennonite (or non-Amish) households receiving the mailed survey. 344 completed surveys were received (43% response ) from non-OOMs.
The survey for both groups consisted of identical questions. It was piloted with a small number of OOM church leaders and community members, with feedback being incorporated into the final version.

Health measure
The physical component summary (PCS) score of the SF-12 health survey was used to measure physical health status because of its brevity and well-established psychometric properties 51 . The SF-12 measures six physical health functional domains: general health perceptions (GH), energy and vitality (VT), physical health impacts to social functioning (SF), physical functioning (PF), physical role limitations (RP), and bodily pain (BP). An algorithm scores the functional domains, standardizing them to a mean of 50 and standard deviation of 10. Higher PCS scores indicate better physical health.

Figure 2: Study Location -Wellesley, Woolwich and Wilmot Townships of Waterloo, Ontario, Canada
The SF-12 has been shown to be reliable in measuring health in many populations and clinical groups 52,53 . Reliability/validity tests designed for the SF-12 54-56 were conducted. For both groups, the instrument met the validity criteria, principal components analysis confirmed the twofactor structure, and known group tests confirmed expected relationships between demographic and health-related variables 57 .

Social determinants of health measures
There were practical restrictions on how determinants could be portrayed. Multiple measures were included in the survey for many SDOH because of their multidimensional nature and to provide alternate measures if significant non-responses were encountered. Various sources were consulted to guide selection of measures, question wording, and response options (Table 1).

Statistical analyses
The statistical software SAS v9.2 was used for all statistical analyses (http://www.sas.com). The PCS scores were calculated using the original (orthogonal) scoring algorithm and employing US population norms 51,53,67 . Hopman et al 68,69 confirm the validity of US-based norms in scoring Canadian applications of the SF-36 (the larger survey upon which the SF-12 is based). The PCS and SDOH measure distributions for the two groups were compared. Multivariate analyses (ordinary least squares regression) were conducted for both groups, with PCS as the (continuous) dependent variable and the SDOH measures as independents. All regressions were restricted to working with the same SDOH measures to ensure comparability between the two groups (rather than maximizing explanatory power using a stepwise procedure to select the variables forming the optimal model). In this way, the degree to which SDOH measures were significant in shaping physical health could be determined, given the presence of the same co-measures.

Ethics approval
Ethics approval was obtained from McMaster University Research Ethics Board, #2009-187.

Social determinants of health measure distribution
Compared to non-OOMs, the full OOM sample (n=1,171) is younger (mean age 43.4 versus 57.7) and has more females (58% versus 51%) and singles (33% versus 5%). Sample differences reflect differences in the recruitment efforts for the two groups. For example, church recruitment for the OOMs captured many singles living on their parents' farm whereas municipal tax rolls for non-OOMs captured people owning their own farm. Sample differences also reflect natural population characteristics, since the OOM population is younger with more females compared to the Ontario population 70 . Table 2 provides the distribution of the SDOH measures used in the regression analyses, and shows that the two groups differ significantly on most SDOH. Some SDOH were excluded from the regressions, such as Education and Literacy because educational attainment did not vary in OOMs, Physical Environment because of high colinearity with other measures or absence of a significant health relationship, and Health Service Use since virtually all respondents (both groups) reported having family doctor access. Also excluded from the regressions were traditional health behaviours such as smoking and alcohol consumption, because no OOMs reported either. Employment type was excluded because the majority of members of both groups were farmers. Regarding employment status, more non-OOMs were unemployed than OOMs (28.5% of non-OOMs versus 10.8% of OOMs).
Since the majority of the unemployed (over 90%) in both groups indicated that retirement was the reason for unemployment, employment status was highly correlated with age and thus excluded from the regressions.
The groups did not differ on income adequacy or degree of job control, with most participants reporting no trouble meeting basic needs and high job control levels. Most members of both groups were married, with the OOMs having more singles. The OOMs assign more importance to the socially oriented sense-of-place measures -rootedness and community -and less to the physical environment. For social capital, the OOMs report lower levels of participation and higher levels of trust and reciprocity. OOMs rarely join formal organizations, yet regularly participate within their community, suggesting that social interaction may better capture participation levels. More social interaction in OOMs is evident in the higher SNI and perceived social support scores. OOMs report more difficulty coping but less stress, which seems counterintuitive, although the stress question may not have captured the full response range or asked about stressors most common in OOMs. OOMs report fewer dietary concerns and childhood diseases. OOMs are shorter (p<0.001 overall, each gender), with women`s weight being similar to non-OOM women and men`s being less than non-OOM men. Compared to non-OOMs, BMI is higher in OOM women (p<0.001) and similar in OOM men. OOMs also report significantly higher spirituality levels.

Health status
Mean PCS scores are higher (p<0.01) in non-OOMs than OOMs, indicating better physical health (Table 3). This is particularly influenced by differences in women, since physical health in non-OOM women is better than in OOM women (p<0.01) and the men's health is similar (p=0. 16).
There is no gender difference within non-OOMs (p=0. 47), yet within OOMs women have lower PCS scores (p<0.01). All statistically significant (p=0.05) PCS score differences are also of potential clinical significance since they exceed onethe minimum (cut-off) for interpretation 67,71 . Potential clinical significance means the difference justifies further investigation, as it may reflect substantive differences in underlying causal mechanism(s). The difference in men's PCS scores, although not statistically significant, may be clinically significant given that it (slightly) exceeds one. PCS scores in both groups are negatively skewed, as in other SF-12 general population studies 56 . The kurtosis coefficient is a measure of how peaked a distribution is, with lower values indicating a flatter (more spread out) distribution. The OOM PCS distribution is flatter than for non-OOMs, indicating wider variation in physical health status within OOMs.
Examining PCS scores by age and gender provides further insight into group differences and patterns. As expected, physical health declines with increasing age (Figs3,4) Key Social determinants of health shaping health Table 4 presents the regression analysis results, with adjusted R-square values of 0.39 and 0.29 for the OOM and non-OOM models respectively. Four SDOH measures were highly significant (p≤0.01) in both regression modelscoping, BMI, age and number of childhood disorders. Stronger coping skills were associated with better physical health, and higher BMI, increasing age and number of childhood diseases were associated with poorer physical health. The significance of age in both models reflects earlier findings in which PCS scores decreased with increasing age in both groups (Figs3,4).
Some SDOH were associated with only one group. Stress was significantly and negatively associated with health in non-OOMs. Within OOMs, increased physical health was associated with decreased spirituality, increased income adequacy and increased reciprocity (given). Decreased physical health in OOMs was associated with reciprocity (received) and being female.
Gender in the OOM model, and its absence in the non-OOM model, is consistent with earlier results showing a gender difference for OOMs only (Table 3).

Discussion
The purpose of this study was to determine whether physical health was better in OOMs, and to identify the key determinants shaping it. Physical health in OOMs was found to be lower than in non-OOMs. Since both groups live in the same location, the individual and cultural characteristics discussed below are among the main factors responsible for this difference.

Individual and cultural determinants of physical health
Physical health in both groups is shaped by age, coping, number of childhood disorders and BMI. Age is linked to declining physical health in all populations, and other studies have found that coping significantly influences physical and mental health 5,24,72,73 . OOMs report more difficulty coping, despite low stress levels and the high self-efficacy often seen in Old Orders 74 . Since the survey may have failed to identify common OOM stressors, chronic exposure to stressors may still underlie their coping difficulties, potentially resulting in increased blood cortisol levels and ultimately cardiovascular disease and other illness 75,76 . Many life-course studies support the association between number of childhood disorders and physical health 77,78 . Childhood illness can represent a lifelong threat, predisposing adults to related conditions or weakening their immune system and increasing general illness susceptibility. Psychosocial effects may also be present, since childhood illness shapes OOM mental health 72 . BMI is associated with poorer physical health, with studies linking BMI to diabetes, cardiovascular disease and premature mortality [79][80][81][82] . Exercise and weight management are critical for controlling BMI and reducing chronic illness risk 79 . BMI does not shape mental health in either group 72 , suggesting minimal psychosocial effects.
Some determinants shape physical health in only one group. Higher stress is linked to poorer health in non-OOMs, and they report higher stress (Table 2). Studies suggest that stress can negatively impact physical health, particularly when it is chronic and co-exists with a lack of control or low social interaction levels 58 . While non-OOM job control levels are high, they may feel a lack of control over broader conditions impacting farmers (eg economic or climate conditions) and they report less social interaction. As discussed above, psychosocial stressors can lead to elevated blood cortisol levels, potentially causing other health conditions.    Within OOMs, income adequacy, reciprocity (given and received), gender and spirituality are associated with physical health. Health improves with income adequacy, with strong support in the literature for this association 22 . Income may significantly shape OOM physical health because of lower incomes, high parity, refusal of government support or high self-reliance. Survey results indicate lower OOM household incomes and larger families. Also, OOMs refuse government assistance such as old age security. Despite a strong mutual aid system providing economic support for families, OOMs are taught self-reliance as a basic virtue 10 . Reluctance to utilize broader safety nets means some OOMs may lack the resources to meet family needs. Absence of a direct incomehealth effect in non-OOMs may reflect higher incomes, access to government support, or the existence of indirect effects 83 . Poetz et al 24 . and the OOM mental health results 72 show an indirect income-health effect mediated by coping, but this could not be tested in non-OOMs because the sample is below 500 84 . Regarding the mechanism underlying the income-physical health relation in OOMs, material conditions are suspected and are, in part, socially produced. However, psychosocial mechanisms cannot be ruled out, since an indirect income-mental health link in OOMs was observed, and some researchers suggest that humans, as social animals, will always attach some psychosocial meaning to material resources 85 .
Reciprocity received and given are negatively and positively associated with physical health in OOMs. The associations may reflect psychosocial effects such as satisfaction from helping others and stress from receiving help due to feeling indebted, being a burden or losing independence 86 . However, neither form of reciprocity significantly impacts OOM mental health 72 , suggesting psychosocial effects are minimal.
Reciprocity-physical health relations more likely reflect health status, where more help is given and received by people with better and worse health. Reciprocity in the OOM model is consistent with the high levels of help they give and receive (Table 2), and originates from 'a sense of community matched by none' (p186) 10 .
Gender appears in the OOM model, with men's physical health exceeding that of women. While many SF-12 studies find lower PCS scores among women, OOM women's scores are also below those of non-OOM women, suggesting they face unique risks. There is little evidence of psychosocial effects, since OOM women's mental health is significantly better than that of non-OOM women 72 . The gender-health effect remains after BMI is included in the model, suggesting risk factors beyond diet and weight management. Perhaps large families are an underlying risk to women: frequent pregnancies are physically demanding, leave less time for recovery, and tie women to the home with cooking, cleaning and caring for children. While Old Order women report lower reproductive-related stress 15 , frequent pregnancies may nevertheless carry a physical toll. Moreover, OOMs may limit or delay necessary reproductive-related health services. Therefore, the mechanism underlying the gender-physical health relation may be linked to reproduction, and rooted in broader societal norms that encourage large families and women to be 'keepers at home' (p109) 87 .
Spirituality appears in the OOM model, with less spiritual people having better physical health. While studies often report positive associations with physical health, most also pertain to religiosity (not spirituality) and have methodological limitations 43 . This study`s cross-sectional nature does not allow determination of directionality, thus health status may drive the spirituality-physical health relation. This interpretation is supported by studies showing that people turn to religion/spirituality to cope with poor health 88 . Interestingly, spirituality is positively associated with OOM mental health 72 . This dual role -protective for mental health and a resource for those with poor health -has been observed in clinical groups such as those with chronic pain 88 . While OOMs are not a clinical group, their lower physical health status might generate results similar to those of a clinical (or sub-clinical) group. This is speculation, however, with more research needed to clarify the costs and benefits of religion/spirituality in various populations.

Summarizing physical health and its determinants
In considering the study's overall results, one might ask: Why do social factors not shape physical health? Health in both groups is influenced more by traditional determinants such as age and income, which is particularly unexpected in OOMs given the strength of many social factors ( Table 2). What might explain this? First, results are sensitive to the health measure used, as evidenced by the finding that self-rated health shows the most consistent linkages with social capital 35 . Manderbacka 89 found that self-rated health reflects physical health, yet Huisman and Deeg 90 suggest it may be mental or physical health depending on cognitive/filtering processes and assessment timing. Self-rated health is not equivalent to this study's measure of physical health (PCS), as it is one of the six functional domains used to derive it. Second, other studies have failed to demonstrate a relationship between physical health and either social capital 91,92 or sense-of-place 59 . Third, many studies focus on a subset of determinants, which will produce different results from those including a more comprehensive list. Fourth, perhaps the rural or farming status of this study's populations affects the relevance of the social factors. An Australian study on rural populations found that social relations and community support had less impact on mental health in farmers than non-farmers, and the researchers suggest this may reflect higher self-reliance among farmers 93 .
Another question arises when comparing regression model intercepts: does the low intercept in the OOM model signal other risk factors not explicitly recognized? While the OOM model has a reasonable R-square (0.39) and includes a number of determinants, the intercept is well below the non-OOM's, significantly depressing OOM PCS scores. Why? The SF-12 instrument met the validity/reliability criteria 57 , and underlying psychosocial effects appear minimal since virtually all members report no discrimination, and OOM mental health compared to that of non-OOMs is equal for men and higher for women 72 . Employment status was excluded from the regressions (due to high colinearity with age), yet it can influence health. However, the models do not change significantly when employment status is included (results available from first author). Lower OOM PCS scores may reflect genetic disorders (not measured in this study) or aspects of the OOM lifestyle that are difficult to measure. For example, families are large, with OOM fertility rates more than double those of non-OOMs. Beyond the physical and economic burden of a large family, the need for offspring to acquire affordable farmland has resulted in one-third of the population leaving Waterloo over the past two decades. This has disrupted the family farm and reduced contact with family and friends, which may impact health directly or do so through coping (eg coping is similar in OOM men and women but better in non-OOMs for both genders). The overall impact of a large family is difficult to determine, however, as some results suggest a positive health impact; for example, when number of children is included in the regression, it significantly (p=0.03) and positively shapes health (OOMs only).
OOMs also differ in health service usage, although the health impacts are unclear. Virtually all OOMs report access to a family physician, so this often-used measure was not employed in this study. The challenge is in incorporating other health service information into a meaningful measurefor example OOMs report fewer visits to hospitals and doctors but more visits to community clinics, and significantly higher usage of alternative services. Local service providers indicate that OOMs often present with more advanced symptoms, attributing this to delays in seeking health services (pers. comm., Waterloo Public Health, 2011). This is also seen in the Amish 74 , unsurprising since both groups lack public health insurance.

Study limitations
Several study limitations should be acknowledged. First, responses rely on self-report and interpretation, and are cross-sectional only, although these same limitations are consistently found in most large population studies. Second, results may be limited by the ways in which various determinants were measured. Since all the determinants were being measured, there were restrictions on the number of measures that could be included in the survey. Third, the non-OOM sample size (344) is below the 500 recommended by the SF-12 developers (for consistency with SF-36 results), with the small sample size potentially contributing to the lack of significance among variables in the non-OOM regression. However, tests conducted for the non-OOMs indicate that the instrument shows acceptable internal consistency, distinguishes between socio-demographic classes of respondents in the expected manner, and exceeds the small convenience samples often used in SF-12 validity studies 55 . Finally, OLS multiple regressions assume normality, yet a few variables show evidence of non-normality. However, since these variables are negatively skewed, commonly employed data transformations will be ineffective in normalizing them 94 .

Conclusions
Despite the limitations, this study highlights important avenues for research and policy action. It indicates the complexity of determinants research, showing that the key determinants depend on populations, health outcomes, and measures included in the analysis. For example, the health outcome can influence directionality, with spirituality and age being negatively associated with physical health and positively associated with mental health. The measures used for determinants and the range of determinants included in the analysis also impact the results. For example, this study found that reciprocity as a single measure (help received and help given combined) was not significant in shaping health. However, when it was broken down into two measures, both were significant and their direction of association with health differed. This study also found that social capital and other social factors are less important than traditional determinants in shaping physical health, yet they appear to shape mental health 72 . This may be due to inherent differences between physical and mental health and/or unique characteristics of rural populations or farmers. Ultimately, the implication of these findings is that there needs to be increased awareness of the various health outcomes, determinants, and measures used to represent them. A targeted approach to research is then required that involves careful specification of the population, health outcome of interest and determinant measures to ensure that policy actions based on the study's findings are effective and achieve desired results.
This study also supports policy action on the determinants that significantly shape health in both groups: coping, number of childhood disorders, BMI and age. These determinants have been found to influence health in many urban populations and people whose life circumstances vary widely, suggesting they transcend the boundaries of OOMs, farmers and rural populations. As such, actions focused on these determinants may offer broad impact across many populations. Approaches could be designed that address the individual, structural and temporal underpinnings of these determinants. For example, psychosocial concerns could be addressed with services that help individuals cope and reduce stress, monitor/control their weight, and maintain their family's health. Structural concerns could be tackled by investing in community resources that promote healthy lifestyles and alleviate broader economic/social conditions challenging peoples' ability to cope or maintain health. Concerns about the perpetuation of disadvantage through the life-course could be addressed by ensuring that programs are integrated and sustainable over the life-course, emphasizing the entire age spectrum from children's health (eg immunization programs, reducing childhood obesity) to healthy aging and age-friendly built environment programs.