Social Science & Medicine 54 (2002) 849–868

Social capital and health (plus wealth, income inequality and regional health governance) Gerry Veenstra Department of Anthropology and Sociology, Centre for Health Services and Policy Research, University of British Columbia, 6303 N W. Marine Drive Vancouver, BC, Canada V6 T 1Z1

Abstract This article describes an empirical exploration of relationships among aspects of thirty health districts in Saskatchewan, Canada. These aspects include social capital, income inequality, wealth, governance by regional health authorities and population health, the primary dependent variable. The social capital index incorporated associational and civic participation, average and median household incomes served as proxies for wealth, the degree of skew in the distribution of household incomes assessed income inequality while the model for effective governance by District Health Boards (DHBs) focused on reflection of health needs, policy making and implementation, fiscal responsibility and the integration and co-ordination of services. I found no evidence of a relationship between social capital in health districts and the performance of DHBs. Among the determinants of health, wealth appeared unrelated to age-standardised mortality rates while income inequality was positively and social capital was negatively related to mortality. Income inequality was not as strongly related to age-standardised mortality after controlling for social capital, and vice versa, suggesting the two may be comingled somehow when it comes to population health, although they were not significantly related to one another. Of the predictors of social capital the distribution of age in districts appeared to be the most salient; of the predictors of age-standardised mortality rates the gender composition of a district was most salient. # 2002 Elsevier Science Ltd. All rights reserved. Keywords: Social capital; Income inequality; Regional health governance; Canada

Introduction In Unhealthy Societies: The Afflictions of Inequality Wilkinson (1996) explored social and other determinants of health among Western nations. Even though income and health are usually found to be strongly related at the individual level, he found that among the developed market economies of the OECD the Gross Domestic Product per capita was only weakly related to life expectancy}wealthier countries were not necessarily much healthier ones. The degree of income inequality, the nature and width of the gap between the rich and the poor, on the other hand, was strongly related to health among

nations.1 As a result Wilkinson suggests that relative income may be more important for peoples’ health than is absolute income. Several studies have additionally found that income inequality and various health indicators are related across settings smaller than the nation. For example, the relationship was found to be strong among

1 Among the nine nations of the Luxembourg Income Study, and including Canada, income inequality (calculated to be the proportion of overall wealth held by the poorest 70% of families) and life expectancy were strongly related (r ¼ 0:86).

0277-9536/02/$ - see front matter # 2002 Elsevier Science Ltd. All rights reserved. PII: S 0 2 7 7 - 9 5 3 6 ( 0 1 ) 0 0 0 4 9 - 1

850

G. Veenstra / Social Science & Medicine 54 (2002) 849–868

the American states (Kennedy, Kawachi, & ProthrowStith, 1996; Kaplan, Pamuk, Lynch, Cohen, & Balfour, 1996) and among American metropolitan areas (Lynch et al., 1998).2 In Canada, Ross et al. (2000) failed to find statistically significant relationships between income inequality and health among Canadian provinces and among Canadian metropolitan areas, however. They suggest this might result from a non-linear relationship between income inequality and mortality (as the Canadian provinces and cities are generally more equitable than American states and cities), or that the relationship depends on social and political characteristics specific to place. Lynch and Kaplan (1997) proffer several types of explanations for the relationship between income inequality and health. First, the individual level relationship between income and health may influence the ecological relationship, although this contribution is thought to be slight (supported empirically by Wolfson et al. (1999)). Second, ‘‘an inequitable income distribution may be associated with a set of social processes and policies that systematically underinvest in human, physical, health and social infrastructure, and this underinvestment may have health consequences’’ (p. 306). Contextual explanations of this kind look to ecological aspects of societies that are associated with income inequality}individual level ‘outcomes’ are further down the explanatory line. Third, ‘‘an inequitable income distribution may have direct consequences on peoples’ perceptions of their social environment that influence their health’’ (p. 306). The hypothesis that relative income is more important than is absolute income focuses upon individuals’ perceptions of their social standing relative to others and is of this form. Wilkinson (1996) and Lynch and Kaplan (1997) propose that societies with a high degree of income inequality are also ones with low social cohesion, and that income inequality affects health by means of this important social resource. Wilkinson describes social cohesion as the social nature of public life, ‘‘dominated by peoples’ involvement in the social, ethical and human 2 Kennedy et al. (1996) used the Robin Hood index, the proportion of aggregate income that must be redistributed from households above the mean and transferred to those below to achieve perfect equality in the distribution of household incomes, to measure degree of income inequality. The correlation between income inequality and all-cause mortality was strong (r ¼ 0:54). Similarly, Kaplan et al. (1996) found a correlation of r ¼ 0:62 between income inequality, measured by the share of aggregate income held by the poorest 50% of households, and age-adjusted total mortality rates. Among metropolitan areas Lynch et al. (1998) found a correlation of r ¼ 0:52 between income inequality and mortality and a correlation of r ¼ 0:21 between per capital income and mortality.

life of the society, rather than being abandoned to market values and transactions. People come together to pursue and contribute to broader, shared social purposes’’ (p. 136). Lynch and Kaplan (1997) describe social capital as ‘‘the stock of investments, resources and networks that produce social cohesion, trust and a willingness to engage in community activities’’ (p. 307). Supposedly, then, social spaces with many networks of participation and much social trust facilitate good health. Testing the hypothesis empirically, Kawachi, Kennedy, Lochner, and Prothrow-Stith (1997) found that social capital was strongly related to both income inequality and mortality among the American states.3 They conclude that social capital mediates the relationship between the other two: greater inequality leads to less participation in the public space and more mistrust which then influence health. To explain the potential influence of social capital on health, Kawachi and Berkman (2000) distinguish between compositional and contextual effects. It could be that areas low in social capital are composed of proportionately more socially isolated individuals, and given the relationship between social isolation and health such areas might then have poorer aggregated health status. If psycho-social attributes of individuals (e.g. self-esteem) are sometimes related to their health then perhaps the psycho-social attributes of social capital manifested in individuals (e.g. trust) translates into health outcomes directly. In one survey of citizens from Saskatchewan, Canada, few individual level relationships were found between self-rated health status and human-level notions of social capital such as social and political trust, civic participation or participation in secondary associations, however (Veenstra, 2000). On the other hand, using a sample of Canadians taken from the entire country, Lavis and Stoddart (1999) found that trust and membership in secondary associations were significantly related to good health in the World Values Survey. Beyond compositional effects at the individual level, Kawachi and Berkman (2000) propose contextual effects that are more in keeping with the structural nature of the social capital described by social theorists such as 3

Social capital was measured with three trust questions: ‘Most people would try to take advantage of you if they got the chance,’ ‘You can’t be too careful in dealing with people’ and ‘People mostly look out for themselves,’ along with the rate of participation in civic associations. The inter-item correlations were positive and significant. The correlation between income inequality, measured by the Robin Hood index, and all-cause mortality was r ¼ 0:65. The correlations between income inequality and per capita group membership and between income inequality and lack of social trust were r ¼ 0:40 and r ¼ 0:73, respectively. Social mistrust (r ¼ 0:77) and per capita group membership (r ¼ 0:49) were both strongly related to all-cause mortality.

G. Veenstra / Social Science & Medicine 54 (2002) 849–868

Coleman (1988).4 They suggest that social capital may influence health-related behaviours by promoting diffusion of health-related information, thus increasing the likelihood that healthy norms of behaviour are adopted. They also propose that socially cohesive communities are more successful at uniting to ensure access to services and amenities. Lavis and Stoddart (1999) suggest that social cohesion may influence the capacity of government to develop and implement policies (and re-distributive policies in particular), and that cohesive communities might better provide for their elderly and their homeless. Wilkinson (1996) proposes, and Kennedy, Kawachi, Prothrow-Stith, Lochner, and Gupta (1998) have confirmed empirically among the American states, that social capital may be related to the incidence of violent crime. These are some, but probably not all, of the ways in which social cohesion and/or social capital may influence the health of populations (see Veenstra (2001) for a fuller description of possible means by which social capital may influence the health of populations). Does income inequality indeed influence social capital? The direction of causality is a point of contention not easily resolved. Kawachi et al. (1997) propose that income inequality has causal priority whereas Wilkinson (1996) suggests that income inequality and social cohesion may be reciprocally related. Coburn (2000) suggests that both reflect the degree of adherence to neo-liberal tenets within societies while Muntaner and Lynch (1999) suggest both reflect the nature of class relations. The appropriate level of analysis is also an open and difficult question. It has not been adequately determined whether income inequality and/or social cohesion/capital predict variability in population health at the levels of region, city or neighbourhood in addition to or other than the levels of nation or state, and also whether they vary together across these levels. Lynch and Kaplan (1997) suggest that the relationship between income inequality and health would weaken as the level of geopolitical aggregation became more local. This makes sense given that neighbourhoods (in Canadian cities at least) tend to be efficient spatial sorting mechanisms along economic lines (Bunting, 1991; Bourne, 1997; MacLachlan & Sawada, 1997), meaning that income inequality in cities may reflect differences among neighbourhoods more than inequality within them. On the other hand, social capital as a social resource may well congregate, at least in part, within neighbourhoods. It has not been 4 ‘Social capital is defined by its function. It is not a single entity but a variety of different entities, with two elements in common: they all consist of some aspect of social structures, and they facilitate certain actions of actors}whether persons or corporate actors}within the structure’ (Coleman, 1988, p. S98).

851

determined whether the relationships hold true among rural areas, either. The income inequality and health relationship does not seem to manifest itself at the provincial or metropolitan level in Canada (Ross et al., 2000) but might at the regional or local level or among rural areas. As far as I know, the social capital and health relationship has not yet been explored at any level beyond the individual in Canada. In this article, then, I present results from an empirical investigation among thirty health districts in the province of Saskatchewan, Canada that explores the degree to which wealth, income inequality and social capital predict population health inequalities at the regional level. To additionally complicate matters, all but one of Canada’s provinces have recently devolved responsibility for health care decision-making from the provincial ministries of health to newly formed regional health authorities (Lomas, Woods, & Veenstra, 1997). When exploring the effects of inequitable income distributions and social capital upon health we might also incorporate consideration of these governing bodies, since presumably effective health authorities will positively influence population health status. This influence may be comingled with the effects of social capital, however. In their study of Italian regions ranging over 20 years, Putnam, Leonardi, and Nanetti (1993) found that those regional governments that performed their tasks effectively were also ones presiding over regions high in social capital. This finding was bolstered empirically by the Rice and Sumberg (1997) study of the American states. Using measures somewhat similar to those used by Putnam et al. (1993) they found a clear link between civicness5 (i.e. social capital) and governmental performance. ‘‘States that are more civic tend to have governments that enact more liberal and innovative policies. The relationship between civicness and performance remains strong even after controlling for political culture, ideology, education, and other factors’’ (p. 99). The work by Putnam et al. (1993) and Rice and Sumberg (1997) refer to comprehensive governments with jurisdiction over more than health. Veenstra and Lomas (1999) apply the perspective to health govern5 Civicness was measured using newspaper circulation as an indicator of concern for public matters; the number of books per capita in public libraries as an indicator of support for civic engagement; the number of community improvement and philanthropic groups per capita to indicate concern for the public good; the percentage of public school teachers who are men and the percentage of state legislators who are women, the number of civil rights groups per capita among the non-white population and a gauge of income distribution to indicate equality; the crime rate, number of lawyers per capita and the default rate on Perkins student loans to assess solidarity and trust; and finally a per capita composite index of 26 different types of non-profit organisations. All inter-item correlations were positive but for one, creating a rather coherent index.

852

G. Veenstra / Social Science & Medicine 54 (2002) 849–868

ance in particular, postulating a theoretical relationship between social capital in health regions and effective governance by corresponding regional health authorities (RHAs) in Canada. Three broad spheres of social capital activity are identified: namely, associationalism, civic participation and collaboration with others to solve community problems. The first and third can be conceptualised and measured at the community level (e.g. dense networks of association and institutionalised norms that facilitate collaborative action) and also at the individual level (e.g. participation in clubs and working with neighbours to better the community). Socialpsychological attributes of individuals that may contribute to the three spheres of social capital activity are identified; namely, trust, identity and commitment, social-psychological concepts that exist in reference to groups of people, or communities. Veenstra and Lomas (1999) hypothesise, following Putnam et al. (1993), that a community rich in social capital will have a positive relationship with the governing process and outcomes of the RHAs. Presumably a highly effective RHA will have a positive effect on the health of the region, making governance another member of the growing array of concepts available to explain variability in population health. The health districts of the southern portion of the Canadian province of Saskatchewan present an opportunity to explore many of the relationships described above. Wealth, income inequality and social capital can be evaluated at a level far smaller than those of nation or state/province and I can include both rural and urban regions. In addition, each region has a governing body responsible for making health-related decisions in the region whose performance as a political institution can be evaluated. The province of Saskatchewan has a population of approximately one million, and the health districts included in this study, shown in Fig. 1, range in size from about 11,000 to 220,000 people. They include among their number two urban districts and four midsized ones}the remainder of the thirty districts explored here can be considered rural. Because of the history of strong co-operative movements in the past (the province was the birthplace of the Cooperative Commonwealth Federation, precursor to the New Democrat Party of Canada) and given that Medicare in Canada was born here I thought Saskatchewan a particularly appropriate place to study relationships between social capital and health. As such this article explores the following research questions, always among the thirty districts but sometimes among the rural districts as well (to complement the work among Canadian cities conducted by Ross et al., 2000): First, following Wilkinson (1996), Kaplan et al. (1996) and Kennedy et al. (1996), is overall wealth

and/or the nature of the distribution of wealth (income inequality) related to population health status among the health districts? Second, following Wilkinson (1996) and Kawachi et al. (1997), is social capital related to population health, and, if the preceding relationship(s) between (the distribution of) wealth and health is empirically supported, does social capital fit within a line of causality including these other attributes of health districts? Third, following Putnam et al. (1993) and Veenstra and Lomas (1999), is social capital related to the performance of the District Health Boards in Saskatchewan, and, if the relationship between social capital and health holds true, does effective governance mediate this relationship? Fourth and finally, also incorporating additional socio-demographic characteristics of health districts, such as the incidence of crime and degree of mobility, for example, what conclusions can we draw about causality? Which aspects of health districts have the most salient influences upon their degree of social capital and the health of their populations? This research design is particularly interesting for two reasons. First, it is the only study I know of that incorporates all of wealth, income inequality, social capital, the effectiveness of regional health authorities and population health in one design. Second, given interest in the salience of social capital at the local level (i.e. neighbourhood, community, city or region rather than province, state or nation) this study broadens our knowledge about the salience of both social capital and income inequality for health at one such level.

Data Social capital Social capital is commonly thought to be composed of ‘trust, norms and networks.’ Wilkinson (1996) and Putnam et al. (1993) have focused particularly upon participation in the public space, both in voluntary secondary associations (meeting with others to pursue common interests) and through civic venues (to further the common good or participate in the political life of the community). Kawachi et al. (1997) and Veenstra and Lomas (1999) have paid particular attention to the importance of social trust that lubricates interactions among people and groups. I created a social capital index (a=0.841), described in the appendix, that measures aggregated associational and civic participation by individuals and the density of associational life. I did not measure trust, identity, commitment or community collaborative problem solving among the

G. Veenstra / Social Science & Medicine 54 (2002) 849–868

853

Fig. 1. Saskatchewan’s health districts.

health districts. Veenstra (1999) describes results from a survey of citizens from eight of Saskatchewan’s health districts that explored the inter-relatedness of the individual-level concepts described in the Veenstra and

Lomas (1999) model of social capital. Social and political trust formed coherent concepts and were strongly related to one another, participation in voluntary associations, civic participation and experi-

854

G. Veenstra / Social Science & Medicine 54 (2002) 849–868

ence collaborating with others were all positively related to one another, and social trust explains some variance for all three forms of activity. These results give some confidence that the social capital index described here tapped into at least some individual-level components of social capital. The fact that the one ecological item in the index, the proxy measure of associational density, was strongly and positively related to most of the other items suggests that the index did indeed measure more than individual-level phenomena. Governance by District Health Boards Veenstra and Lomas (1999) identified four dimensions of regional health authority effectiveness. In general, an RHA will score highly in the dimension ‘reflecting health needs’ if it accurately reflects health needs, preferences, and perceptions of health needs of residents. An RHA will score highly in the dimension ‘policy-making and implementation’ if it makes good decisions that are then effectively implemented. It will score highly in the dimension ‘fiscal responsibility’ if it operates and achieves goals within a constrained budget; passes the budget on time; allows the community to know the choices that have been made; and handles monies well and properly. Finally, an RHA will score highly in the dimension ‘integration and co-ordination’ if it identifies duplicated services and provides integrated links between duplicated services (pp. 7–9). My measures of the performance of the District Health Boards in Saskatchewan are described in the appendix. As some of the variables measuring performance were continuous and others categorical, because the intersection of cases left too few for meaningful analysis, and, most importantly, because the inter-item correlations among the continuous variables were not statistically significant (see the appendix), I have kept the governance effectiveness indicators separate in subsequent analysis. Even so, the lack of coherence among the measures implies that my attempts to measure board performance failed, possibly because the DHBs were too new to have differentiated one from another with respect to political performance at the time of evaluation (they were created in 1993 and evaluated in 1995–97). Population health status To measure population health status I obtained, for 1993, the mortality rate per 1000 residents (both crude and standardised for age in five-year age categories). Mortality rates describe the proportion of the populace who died in a given year and serve as an indirect indicator of overall levels of health. I also obtained the percentage of births in 1993 where the child weighed fewer than 2501 grams. For 1993/94 I obtained the number of district residents per 1000 who had received

community mental health services from Saskatchewan Health Regional Staff and the number per 1000 hospitalised in specialised psychiatric units. Similarly, from 1993/94 I included the number of inpatient days and outpatient admits per 1000 who utilised alcohol and drug services. Additional socio-demographic variables, including wealth and income inequality, are described in the appendix.

Analysis Throughout the following analyses I have utilised both Pearson’s correlation and Kendall’s t to describe bivariate relationships between continuous variables. Kendall’s t is a non-parametric measure of association that is more resistant to influential points than is Pearson’s r but less powerful at discerning relationships. I have presented both measures of association to indicate the possible presence of influential points in the absence of corresponding scatter plots and also to allow for comparison of the strengths of correlations (as the two forms of correlation cannot be compared with one another). Although tests of significance are not strictly necessary when working with a population rather than a sample from one, I have included the pvalues as a way of gauging the possibility that demonstrated relationships might have occurred by chance. Bivariate analysis: wealth, income inequality and health Even though income was significantly related to selfrated health status in the individual-level survey of citizens in eight districts (Veenstra, 2000), the average and median household incomes of the thirty health districts were not significantly related to any of the population health measures (Table 1 and Fig. 2).6 Among the twenty-six health districts which are least urban (excluding Saskatoon, Regina, Moose Jaw/ Thunder Creek and Prince Albert) the same result held true. The income data is from 1995, presented in the 1996 Census, and the health data is from 1993, however, throwing causal linkages under some suspicion. Even so, in this scenario, as in Wilkinson’s (1996) study of Western nations, the individual-level relationship between income and health does not seem to translate naturally into the corresponding ecological-level one, noting, however, that average or median household incomes are not necessarily adequate measures of a community’s wealth and that the nature of incomes and 6

The crude mortality rate was negatively but not quite significantly related to the median income of health districts (r ¼ 0:350, n ¼ 30, p ¼ 0:058; t¼ 0:235, p ¼ 0:069).

855

G. Veenstra / Social Science & Medicine 54 (2002) 849–868 Table 1 Correlations among characteristics of health districtsa Average income

Income inequality

Social capital Governing effectiveness Survey1

Average income

1.0

Income inequality 0.979 (p50:001) n ¼ 30

Survey2

Age-standardised mortality rate Mailing

Excess

0.876 ðp50:001Þ

0.108 (p ¼ 0:409)

0.020 0.048 0.174 0.244 0.007 (p ¼ 0:884) (p ¼ 0:708) (p ¼ 0:284) (p ¼ 0:063) (p ¼ 0:957)

1.0

0.197 (p ¼ 0:133)

0.106 0.255 0.079 0.106 0.241 (p ¼ 0:440) (p ¼ 0:048) (p ¼ 0:626) (p ¼ 0:420) (p ¼ 0:061)

Social capital

0.063 0.232 1.0 (p ¼ 0:747) (p ¼ 0:225) n ¼ 29 n ¼ 29

0.157 0.084 0.164 0.200 0.241 (p ¼ 0:261) (p ¼ 0:524) (p ¼ 0:314) (p ¼ 0:129) (p ¼ 0:066)

Survey1

0.052 0.164 0.271 (p ¼ 0:795) (p ¼ 0:413) (p ¼ 0:180) n ¼ 27 n ¼ 27 n ¼ 26

1.0

Survey2

0.138 0.369 0.056 (p ¼ 0:466) (p ¼ 0:045) (p ¼ 0:774) n ¼ 30 n ¼ 30 n ¼ 29

50.001 1.0 (p ¼ 0:999) n ¼ 27

Mailing

0.145 0.058 0.210 (p ¼ 0:542) (p ¼ 0:810) (p ¼ 0:375) n ¼ 20 n ¼ 20 n ¼ 20

0.024 0.108 1.0 (p ¼ 0:927) (p ¼ 0:650) n ¼ 17 n ¼ 29

Excess

0.314 0.142 0.314 (p ¼ 0:097) (p ¼ 0:463) (p ¼ 0:097) n ¼ 29 n ¼ 29 n ¼ 29

0.096 0.213 0.173 (p ¼ 0:641) (p ¼ 0:268) (p ¼ 0:467) 1.0 n ¼ 26 n ¼ 29 n ¼ 20

Age-standardised mortality rate

0.066 0.554 0.327 (p ¼ 0:728) (p ¼ 0:001) (p ¼ 0:083) n ¼ 30 n ¼ 30 n ¼ 29

0.063 0.342 0.028 0.284 10.0 (p ¼ 0:754) (p ¼ 0:064) (p ¼ 0:907) (p ¼ 0:136) n ¼ 27 n ¼ 30 n ¼ 20 n ¼ 29

a

0.106 0.030 0.012 0.020 (p ¼ 0:440) (p ¼ 0:869) (p ¼ 0:930) (p ¼ 0:884)

0.121 0.126 0.260 (p ¼ 0:455) (p ¼ 0:339) (p ¼ 0:044)

0.280 0.005 (p ¼ 0:085) (p ¼ 0:974)

0.180 (p ¼ 0:171)

Correlations are Pearson’s r (below the diagonal) and Kendall’s t (above the diagonal).

their distributions in 1995 may have changed during the two-year period. The crude mortality rate was positively related to the degree of income inequality (r ¼ 0:546, p ¼ 0:002; t ¼ 0:281, p ¼ 0:029; n ¼ 30), meaning that a greater skew in the distribution of income toward the wealthy end of the spectrum corresponded with a higher rate of overall mortality. (The results were similar among the rural districts (r ¼ 0:582, p ¼ 0:002; t ¼ 0:302, p ¼ 0:031; n ¼ 26).) After standardising the mortality rate for age, income inequality remained significantly related to the mortality rate (Fig. 3 and Table 1; such that a unit increase in income inequality corresponded with an increase of 11.3 in the mortality rate). (Similar results were obtained among the rural districts (r ¼ 0:587, p ¼ 0:002; t ¼ 0:323, p ¼ 0:021; n ¼ 26).) Fig. 3 shows the influence of one point (the Lloydmin-

ster health district7). Without Lloydminster, income inequality was less strongly related to the standardised mortality rate (r ¼ 0:351, p ¼ 0:062; t ¼ 0:187, p ¼ 0:154; n ¼ 29; the mortality rate increased by 6.9 for every unit increase in income inequality). (Again, similar results were obtained among the rural districts (r ¼ 0:394, p ¼ 0:051; t ¼ 0:267, p ¼ 0:062; n ¼ 25).) The percentage of births at a low weight and the use of mental health and alcohol and drug services were not significantly related to any of the income-related variables. Thus degree of wealth does not appear to 7

The Lloydminster health district is a special case. It was formed after the others included in this study, and is comprised of the Saskatchewan portion of the town of Lloydminster. The town straddles the border with Alberta and shares health-care responsibilities with that province.

856

G. Veenstra / Social Science & Medicine 54 (2002) 849–868

Fig. 2. Scatter plot of average household income and age-standardised mortality rates.

predict much, if any, variation in age-adjusted mortality rates among Saskatchewan’s health districts, but the nature of the distribution of wealth may play a role in such explanation. Bivariate analysis: social capital and health Among the twenty-nine districts with a measure of social capital (thus excluding Lloydminster), the social capital index was positively related to the crude mortality rate (r ¼ 0:482, p ¼ 0:008; t ¼ 0:269, p ¼ 0:083; n ¼ 29; Fig. 4) but negatively (although not quite significantly) related to the mortality rate standardised for age (Table 1 and Fig. 5; such that a unit increase in the social capital index corresponded with a decrease of 0.113 in the mortality rate).8 This is made all 8

Upon breaking the social capital index into its constituent parts, the crude mortality rate was related to the average of the three voting variables (r ¼ 0:649, n ¼ 30, p50:001; t ¼ 0:437, p ¼ 0:001) but not to the proxy for the number of clubs nor to the aggregated participation index from the NPHS. The agestandardised mortality rate was almost significantly related to the proxy variable for the number of clubs (r ¼ 0:356, p ¼ 0:058; t ¼ 0:236, p ¼ 0:072) n ¼ 29, but was unrelated to the other social capital indicators.

the more intriguing by noting that the crude and standardised mortality rates were positively related to one another (r ¼ 0:382, p ¼ 0:054; t ¼ 0:254, p ¼ 0:053; n ¼ 29). The turnaround after standardising for age suggests that the age distribution of districts will be an important consideration when describing the prevalence of social capital. (Among the rural districts the story is a different one, however. The relationships between the social capital index and the crude mortality rate (r ¼ 0:373, p ¼ 0:066; t ¼ 150, p ¼ 0:293; n ¼ 25) and standardised mortality rate (r ¼ 0:223, p ¼ 0:285; t ¼ 0:180, p ¼ 0:207; n ¼ 25) were less interesting in this setting.) Taken together with the finding that all of trust, voluntary and civic participation were unrelated to self-rated health status in the individual-level survey in eight districts (Veenstra, 2000), we have some indication of an ecological-level relationship (social capital and population health) that does not necessarily translate into an individual-level one (trust and/or participation in the public space and health). While noting that the social capital measures pertain to 1995–97 and the mortality rates to 1993 it seems reasonable to conclude that social capital has a causal effect on health rather than to conclude that healthier places influence the degree of social capital; Putnam et al. (1993) trace the prevalence of social capital within Italian regions back

G. Veenstra / Social Science & Medicine 54 (2002) 849–868

Fig. 3. Scatter plot of income inequality and age-standardised mortality rates.

Fig. 4. Scatter plot of social capital and crude mortality rates.

857

858

G. Veenstra / Social Science & Medicine 54 (2002) 849–868

Fig. 5. Scatter plot of social capital and age-standardised mortality rates.

several hundred years, implying that the social nature of places has a rather stable character. The social capital index was not related to the low birth-weight rate of 1993 but was negatively related to the proportion of district residents who received mental health services in 1993/94 (r ¼ 0:426, p ¼ 0:021; t ¼ 0:310, p ¼ 0:018; n ¼ 29) and to the proportion of residents who were hospitalised for psychiatric disorders (r ¼ 0:629, p50:001; t ¼ 0:527, p50:001; n ¼ 29). Finally, the social capital index was negatively related to alcohol and drug services utilisation on both an inpatient (r ¼ 0:638, p50:001; t ¼ 0:399, p ¼ 0:002; n ¼ 29) and outpatient basis (r ¼ 0:386, p ¼ 0:038; t ¼ 0:195, p ¼ 0:138; n ¼ 29). (Results were very similar among the rural districts.) Thus higher social capital was associated with a higher crude mortality rate, (possibly) a lower mortality rate standardised for age and fewer encounters with both the mental health system and alcohol and drug services among the health districts. Bivariate analysis: social capital and effective governance As hypothesised above, I would like to explore whether the performance of DHBs affects the health of residents, and, if so, whether it mediates a relationship between social capital and health. Given that the health status indicators precede the governance measures in time (and are indeed coincident with the birth of the

boards) I cannot pursue this line of inquiry in its entirety. Rather, I can explore whether social capital and governing effectiveness are related, following Putnam et al. (1993), which, if supported empirically, provides some support for this line of causality. Tables 1 and 2 show that the social capital index was not related to any of the effective governance indicators, effectively eliminating this line of inquiry from further consideration. Multivariate analysis: social capital The social capital index was not significantly related to either of the economic indicators, each of which might be argued to have causal precedence over social capital (although this direction of causality could be and has been contested). Table 3 shows that social capital scores were higher in districts with proportionately fewer females, lower rates of mobility, fewer assaults and less total crime, fewer single parent homes and fewer renters versus homeowners. Social capital scores were also higher in districts with smaller populations, a lesser or negative rate of population change, a lower population density, proportionately fewer residents who live in towns, proportionately fewer youth and/or more people older than 65 years of age and proportionately fewer agnostics. After exploring interrelationships among these socio-demographic characteristics of districts, the broad picture seems to be that districts with high rates of mobility also tend to be places with a larger population,

859

G. Veenstra / Social Science & Medicine 54 (2002) 849–868 Table 2 Relationships between social capital and DHB performance DHB performance dimension and measure

Social capital index

Reflection of health needs

Statistical test not applicable Statistical test not applicable p ¼ 0:408 (1,20)

Policy making and implementation

Fiscal responsibility

Integration and co-ordination of services

Heard representations from public Held two public meetings as required Had an adequate policy making and development self-review process Had written rules and procedures in place as required by the auditors Submitted required documents in 1995 Submitted documents on time in 1996 Safeguarded its assets and kept its accounts in good order, in 1995 Safeguarded assets in 1996 Had proper written agreements with all service providers in 1995 Had proper written agreements with all services providers in 1996

higher crime rates and proportionately more single parents, rented accommodation, women and elderly people. Such places tend to have lower social capital scores. Veenstra (1999) found strong individual-level relationships between civic and associational participation and age in a survey of citizens in eight of these thirty health districts. Given these findings and the relationships between other characteristics of districts and the distribution of age, the ecological relationships just described may partially reflect individual-level relationships between age and activity (as four of the five items in the social capital index are aggregated individual-level actions). I will consider only gender composition, the number of residents aged 65 and over, and the percentage of the population living in rural areas, each of which might be deemed to have causal precedence over, and no conceptual overlap with, the social capital index. Taken together in a multiple regression (R2 ¼ 0:661; p50:001) and therefore controlling for one another, only the age variable remained a significant predictor of social capital (p ¼ 0:001). Thus the distribution of age shows itself to be a primary consideration when describing the social capital of communities: ‘older’ communities seem to score higher on the social capital index than do ‘younger’ ones.

Multivariate analysis: age-standardised mortality rate In this section I will focus on one population health dependent variable: the mortality rate standardised for age. Although utilisation of mental health and drug and alcohol services might be considered indicators of health problems in a district, they are less obviously indicators of population health since the availability of services will

p ¼ 0:500 (1,15) p ¼ 0:306 (1,15) p ¼ 0:896 (1,15) p ¼ 0:280 (1,15) p ¼ 0:550 (1,15) p ¼ 0:531 (1,15) p ¼ 0:355 (1,15)

influence their delivery. The low birth-rate indicator was unrelated to the primary variables of interest. As was shown earlier, of the economic characteristics of health districts only income inequality was related to the mortality rate, such that greater inequality corresponded with a higher rate. The social capital index was almost significantly related to the mortality rate, such that greater social capital corresponded with fewer deaths (B ¼ 0:113, b ¼ 0:327, p ¼ 0:083 in the linear regression). Because the DHBs were created in 1993 and the mortality rates are from the same year, political performance cannot be included in the lines of causality converging upon population health status. Of the sociodemographic characteristics of the health districts, Table 3 shows that health districts with proportionately more females had higher age-standardised mortality rates, as did those districts with proportionately more singleperson households and a higher rate of population change. For the following multivariate tests I have removed the health district (Lloydminster) for which I do not have a score on the social capital index. As shown above, without this case the zero-order relationship between income inequality and mortality weakened (B ¼ 6:893, b ¼ 0:351, p ¼ 0:062 in the linear regression). Without this case the relevant socio-demographic characteristics of health districts for mortality were the proportion of the populace who are female (r ¼ 0:402, p ¼ 0:031; t ¼ 0:260, p ¼ 0:050), the total crime rate (r ¼ 0:267, p ¼ 0:162; t ¼ 0:323, p ¼ 0:014), the proportion of families with a single parent (r ¼ 0:373, p ¼ 0:046; t ¼ 0:274, p ¼ 0:037) and the proportion of renters (r ¼ 0:373, p ¼ 0:046; t ¼ 0:190, p ¼ 0:148). Can we further elucidate the relationship between income inequality and mortality? First, the inter-relationships among median household incomes

860

Table 3 Socio-demographic predictors of social capital and population healtha Age-standardised mortality rate 1993

Low birth-weight rate 1993

Mental health service use 1993

Mental health hospital use 1993

Alcohol and drug service inpatients 1993

Alcohol and drug service outpatients 1993

Income inequality

0.232 (p ¼ 0:225) 0.197 (p ¼ 0:133) n ¼ 29

0.554 (p ¼ 0:001) 0.241 (p ¼ 0:061) n ¼ 30

0.171 (p ¼ 0:366) 0.210 (p ¼ 0:104) n¼3

0.037 (p ¼ 0:845) 0.071 (p ¼ 0:580) n ¼ 30

0.238 (p ¼ 0:205) 0.264 (p ¼ 0:040) n ¼ 30

0.083 (p ¼ 0:662) 0.011 (p ¼ 0:929) n ¼ 30

0.107 (p ¼ 0:573) 0.032 (p ¼ 0:803) n ¼ 30

Gender (% female)

0.392 (p ¼ 0:035) 0.240 (p ¼ 0:071) n ¼ 29

0.542 (p ¼ 0:002) 0.310 (p ¼ 0:017) n ¼ 30

0.061 (p ¼ 0:747) 0.096 (p ¼ 0:463) n ¼ 30

0.324 (p ¼ 0:081) 0.110 (p ¼ 0:401) n ¼ 30

0.569 (p ¼ 0:001) 0.333 (p ¼ 0:011) n ¼ 30

0.163 (p ¼ 0:388) 0.105 (p ¼ 0:421) n ¼ 30

0.037 (p ¼ 0:845) 0.061 (p ¼ 0:642) n ¼ 30

Mobility (1 year)

0.812 (p50:001) 0.582 (p50:001) n ¼ 26

0.011 (p ¼ 0:957) 0.100 (p ¼ 0:466) n ¼ 27

0.140 (p ¼ 0:486) 0.238 (p ¼ 0:083) n ¼ 27

0.395 (p ¼ 0:041) 0.322 (p ¼ 0:018) n ¼ 27

0.666 (p50:001) 0.311 (p ¼ 0:023) n ¼ 27

0.489 (p ¼ 0:010) 0.425 (p ¼ 0:002) n ¼ 27

0.489 (p ¼ 0:010) 0.228 (p ¼ 0:095) n ¼ 27

Mobility (5 years)

0.715 (p50:001) 0.489 (p50:001) n ¼ 26

0.338 (p ¼ 0:091) 0.145 (p ¼ 0:300) n ¼ 26

0.179 (p ¼ 0:383) 0.251 (p ¼ 0:074) n ¼ 26

0.771 (p50:001) 0.495 (p50:001) n ¼ 26

0.711 (p50:001) 0.385 (p ¼ 0:006) n ¼ 26

0.606 (p ¼ 0:001) 0.342 (p ¼ 0:014) n ¼ 26

0.652 (p50:001) 0.247 (p ¼ 0:078) n ¼ 26

Crime-rate (assault)

0.697 (p50:001) 0.586 (p50:001) n ¼ 29

0.172 (p ¼ 0:364) 0.102 (p ¼ 0:432) n ¼ 30

0.243 (p ¼ 0:196) 0.139 (p ¼ 0:284) n ¼ 30

0.168 (p ¼ 0:376) 0.106 (p ¼ 0:412) n ¼ 30

0.190 (p ¼ 0:316) 0.240 (p ¼ 0:063) n ¼ 30

0.545 (p ¼ 0:002) 0.406 (p ¼ 0:002) n ¼ 30

0.435 (p ¼ 0:016) 0.192 (p ¼ 0:138) n ¼ 30

Crime-rate (total)

0.778 (p50:001) 0.520 (p50:001) n ¼ 29

0.231 (p ¼ 0:219) 0.235 (p ¼ 0:069) n ¼ 30

0.235 (p ¼ 0:211) 0.282 (p ¼ 0:029) n ¼ 30

0.280 (p ¼ 0:135) 0.285 (p ¼ 0:027) n ¼ 30

0.204 (p ¼ 0:280) 0.258 (p ¼ 0:046) n ¼ 30

0.619 (p50:001) 0.525 (p50:001) n ¼ 30

0.449 (p ¼ 0:013) 0.205 (p ¼ 0:112) n ¼ 30

G. Veenstra / Social Science & Medicine 54 (2002) 849–868

Social capital index

Table 3 (continued) Low birth-weight rate 1993

Mental health service use 1993

Mental health hospital use 1993

Alcohol and drug service inpatients 1993

0.260 (p ¼ 0:173) 0.094 (p ¼ 0:475) n ¼ 29

0.398 (p ¼ 0:030) 0.229 (p ¼ 0:077) n ¼ 30

0.161 (p ¼ 0:394) -0.058 (p ¼ 0:655) n ¼ 30

0.086 (p ¼ 0:653) 0.039 (p ¼ 0:761) n ¼ 30

0.047 (p ¼ 0:804) 0.030 (p ¼ 0:816) n ¼ 30

0.164 (p ¼ 0:386) 0.049 (p ¼ 0:708) n ¼ 30

(p ¼ 0:018) 0.264 (p ¼ 0:042) n ¼ 30

Single-parent homes

0.824 (p50:001) 0.700 (p50:001) n ¼ 29

0.284 (p ¼ 0:129) 0.235 (p ¼ 0:069) n ¼ 30

0.342 (p ¼ 0:064) 0.269 (p ¼ 0:038) n ¼ 30

0.363 (p ¼ 0:049) 0.194 (p ¼ 0:134) n ¼ 30

0.633 (p50:001) 0.443 (p ¼ 0:001) n ¼ 30

0.719 (p50:001) 0.484 (p50:001) n ¼ 30

0.351 (p ¼ 0:057) 0.109 (p ¼ 0:401) n ¼ 30

Ratio of homeowners to renters

0.640 (p50:001) 0.453 (p ¼ 0:001) n ¼ 29

0.091 (p ¼ 0:634) 0.122 (p ¼ 0:344) n ¼ 30

0.178 (p ¼ 0:345) 0.219 (p ¼ 0:090) n ¼ 30

0.501 (p ¼ 0:005) 0.343 (p ¼ 0:008) n ¼ 30

0.465 (p ¼ 0:010) 0.246 (p ¼ 0:056) n ¼ 30

0.570 (p ¼ 0:001) 0.370 (p ¼ 0:004) n ¼ 30

0.456 (p ¼ 0:011) 0.207 (p ¼ 0:108) n ¼ 30

Population

0.465 (p ¼ 0:011) 0.399 (p ¼ 0:002) n ¼ 29

0.218 (p ¼ 0:248) 0.090 (p ¼ 0:487) n ¼ 30

0.144 (p ¼ 0:447) 0.182 (p ¼ 0:158) n ¼ 30

0.014 (p ¼ 0:940) 0.301 (p ¼ 0:019) n ¼ 30

0.283 (p ¼ 0:130) 0.333 (p ¼ 0:010) n ¼ 30

0.401 (p ¼ 0:028) 0.228 (p ¼ 0:077) n ¼ 30

0.161 (p ¼ 0:395) 0.078 (p ¼ 0:544) n ¼ 30

Rate of population change

0.758 (p50:001) 0.591 (p50:001) n ¼ 29

0.488 (p ¼ 0:010) 0.323 (p ¼ 0:018) n ¼ 27

0.369 (p ¼ 0:045) 0.307 (p ¼ 0:018) n ¼ 30

0.080 (p ¼ 0:674) 0.149 (p ¼ 0:246) n ¼ 30

0.172 (p ¼ 0:365) 0.172 (p ¼ 0:181) n ¼ 30

0.627 (p50:001) 0.389 (p ¼ 0:003) n ¼ 30

0.199 (p ¼ 0:291) 0.014 (p ¼ 0:915) n ¼ 30

Dependency ratio

0.549 (p ¼ 0:002) 0.328 (p ¼ 0:014) n ¼ 29

0.093 (p ¼ 0:626) 0.040 (p ¼ 0:761) n ¼ 30

0.024 (p ¼ 0:900) 0.106 (p ¼ 0:420) n ¼ 30

0.319 (p ¼ 0:086) 0.370 (p ¼ 0:005) n ¼ 30

0.377 (p ¼ 0:040) 0.370 (p ¼ 0:005) n ¼ 30

0.560 (p ¼ 0:001) 0.346 (p ¼ 0:008) n ¼ 30

0.329 (p ¼ 0:076) 0.349 (p ¼ 0:008) n ¼ 30

Population density

0.486 (p ¼ 0:008) 0.422 (p ¼ 0:001) n ¼ 29

0.088 (p ¼ 0:642) 0.041 (p ¼ 0:748) n ¼ 30

0.136 (p ¼ 0:472) 0.240 (p ¼ 0:063) n ¼ 30

0.016 (p ¼ 0:934) 0.124 (p ¼ 0:335) n ¼ 30

0.296 (p ¼ 0:112) 0.377 (p ¼ 0:003) n ¼ 30

0.447 (p ¼ 0:013) 0.336 (p ¼ 0:009) n ¼ 30

0.058 (p ¼ 0:759) 0.090 (p ¼ 0:486) n ¼ 30

Single-person households

Alcohol and drug service outpatients 1993

861

Age-standardised mortality rate 1993

G. Veenstra / Social Science & Medicine 54 (2002) 849–868

Social capital index

862

Table 3 (continued) Age-standardised mortality rate 1993

Low birth-weight rate 1993

Mental health service use 1993

Mental health hospital use 1993

Alcohol and drug service inpatients 1993

Alcohol and drug service outpatients 1993

Population in town

0.440 (p ¼ 0:017) 0.262 (p ¼ 0:047) n ¼ 29

0.097 (p ¼ 0:611) 0.081 (p ¼ 0:532) n ¼ 30

0.087 (p ¼ 0:649) 0.090 (p ¼ 0:486) n ¼ 30

0.451 (p ¼ 0:012) 0.320 (p ¼ 0:013) n ¼ 30

0.435 (p ¼ 0:016) 0.256 (p ¼ 0:048) n ¼ 30

0.401 (p ¼ 0:028) 0.237 (p ¼ 0:066) n ¼ 30

0.314 (p ¼ 0:091) 0.231 (p ¼ 0:074) n ¼ 30

Population in rural area

0.673 (p50:001) 0.481 (p50:001) n ¼ 29

0.246 (p ¼ 0:190) 0.175 (p ¼ 0:175) n ¼ 30

0.242 (p ¼ 0:198) 0.282 (p ¼ 0:029) n ¼ 30

0.378 (p ¼ 0:040) 0.331 (p ¼ 0:010) n ¼ 30

0.491 (p ¼ 0:006) 0.322 (p ¼ 0:012) n ¼ 30

0.578 (p ¼ 0:001) 0.410 (p ¼ 0:001) n ¼ 30

0.190 (p ¼ 0:316) 0.122 (p ¼ 0:344) n ¼ 30

Percent 14 and under

0.401 (p ¼ 0:031) 0.279 (p ¼ 0:034) n ¼ 29

0.146 (p ¼ 0:443) 0.032 (p ¼ 0:803) n ¼ 30

0.387 (p ¼ 0:035) 0.148 (p ¼ 0:253) n ¼ 30

0.013 (p ¼ 0:944) 0.083 (p ¼ 0:520) n ¼ 30

0.073 (p ¼ 0:701) 0.055 (p ¼ 0:668) n ¼ 30

0.184 (p ¼ 0:330) 0.134 (p ¼ 0:301) n ¼ 30

0.166 (p ¼ 0:381) 0.035 (p ¼ 0:789) n ¼ 30

Percent 65 and older

0.720 (p50:001) 0.494 (p50:001) n ¼ 29

0.181 (p ¼ 0:340) 0.016 (p ¼ 0:901) n ¼ 30

0.326 (p ¼ 0:078) 0.248 (p ¼ 0:056) n ¼ 30

0.189 (p ¼ 0:316) 0.168 (p ¼ 0:193) n ¼ 30

0.181 (p ¼ 0:339) 0.141 (p ¼ 0:276) n ¼ 30

0.524 (p ¼ 0:003) 0.293 (p ¼ 0:023) n ¼ 30

0.331 (p ¼ 0:074) 0.134 (p ¼ 0:301) n ¼ 30

Religion: None

0.653 (p50:001) 0.442 (p50:001) n ¼ 29

0.027 (p ¼ 0:886) 0.137 (p ¼ 0:292) n ¼ 30

0.184 (p ¼ 0:329) 0.156 (p ¼ 0:231) n ¼ 30

0.218 (p ¼ 0:248) 0.220 (p ¼ 0:090) n ¼ 30

0.158 (p ¼ 0:403) 0.113 (p ¼ 0:381) n ¼ 30

0.603 (p50:001) 0.419 (p ¼ 0:001) n ¼ 30

0.295 (p ¼ 0:113) 0.204 (p ¼ 0:116) n ¼ 30

a

Correlations are Pearson’s r and Kendall’s t (in italics). Average income, the crime rate for liquor offences and the percentage of the population who speak English, live in Indian bands, and profess Roman Catholic, Protestant or ‘other’ religious affiliation were not significantly related to any of the dependent variables.

G. Veenstra / Social Science & Medicine 54 (2002) 849–868

Social capital index

G. Veenstra / Social Science & Medicine 54 (2002) 849–868

(a proxy for wealth), the distribution of income and health are of special interest. Taken together in a multiple regression, income inequality was significantly related to mortality (B ¼ 11:842, b ¼ 0:603 and p ¼ 0:025) while median household income was not (p ¼ 0:175). Thus income inequality became a better predictor of mortality after controlling for this crude measure of wealth. Second, does social capital mediate the relationship between income inequality and health, or vice versa? The relationships between inequality and mortality and between social capital and morality were of relatively equal strength, but the two independent variables were not significantly related to one another (Table 1). Taken together in a multiple regression, neither income inequality (B ¼ 5:707, b ¼ 0:290, p ¼ 0:122) nor social capital (B ¼ 0:089, b ¼ 0:260, p ¼ 0:165) were significant predictors of mortality. Thus one or each of social capital and income inequality may have some influence on the other’s relationship with mortality. Third, the gender composition of a district might be considered either a consequence of the mortality rate, antecedent to the income inequality and health relationship, causally prior to both inequality and population health or an unrelated independent variable; the weak relationship between gender composition and income inequality (r ¼ 0:188, p ¼ 0:329; t ¼ 0:145, p ¼ 0:275) supports the first and last suppositions. Upon controlling for gender in a multiple regression income inequality lost significance (B ¼ 5:608, b ¼ 0:285, p ¼ 0:113), supporting the notion that income inequality may mediate the gender and mortality relationship in part or that some of the income inequality and health relationship found herein is spurious. Fourth, others have argued that one of the ways by which income inequality influences population health status might be through an effect on crime (e.g. Kennedy et al., 1998). It makes sense to me to consider crime as either an unrelated independent variable or as one intervening between inequality and health rather than as a variable that affects both inequality and health; the non-relationship between inequality and crime rates (r ¼ 0:036, p ¼ 0:852; t ¼ 0:042, p ¼ 0:750) supports the first supposition. As expected, upon controlling for total crime rates in a multiple regression the influence of income inequality on mortality remained relatively unchanged (B ¼ 6:711, b ¼ 0:342, p ¼ 0:064). Can we further elucidate the relationship between social capital and mortality? First, social capital and total crime rates had relationships with mortality and were strongly related to one another. Others have suggested that crime may also mediate the relationship between social capital and health (e.g. Kennedy et al., 1998). Upon controlling for total crime rates in a multiple regression, neither crime rates (p ¼ 0:917) nor social capital (B ¼ 0:104, b ¼ 0:303, p ¼ 0:314) were

863

significant, although the beta for the latter changed only a little upon adding the control. This suggests that, in this instance at least, crime probably does not mediate the relationship between social capital and health. Second, as with income inequality, the gender composition of districts may be considered a consequence of the mortality rate, antecedent to the social capital and health relationship or causally prior to both. Upon controlling for gender in a multiple regression, social capital lost power to predict variability in mortality (B ¼ 0:069, b ¼ 0:200, p ¼ 0:304). Thus I would argue that either some of the relationship between social capital and mortality is spurious or that social capital provides another means by which the gender composition of a community influences its mortality rate. Finally, taken together in a multiple regression (R2 ¼ 0:283, p ¼ 0:081) to simply determine which variable has the greatest explanatory power while controlling for the others, neither income inequality (B ¼ 5:828, b ¼ 0:297, p ¼ 0:118), social capital (B ¼ 0:027, b ¼ 0:079, p ¼ 0:806), gender (B ¼ 0:294, b ¼ 0:335, p ¼ 0:098) nor total crime (B ¼ 0:004, b ¼ 0:264, p ¼ 0:374) had coefficients that were significantly different from zero. Even so, of the four the social capital index had by far the least, and gender composition the most, explanatory power. Returning to the research questions broached in the introduction, the distribution of wealth, albeit roughly measured, proved to be a moderate predictor of mortality, more so after controlling for the proxy measure of wealth. Overall wealth did not predict much, if any, variation in mortality rates. Second, the social capital index proved itself to be a somewhat meaningful predictor of mortality, with some indication of an effect on mortality co-mingled with that from income inequality. Third, effective governance by regional health authorities did not contribute to understanding health inequalities in this context. Fourth and finally, additionally incorporating socio-demographic characteristics into multivariate analysis, the gender composition of a health district proved to be the most salient predictor of age-standardised mortality rates (perhaps not surprising given that the mortality rates were not sex-adjusted).

Discussion Social capital is a popular theoretical concept that has been empirically linked to all of income inequality, population health status and effective governance in various settings using various measures of social capital. The common conceptual theme in this work is that many phenomenon that we might deem important, such as population health, economic growth or vital government, are predicated upon the nature of social

864

G. Veenstra / Social Science & Medicine 54 (2002) 849–868

relationships in the civil space. This diffuse and nebulous space, wherein individuals leave the sanctuary of their homes to participate and socialise with others, either for politically inspired reasons or for their own enjoyment, may set the context within which these and other important ends are accomplished. It may provide a culture of political participation within which democracy and business can flourish or may promote cohesive communities that promote the health of residents through complex and subtle means such as those suggested above. Social capital and regional health governance One plausible line of causality starts with social capital in communities influencing the performance of political institutions which in turn influence the health status of community residents. There did not appear to be a relationship between social capital, as I measured it, and the performance of the District Health Boards in Saskatchewan, as I evaluated it; every deductive test was non-significant. This may be because I failed to adequately measure the concepts; that social capital, as I defined it, is not related to governance in general; or the boards are too new to reflect social capital influences. The first and third possibilities seem most reasonable in this instance. I have, however, failed to confirm Putnam’s findings of a relationship between social capital and effective regional governance in this particular context. There were differences between my work and Putnam’s that may be relevant to my failure to replicate his results. The interviews with regional councillors conducted by Putnam were likely more informative than the surveys of board members, although the surveys may provide better comparative data. Putnam also spent much energy interviewing community leaders (nonpoliticians, for the most part)}three waves of interviews with community leaders in the six regions, and one mailed survey of community leaders in all 20 regions; I did not obtain any similar information or perspectives. Putnam collected his data over the course of 15 years; I collected it in one, and, although my data sources span a range of approximately six years, the majority of my measures are from 1996 and 1997. Also, Putnam’s regional governments were evaluated over the course of 15 years whereas I evaluated DHBs over only three years, and as soon as two years after their conception. The scenarios for the tests of the relationship between social capital and the performance of political institutions were also quite different (regions of a large country versus regions in one small province). However, in both settings the regions were created in an attempt to match historical regions. Putnam et al. (1993) claim that ‘‘the borders of the new governments largely corresponded to the territories of historic regions of the peninsula,

including such celebrated principalities as Tuscany and Lombardy’’ (p. 5). In Saskatchewan community leaders were asked to be part of the district creation process so that the districts might reflect historical regions in the province. The cohesiveness of the social capital index supports the notion that the health districts conform to some social topography. Social capital, income inequality and health Although the role of the performance of DHBs as an intervening variable between social capital and health has been removed from consideration, the role of social capital as a predictor of health via other means has not. The health districts vary with respect to the distribution of age within them, such that standardising for age when calculating mortality rates makes a significant difference in a health district’s population health ‘score’. Given that the social capital index was strongly related to the age distribution, such that ‘older’ districts might be deemed to have a greater store of social capital, it is not surprising that social capital had rather different relationships with the crude and age-standardised mortality rates. The extremity of the difference is surprising, however. Before controlling for age, high social capital districts had higher rates of mortality than did low social capital areas; the relationship completely changed direction after controlling for age. Strong relationships have been found elsewhere (e.g. Putnam, 1996; Veenstra, 1999) between age and both trust and participation in the civil space. Putnam (1996) argues for a cohort effect}earlier generations are simply more trusting than later ones. Given this, we are reminded anew of the importance of age when exploring social capital: ‘older’ districts appear to have a more vibrant grassroots political landscape and a greater density of secondary associations. How might social capital influence health? This question touches on basic sociological discourse. Social capital and/or social cohesion fall on the ‘consensual’ side of sociological theory, which can be traced through structural functionalism as far back as Durkheim. Income inequality, on the other hand, might be considered a specific indicator of a more general social inequality. It could reflect the nature of class relations and exploitation, as Marx (and Coburn (2000) and Muntaner and Lynch (1999)) might argue, or power struggles and domination among certain interest groups, as feminist or conflict theorists might argue. In comparisons of societies or communities with governments, income inequality likely reflects government policies toward the redistribution of wealth as well. (This is probably not the case among Saskatchewan’s health districts, however, as all health districts share the same federal and provincial government responsible for redistributing wealth.) The consensual and conflictual

G. Veenstra / Social Science & Medicine 54 (2002) 849–868

sides in sociological discourse do not agree on the driving causal forces of society, making interpretation of empirical relationships in this instance extremely problematic. I can say no more than to note that in this instance social capital and income inequality may be comingled when it comes to predicting population health but that the nature of their inter-relationship remains undetermined, and that income inequality and the social capital index seem to have relationships with mortality that are similar in strength. After adding other controls, however, income inequality proved to have more predictive power than did the social capital index. Previous work among Canadian provinces and among Canadian cities shows small promise for income inequality as a predictor of population health within Canada (Ross et al., 2000). In this article I have shown that the level of analysis when exploring relationship between income inequality and health is an important consideration. The inclusion of rural areas and attention to a smaller (with respect to population size) unit of analysis brings the relevance of income inequality as a predictor of a population’s health back on stage in Canada, and supports further efforts toward understanding the role of civil society in the social production of the health of Canadians.

Acknowledgements I would like to thank Jonathan Lomas, Ralph Matthews, John Fox and David Streiner for serving on my dissertation committee. Thanks also go to the HEALNet-Regional Health Planning Theme, Provincial Auditor and District Health Boards of Saskatchewan for facilitating data collection. Gabriela Pechlaner collected the income information for the health districts. I was supported by Social Sciences and Humanities Research Council of Canada doctoral fellowships in 1996–98 and by SSHRC and CHSPR postdoctoral fellowships in 1999/2000.

Appendix. A Measures of social capital I was unable to obtain a listing of all clubs and voluntary associations in the province and so contacted parent associations9 with subsidiary groups scattered throughout the province instead. I counted the number 9

In the summer of 1997 I received information from 13 parent associations: arts and crafts festivals, music companies, theatre companies, arts councils, choral companies, multicultural councils, ringette clubs, scouts clubs, 4-H clubs, squash clubs, girl guide clubs, boxing clubs and Women’s Institutes.

865

of such subsidiary groups within each health district as a proxy measure of ‘associational density,’ controlling for population size (I called this variable clubs1). I also obtained from Statistics Canada, by health district, a derived ‘social involvement’ index (clubs2) that combined two questions on the 1995 National Population Health Survey that pertain to associational activity. These were (i) ‘‘Are you a member of any voluntary organizations or associations?’’ and (ii) ‘‘How often do you participate in meetings/activities sponsored by these groups?’’.10 For one health district (Gabriel Springs) I was unable to obtain data for the latter item and so substituted in the average score for the other districts. To measure ‘civic participation’ I determined, by health district, the percentage of eligible voters who voted in (1) the federal election of 1993 (vote1), (2) the provincial election of 1995 (vote2) and (3) the District Health Board elections of 1995 (vote3). For one health district (Lloydminster) I was unable to obtain any voting information and therefore did not record a social capital score for that district. Table 4 describes the resultant coherent social capital index among twenty-eight health districts (including Gabriel Springs changes the correlations only slightly). The mean score for the index was 5.22 with a standard deviation of 1.95. Measures of District Health Board performance I created two indices (survey1 and survey2) that describe board members’ assessments of their own board’s ‘comprehensive performance’, aggregated to the level of the board. In 1995 Lomas et al. (1997) conducted a survey of RHA board members in five Canadian provinces with an overall response rate of 64% among Saskatchewan District Health Board members. I aggregated responses by DHB (N ¼ 27) and created an index (survey1) with the following items. The Pearson’s correlation for each item represents its correlation with the index minus that item. The mean of the inter-item correlations was r ¼ 0:347 and Cronbach’s a was 0.795. 1. We have sufficient needs assessment information (r ¼ 0:621) 2. I am confident my board makes good decisions (r ¼ 0:587) 3. I am confident board decisions are better than the province’s were (r ¼ 0:568) 4. We have sufficient citizens’ preferences information (r ¼ 0:560) 10

The number of responses to the NPHS per district ranged from a low of four to a high of 231. This means that, for many of the districts, the standard errors were quite large for the district summaries on these questions. The data should be interpreted with this limitation in mind.

866

G. Veenstra / Social Science & Medicine 54 (2002) 849–868

Table 4 Social capital indexa Clubs1

Clubs2

Vote1

Vote2

Vote3

Rest

Clubs1

1.0

0.120 (p ¼ 0:374)

0.540 (p50:001)

0.476 (p50:001)

0.389 (p ¼ 0:004)

0.672

Clubs2

0.185 (p ¼ 0:345)

1.0

0.242 (p=0.072)

0.247 (p ¼ 0:066)

0.293 (p ¼ 0:030)

0.418

Vote1

0.703 (p50:001)

0.385 (p ¼ 0:043)

1.0

0.471 (p50:001)

0.379 (p ¼ 0:005)

0.738

Vote2

0.624 (p50:001)

0.459 (p ¼ 0:014)

0.716 (p50:001)

1.0

0.453 (p ¼ 0:001)

0.794

0.404 (p ¼ 0:033)

0.448 (p ¼ 0:017)

0.619 (p50:001)

1.0

0.626

Vote3

0.536 (p ¼ 0:003) N ¼ 28, Cronbach’s a=0.841 a

Correlations are Pearson’s r (below the diagonal) and Kendall’s t (above the diagonal). The column ‘rest’ represents the Pearson’s correlation between each item and the remainder of the index minus that item.

5. Our board is at least quite involved in ensuring effectiveness and efficiency of services (r ¼ 0:547) 6. We have sufficient key informants’ opinions (r ¼ 0:541) 7. We have sufficient population needs information (r ¼ 0:379) 8. I influence my board’s decisions (r ¼ 0:342) In 1997 the HEALNet-RHP Theme conducted a follow-up survey of board members in Saskatchewan with an overall response rate of 77%. I aggregated responses by DHB (N ¼ 30) and created an index (survey2) with the following items. The Pearson’s correlation for each item represents its correlation with the index minus that item. The mean of the inter-item correlations was r ¼ 0:409 and Cronbach’s a was 0.902. 1. Our board is good at long range planning (r ¼ 0:827) 2. I am confident that our board generally makes good decisions (r ¼ 0:779) 3. Our board manages its money well (r ¼ 0:787) 4. Our board can be considered creative in addressing problems (r ¼ 0:755) 5. Our board effectively communicates the rationale for our decisions to district residents (r ¼ 0:681) 6. Board meetings are run efficiently and effectively (r ¼ 0:594) 7. Our board has adequate mechanisms for board evaluation (r ¼ 0:574) 8. Our board’s values reflect the values of the district (r ¼ 0:570) 9. The level of board involvement in allocating funds should not increase (r ¼ 0:567)

10.The level of board involvement in planning programs and services should not increase (r ¼ 0:566) 11.Our board is responsive to wishes of district residents (r ¼ 0:548) 12.Our board has an accurate understanding of what district residents want for the health care system (r ¼ 0:437) 13.The level of board involvement in ensuring service effectiveness and efficiency should not increase (r ¼ 0:353) In addition, to measure ‘reflection of health needs’, from an analysis of the minutes taken in DHB meetings11 during 1995/96 I determined whether the board heard or received any representations from the public as a measure of their commitment to determine need. Similarly, from the Provincial Auditor’s evaluations of the boards12 I determined whether the board held two public meetings as prescribed by Saskatchewan’s Health District Act. To measure ‘policy making and implementation’, in the analysis of board minutes I determined whether the board had a systematic policy review or development process in place as a measure of their reflexivity with respect to creating good policy. To assess bureaucratic 11 This analysis was conducted by the HEALNet-RHP Theme in Saskatoon, Saskatchewan. 12 In Saskatchewan, every DHB is responsible, each year, for hiring an auditor from their district to audit the DHB’s financial situation using guidelines prescribed by the Provincial Auditor. For the 1995 fiscal year the Provincial Auditor made available observations for 18 boards and for 1996 made available 20.

867

G. Veenstra / Social Science & Medicine 54 (2002) 849–868

efficiency, on November 7, 1997 I sent a letter to each of the thirty DHBs requesting some information (contact names and addresses for several small towns in the district) and recorded the date at which I first received a reply from the board. When they mailed a package I noted the date on the postmark; when they called or emailed I noted the date of communication. A board’s score was the total number of days beyond Nov. 7 at which I first heard from the board}the lower the score the better the DHB performed in this test (mailing). This replicated Putnam’s bureaucratic efficiency test conducted in Italy, in which he approached the regional governments and noted how long it took them to fulfil an administrative request. To additionally measure organisational integrity, from the 1995 audits I also determined whether the board’s written rules and procedures were adequate, again from the Provincial Auditor’s point of view. To measure ‘fiscal responsibility’ I calculated the percent in which there was an excess of revenue over expenses, or vice versa, for the fiscal year ending on March 31, 1995 (excess). This allowed me to determine how well the board was able to conduct business while staying within the parameters dictated by revenues. The boards also have fiscal responsibilities to the Ministry of Health mandated by the Health Act. From the 1995 audit I determined whether the board was late submitting required documents to the Ministry of Health and whether the accounts, records, banking and investments were done properly. From the 1996 audit I determined whether the board adequately safeguarded assets, upheld its accountabilities to residents and Minister of Health and was late submitting required documents to the Ministry. Finally, to measure ‘integration and coordination of services’, from the 1995 and 1996 audits I determined whether the board had proper written service agreements with all of its providers. I argue that proper agreements facilitate co-ordination, which is more difficult to achieve when a board is unaware of which services are being offered by whom (Table 5).

Table 5 Correlations among the continuous measures of DHB effectivenessa Survey1

Survey2

Mailing

Excess

Survey1 1.0

0.106 (p ¼ 0:440)

0.030 0.012 (p ¼ 0:869) (p ¼ 0:930)

Survey2 50.001 (p ¼ 0:999) n ¼ 27

1.0

-0.121 0.126 (p ¼ 0:455) (p ¼ 0:339)

Mailing

0.024 0.108 ( p ¼ 0:927) (p ¼ 0:650) n ¼ 27 n ¼ 20

Excess

0.096 0.213 0.173 1.0 ( p ¼ 0:641) ( p ¼ 0:268) (p ¼ 0:467) n ¼ 26 n ¼ 29 n ¼ 20

1.0

0.280 (p ¼ 0:085)

a

Correlations are Pearson’s r (below the diagonal) and Kendall’s t (above the diagonal).

$33,865 with a standard deviation of 4796; for the mean household income the corresponding figures were $40,189 and 4798. I measured income inequality by calculating the difference between the median and the mean, dividing by the median and multiplying by 100. This measure assesses the degree of skew in the distribution of household income. I could not access the original Census data files that would enable me to calculate the Gini coefficient, the proportion of total wealth in a health district owned by the poorest 70% of households, the true median income or the standard deviation of household incomes. The median I used was not the true median, but rather a weighted average of the medians of the health districts’ constituent CSDs. The mean score on this measure of income inequality was 0.19 with a standard deviation of 0.04.

Wealth and income inequality characteristics of the health districts From the 1996 Canadian Census I calculated the average and median household incomes for the health districts by aggregating Census Sub Divisions to the level of health district. 263 of the 819 CSDs (32.1%) had income data suppressed due to the small N (fewer than about 50 households) and so were not included. These 263 CSDs represent approximately 4.8% of the total number of households in the health districts. Several CSDs straddle health district boundaries, in which case I assigned the CSD to the health district which claimed more than 50% of its households. The mean score for the health districts’ median household incomes was

Socio-demographic characteristics of health districts From the Saskatchewan Ministry of Health I obtained for each health district the population size in 1994, population density (number of people per square kilometre), gender composition (percent female), age distribution and dependency ratio (the number of residents under age 15 or over age 65 divided by the remainder). I obtained the percentage of the population living in Indian bands, rural communities and in communities larger than 1000 persons, and also the rate of population change between 1984 and 1994. For 1993 I obtained rates for liquor-related offences, assault cases, and overall crime rates.

868

G. Veenstra / Social Science & Medicine 54 (2002) 849–868

From the 1991 Census I obtained the percentage of health district residents who own their home, live in a single-person household, live in a single-parent household, claim English as their mother tongue and claim Catholic, Protestant, other or no religious affiliation. I also obtained the one- and five-year rates of mobility (proportion of residents who have lived in their same home for the past year(s)).

References Bourne, L. S. (1997). Social inequalities, polarization and the redistribution of income within cities: A Canadian example. In B. A. Babcock, M. H. Browett, (Eds.), Developing small area indicators for policy research in Australia. Bunting, T. E. (1991). Social differentiation in Canadian cities. In T. E. Bunting, P. Filion, (Eds.), Canadian cities in transition. Toronto: Oxford University Press. Coburn, D. (2000). Income inequality, social cohesion and the health status of populations: The role of neo-liberalism. Social Science and Medicine, 51, 135–146. Coleman, J. S. (1988). Social capital in the creation of human capital. American Journal of Sociology, 94, S95–S120. Kaplan, G. A., Pamuk, E., Lynch, J. W., Cohen, R. D., & Balfour, J. L. (1996). Income inequality and mortality in the United States. British Medical Journal,, 312, 999–1003. Kawachi, I., Berkman, L. F. (2000). Social cohesion, social capital and health. In: L. F. Berkman, I. Kawachi (Eds.), Social epidemiology. New York: Oxford University Press. Kawachi, I., Kennedy, B., Lochner, K., & Prothrow-Stith, D. (1997). Social capital, income inequality, and mortality. American Journal of Public Health, 87, 1491–1498. Kennedy, B. P., Kawachi, I., & Prothrow-Stith, D. (1996). Income distribution and mortality: Cross-sectional ecological study of the Robin Hood index in the United States. British Medical Journal, 213, 1004–1007. Kennedy, B. P., Kawachi, I., Prothrow-Stith, D., Lochner, K., & Gupta, V. (1998). Social capital, income inequality, and firearm violent crime. Social Science and Medicine, 47, 7–17. Lavis, J., Stoddart, G., 1999. Social cohesion and health. CHEPA Working Papers Series 99-09, McMaster University.

Lomas, J., Woods, J., & Veenstra, G. (1997). Devolving authority for health in Canadas provinces. I Overview and introduction to the issues. Canadian Medical Association Journal, 156, 371–377. Lynch, J. W., & Kaplan, G. A. (1997). Understanding how inequality in the distribution of income affects health. Journal of Health Psychology, 2, 297–314. Lynch, J. W., Kaplan, G. A., Pamuk, E. R., Cohen, R. D., Heck, K. E., Balfour, J. L., & Yen, I. H. (1998). Income inequality and mortality in metropolitan areas of the United States. American Journal of Public Health, 88, 1074–1079. MacLachlan, I., & Sawada, R. (1997). Measures of income inequality and social polarization in Canadian metropolitan areas. The Canadian Geographer, 41, 377–397. Muntaner, C., & Lynch, J. (1999). Income inequality, social cohesion, and class relations: A critique of Wilkinsons neoDurkheimian research program. International Journal of Health Services, 29, 59–81. Putnam, R. D. (1996). The strange disappearance of civic America. The American Prospect, 24, 34–48. Putnam, R. D., Leonardi, R., & Nanetti, R. Y. (1993). Making democracy work: Civic traditions in modern Italy. New Jersey: Princeton University Press. Rice, T. W., & Sumberg, A. F. (1997). Civic culture and government performance in the American states. Publius, 27, 99–114. Ross, N. A., Wolfson, M. C., Dunn, J. R., Berthelot, J., Kaplan, G. A., & Lynch, J. W. (2000). Relation between income inequality and mortality in Canada and in the United States: Cross-sectional assessment using census data and vital statistics. British Medical Journal, 320, 898–902. Veenstra, G., & Lomas, J. (1999). Home is where the governing is: Social capital and regional health governance. Health and Place, 5, 1–12. Veenstra, G. (1999). Social capital 3: An empirical exploration of relationships among individual-level social-psychological and action components of social capital. University of British Columbia CHSPR Working Papers Series HPRU99:3D. Veenstra, G. (2000). Social capital and health: An individuallevel analysis. Social Science and Medicine, 50, 619–629. Veenstra, G. (2001). Social capital and health. isuma: Canadian Journal of Policy Research, 2, 72–81. Wilkinson, R. G. (1996). Unhealthy societies: The afflictions of inequality. London: Routledge.