Previous literature

A recent literature review of the link between social capital (one measure of psycho-social resources) and health (Islam et al., Reference Islam, Merlo, Kawachi, Lindstrom and Gerdtham2006) identifies two main approaches: the first one posits that what matters for health are the level and quality of psycho-social resources (e.g. civic engagement, trust, public services) available at the aggregate level; in such a conceptual framework social inequalities in health stem from the uneven distribution of socio-economic statuses across geographic areas (Kawachi and Berkman, Reference Kawachi and Berkman2003, Veenstra et al., Reference Veenstra, Luginaah, Wakefield, Birch, Eyles and Elliott2005). The alternative view, to which this study belongs, is interested in the compositional effect and measures access to psycho-social resources at the individual level rather than the availability of these resources at the aggregate level.Footnote ¹

Islam et al. (2006) identified nine published articles on the link between individual access to social capital and individual health. After this literature review was published, Dunn et al. (2006) published a study analyzing self-assessed position in the reference group and self-assessed health based on individual-level Canadian data. We also identified Lavis and Stoddart (2003), not mentioned in Islam et al. (2006), and two studies linking social support and job-related demand to health in France (Paterniti et al. Reference Paterniti, Niedhammer, Lang and Consoli2002, Melchior et al. Reference Melchior, Niedhammer, Berkman and Goldberg2003, both based on a longitudinal survey of employees of the state-owned power company, EDF). We present briefly the main findings of these 13 studies (studies are summarized in Table 1 according to country of observation, the variables used to measure health, and the type of psycho-social resources entered in the model).

Table 1.

Previous studies – main characteristics

First, psycho-social resources are measured and defined in a variety of ways across these studies, reflecting the lack of consensus among social scientists (even among economists) on what social capital really is (Scheffler, this issue). Second, all studies show at least one significant positive relationship between access to psycho-social resources and current health. However, there is no general agreement about which type of psycho-social resource (social support, social capital, or sense of control and perceived position in the social hierarchy) is more strongly associated with health. Only one study (Rose, Reference Rose2000) finds a negative impact of access to social support on health: searching information from friends increases the likelihood of being in poor health in Russia, whereas trust and sense of control have the expected positive impact on health. Interestingly, this is also the only study attempting to measure all three types of psycho-social resources. Lavis and Stoddart (2003) compare trust and civic engagement and find the former dimension of social capital to be strongly correlated with health, but not the latter. They also show that France is one of the seven countries included in their study where the correlation is the weakest. Causal relationship is suggested in three longitudinal studies. The risk of coronary heart disease remains significantly higher in a ten-year follow-up among individuals with low participation in Sweden (Sundquist et al., Reference Sundquist, Lindström, Malmström, Johansson and Sundquist2004). In France, depressive symptoms and the probability of being on sick leave are higher in a three-year follow up among employees with perceived high job demands, low social support, and, for male workers only, low sense of control at work, controlling for occupation and personality traits (Paterniti et al. 2002, Melchior et al. 2003). One study (Liukkonen et al., Reference Liukkonen, Virtanen, Kivimaki, Pentti and Vahtera2004) cannot find any causal impact of perceived security of employment and the level of expected support from co-workers on self-assessed health and psychological distress in a four-year follow-up study of public employees in Finland.

Data and method

Data

The analysis is based on a population survey, representative of the French population, the Health, Health Care and Insurance Survey (ESPS: ‘Enquête sur la santé et la protection sociale’), coordinated by the Institute for Research and Information in Health Economics (IRDES). We use the 2004 wave, which included a set of questions on psycho-social resources.

The survey sample, comprised of 8,141 households, is based on a random draw from administrative files of the main sickness funds to which over 90% of the population living in France belong (Allonier et al., Reference Allison2006). Individuals drawn at random from the administrative files are used to identify households. The socio-economic questionnaire is answered by one key informant in each household (aged at least 18), who needs not be the individual selected at random and self-selects voluntarily. Questions on health status are collected through a self-administered questionnaire completed individually by each household member. Questions on psycho-social resources are answered by the key informant for him or herself only. All key informants were asked questions on civic engagement, trust, and social support, but only employed key informants were asked questions on sense of control at work and comparisons within the reference group, since these dimensions make sense for this population only.

Since our main objective is to assess the respective roles of a variety of psycho-social resources on health status, we restrict our analysis to the population of employed individuals aged 18 to 64, who reported their health status: 3,489 individuals (1,418 males and 2,071 females).Footnote ² This sample is representative of the population of employed key informants, aged 18 to 64, and not of the general population living in France. Table 2 below shows the main characteristics of the sample and it appears that women are over-represented as well as individuals from higher socio-economic status. These biases are controlled for in our analyses and there is no strong reason to suspect that unobserved response biases could affect the associations between psycho-social resources and health.

Table 2.

Descriptive statistics: socio-demographic characteristics of the sample

Psycho-social resources measures

Our main objective was to assess the respective roles of a variety of psycho-social resources in the variations of health across individuals. Most studies pick one dimension (social capital, social support, or sense of control) and measure its association with health, but we wanted to compare all these three dimensions with each other in their effect on health. The objective was therefore to capture as many psycho-social resources as possible, subject to the constraint that the survey already includes many questions and cannot exceed a reasonable length. As a result, we chose to assess our three dimensions (social capital, social support, and position in the social order) with two or three questions each (hence six to nine questions added to the survey overall). We present the relevant section of the questionnaire in Appendix 1.

To help us in this endeavour we invited a group of international experts in the field to a series of workshops organized around three issues: how to ask about access to social capital and social support, and about sense of control over one’s life?

Social capital is often measured at the individual level through civic engagement (participation to collective endeavours) and/or level of trust in the community, following Putnam’s (Reference Putnam1993) definition that social capital ‘refers to features of social organization, such as trust, norms and networks that can improve the efficiency of society by facilitating coordinated actions’. We asked two questions on civic engagement and one on trust.

For civic engagement, we use one question on participation (including political groups) plus one to distinguish participation as member and participation as person in charge to measure the degree of involvement of the individual in collective activities. The rationale for including political groups or unions is that in France participation in religious associations is not as common as in North America. These are standard questions in questionnaires on social capital, validated in terms of reliability and internal consistency (Blaxter et al., Reference Blaxter, Poland and Curran2001; Putnam, Reference Putnam2001).

Trust is measured in surveys through a mind experiment (would you trust your neighbours to look after your children?) or as a broad assessment of how trustworthy are others in general (Scheffler, this issue).Footnote ³ We opted for a mind experiment that is not restricted to parents of young children and involves institutions as well as other individuals (La Porta et al., Reference La Porta, Lopez-de-Silanes and Shleifer1997 and Rothstein, Reference Rothstein2000 show that trust in institutions is an important component of psycho-social resources): what to do in case of losing a wallet? Moreover we wanted to measure the level of reliance on institutions as they exist, where the individual thinks they are relevant (be it where s/he works or lives) and not only in the residence area. There is no validation for this variable so far.

For social and emotional support, we used a series of standard questions, adapted from the GAZEL survey in France (Melchior et al., 2003) and the GLOBE study in the Netherlands. A first question investigates the number of contacts and a second one asks about emotional support and the reasons for the lack of it (Berkman and Glass, 2000).

Sense of control is less often measured in general population surveys. We wanted to measure the sense the individual has of his/her position in the social hierarchy, and whether he/she can control his/her life. The first dimension can be measured as relative deprivation, compared to an objective group of reference (similar individuals). The underlying reasoning here is that feelings of under-achievement can be the individual pathway through which income or social inequalities affect health and explain why in more unequal societies individuals at the lower end of social hierarchy end up in poorer health, even if they are above the absolute material poverty threshold. In the literature, such groups are comprised of people of the same age and educational level (Eibner and Evans, Reference Eibner and Evans2005), and we summarized these characteristics by asking about people with the same skills; therefore, we measure here a sense of being treated fairly or of discrepancies between effort and reward and we expect that perceived discrepancies signal a sense that the individual does not control his/her life. We also wanted to measure the impact of confrontations with social hierarchies on health, as stated in Marmot and Wilkinson, (2005) (see also Ellaway et al., Reference Ellaway, McKay, Macintyre, Kearns and Hiscock2004, for social comparisons of homes and psycho-social health): a perception of being dominated rather than dominant can diminish the sense of control and autonomy and be detrimental for health. Therefore we asked respondents whether they compared their income with other people (self-selected group of reference) and what was the outcome of the comparison – the reliability and internal consistency of the question was not tested. A third question adapted from the GAZEL survey (Melchior et al., 2003) asked about the sense of control at work. Descriptive statistics on these measures are presented in Table 3.

Table 3.

Descriptive statistics: psycho-social determinants of health

Note: (*) Interviewers suggested possible contacts from a list including parents, parents in law, children, siblings, grand-parents, other relatives, friends, neighbours, work-mates, others, and respondents answered yes or no.

Findings

The findings of the baseline analysis of the material and biological determinants of health (model 1, Table 4) show that the probability of reporting poor or bad health increases with age and is higher for females; all socio-economic variables have a significant effect of expected sign. The only socio-economic variable that is significant at the 5% level is income, less well-off individuals being more likely to be in the poorer health category. Some educational and SES categories are significant at the 10% level: individuals with primary educational level report poorer health status compared to individuals with post-secondary education level and skilled and unskilled blue-collar workers compared to professionals, managers, and intellectual professions.

Table 4.

Determinants of the probability of being in poor self-assessed health, odds ratios associated to material and psycho-social determinants of health (multivariate analysis) – employed population

Notes: Significance levels: *10%, ** 5%, *** 1%, ****0.1%.

The second column of Table 4 gives the results of the second analysis where all psycho-social resources are entered simultaneously in the regression of self-assessed health, controlling for age, sex, SES, education level, and household income among employed individuals. Compared to our baseline analysis where only socio-economic determinants (SES, education, and income) are entered, only income remains significant once psycho-social resources are introduced. Introducing psycho-social resources into the model substantially improves the fit: the likelihood ratio test confirms that the two models are significantly different at 0.001 % and the max-rescaled R-square increases by 24.6%. A test of multicollinearity in the set of independent variables has been performed (tolerance and variance inflation factor, see Allison, 1999) and concludes that the correlation structure is not such that it may affect the robustness of our findings, despite the clear correlation across variables describing access to material and psycho-social resources.

Within psycho-social resources, the strongest and most precisely estimated effect is that of the sense of control on work: individuals who strongly disagree that they have autonomy in their work have a much higher relative risk of being in the poorer health category (with an odds ratio equal to 1.9 and significantly different from 0 at the 0.1% level). This effect can partly explain the absence of any significant effect of SES.

Second to sense of control is social support: lacking emotional support is strongly associated with the probability of being in the poorer health category (with an odds ratio greater than 1.5 but at the 5% level only). The lack of civic engagement is also associated with poorer health category with an odds ratio of 1.4 (at the 0.1% level). Findings for the influence of the sense of one’s position in social hierarchy are less intuitive: making more or less than one’s reference group are both associated with a higher probability of being in the poorer health category (with odds ratios around 1.5, but at the 10% level of significance only for making less). On the contrary, perceived relative deprivation (making less than individuals with the same skills) and reliance on the community (believing that it is useful to go to the lost and found for a lost wallet) don’t have any significant association with self-assessed health.

Conducting separate analyses for both sexes we found associations of the same sign and magnitude (Table 5). The only difference is that social support does not work identically for men and women: we are able to measure an influence of the number of contacts on health for men (more contacts decrease the probability of declaring poorer health) but emotional support influences health for women only.

Table 5.

Determinants of the probability of being in poor self-assessed health, odds ratios associated to material and psycho-social determinants of health (multivariate analysis) – Employed population – men and women separately

Notes: Significance levels: *10%, ** 5%, *** 1%, ****0.1%.

The analysis of the determinants of poor access to psycho-social resources (Tables 6a and 6b) confirms that individuals with higher levels of education, SES or income also benefit on average from access to significantly better psycho-social resources: They participate significantly more often, rely significantly more on their community, have more emotional support and have a significantly better sense of control at work. The main exceptions are comparisons with peers or within the reference group for which intermediary groups (2nd level of secondary school or quintiles 3 and 4 of income) are at a disadvantage compared to extreme groups.

Table 6a.

Determinants of access to psycho-social resources (employed population, 18–64)

Notes: Significance levels: *10%, ** 5%, *** 1%, ****0.1%.

Table 6b.

Determinants of access to psycho-social resources (employed population, 18–64)

Notes: Significance levels: *10%, ** 5%, *** 1%, ****0.1%.

Discussion

Using a unique set of measures of psycho-social factors at the individual level this study provides empirical evidence of the link between the subjective perception of some psycho-social resources and health in France, controlling for other, more material, factors of health, such as income, age, gender, education, and SES. Since we also show that access to these resources is uneven across the population and strongly influenced by social status, those factors can partly explain the high level of social health inequalities in France.

Our results are consistent with several previous studies. As suggested by Putnam (1993), social capital, assessed by civic engagement, seems to have a protective effect on health, even though community trust is not associated with self-assessed health. Consistent with the findings of Berkman and Glass (Reference Berkman, Glass, Berkman and Kawachi2000), health status is positively associated with access to emotional support, rather than with the size of social networks. We find a weak association between self-assessed health and relative deprivation, whatever the reference group considered (peers, family, or friends) contrary to the conclusions of Dunn et al. (Reference Dunn, Veenstra and Ross2006), Deaton (Reference Deaton2001), Eibner and Evans (2005), and Elstad et al. (Reference Elstad, Dahl and Hofoss2006), but similar to Miller and Paxson (Reference Miller and Paxson2006). Finally, our main finding is that, controlling for all other resources, the lack of sense of control at work is strongly associated with a higher probability of being in poor or bad health and seems to be the most important psycho-social resource to be studied in a causal analysis of social inequalities in health, in accord with the model proposed by Karasek (Karasek, Reference Karasek1979, Karasek and Theorell, Reference Karasek and Theorell1990).

If the associations we measure indicate a causal pathway from psycho-social resources to health (which, to repeat again, we do not claim we have demonstrated), public health policies aimed at reducing social inequalities of health in France should work on reducing the sense of disenfranchising (lack of sense of control over one’s life and lack of civic engagement) as well as at reducing income inequalities, rather than enhancing the level of institutional trust. As recently and dramatically demonstrated by the riots in the fall of 2005, localized in a small number of urban areas, and less dramatically by the ILO survey on discrimination by employers in France (Cediey and Foroni, Reference Cediey and Foroni2006), the unequal distribution of rights and blatant discrimination based on ethnicity and immigrant status are important factors of unequal access to psycho-social resources in France. Thus, public policies should seek to address unequal access to psycho-social resources and this could, in turn, help to reduce health inequalities.

However, the causal nature of the association is a big if: based on currently available evidence we cannot go beyond associations and the next steps in our research agenda will be to overcome biases precluding the causal interpretation of our findings. First, using a variety of less subjective measures of health (chronic conditions, risk factors, clinicians’ assessments, and health-related behaviours) we will test that the association between health and the lack of psycho-social resources is not due to a ‘pessimistic bias’ (the same individuals tend to pessimistically assess both their health status and their capacity to muster social resources). Second, we will use subsequent waves of the data to exploit the longitudinal nature of the sample: if we can show that changes in health (controlling for baseline health) are linked to the initial level psycho-social resources, this will provide further evidence of a causal impact of social capital on health. Third, we will use aggregate level measures of psycho-social resources as instruments (likely not influenced by individual health) to further limit the simultaneity bias between individual health and individual access to social capital.

In case these further investigations prove the causal pathway behind our observed associations, we will conduct a partial effects analysis in order to assess to what extent social inequalities in health can be explained by social inequalities in access to psycho-social resources.

Last, we will investigate the determinants of access to psycho-social resources, more specifically those based on residence, race-ethnicity, immigrant status, language, and religion, which are seen as major social markers in contemporary France.