This is one of the first studies that systematically examined the mediating effect of social relationships on the association between SES and subjective health using data from a 5 year follow-up. Our results indicate a mediating effect of social relationships, i.e. social relations contribute to the explanation of socioeconomic inequalities in subjective health. When measures of social relations were introduced as mediators into the regression models, percentage reductions of the odds ratios between 2% and 30% were observed in the overall sample. In most cases percent reductions exceeded 10%. If the associations between indicators of SES and general subjective health were to be independent of social relations, a variation in effect size would not have been found after the introduction of SII, emotional and instrumental support into the regression models.
Former studies have not consistently revealed such a mediating effect of social relationships on health. Some studies found only slight mediating effects of social relations [3
], while others showed no contribution of social relations to the explanation of health inequalities [24
]. However, these studies differ in terms of study design, measurement of social relations and health as well as study region.
Regarding gender differences, SES indicators are differently associated with subjective health in men and women. On the one hand, a low equivalent household income leads to stronger OR for poor subjective health in men than in women. On the other hand, less than 10 years of education are more strongly related to poor subjective health in women than in men (see Tables and ). Especially with regard to the specific age-group of the study, which is characterised by a lower degree of labour participation in women, one could imagine that income is of higher importance to men, as it might more directly reflect their success as “breadwinners”. The social status of women in this age-group might be more accurately assessed by their educational background. This might help to explain, why these indicators are differently associated with health in men and women. The mediator analyses revealed similar results as mediating effects of social relationships were detected for both men and women, though varying in effect size. For example, when equivalent household income was used as SES measure, a percentage reduction of up to 50% was found in women, while in men the percentage reduction was 15% in the lowest income group (see Tables and ). Due to a reduced sample size in the gender-specific analyses, the effects in the first basic model rarely reached significance. Therefore, percentage reductions were not calculated in most cases. Generally, the introduction of social relations reduces the association of SES and subjective health in women and in men. Two exceptions can be found in women (see Table ). The introduction of instrumental and emotional support (Model 3) leads to a small percentage increase of the respective OR in women with less than 10 years of education. Secondly, the introduction of the SII leads to a 7% increase of the OR for poor health among female qualified employees compared to managers and professionals. As these increases are minor and no particular pattern can be observed, we are careful in drawing conclusions from these findings. In our view it is important to further investigate gender differences in the association between SES, social relations and health in future studies, because we are far from understanding the particular mechanisms in men and women [19
Regarding the explanatory contribution of social support and social integration, results are inconsistent. Overall, the simultaneous introduction of both aspects of social relations leads to largest percentage reductions. However, it remains unclear, which aspect of social relations contributes most to the explanation of inequalities in subjective health. For example when SES is measured by education, the introduction of the SII alone leads to percentage reductions of about 15%, while the introduction of instrumental and emotional support shows marginal percentage reductions of between 2% and 6% (see Table ). When SES is measured by occupational status, the introduction of social support leads to stronger percentage reductions in technicians and qualified employees than does the introduction of SII, while for the unskilled the opposite is true. We additionally analysed in which way the two indicators of functional aspects of social relations, emotional and instrumental support, contribute differently to the explanation of socioeconomic inequalities in subjective health. Hence, emotional and instrumental support were introduced separately into logistic regression models (results not shown). The explanatory contribution proved to be very similar, with no clear pattern of differences regarding the explanatory contribution of instrumental and emotional support.
Earlier research has led to a vast body of evidence showing different associations of functional and structural aspects of social relations with different health indicators [4
]. While structural aspects of social relations may facilitate the availability of help, social support might more directly affect health behaviour and psychological mechanisms such as feelings of self-esteem and coping [9
]. It has been highlighted that especially for ones feeling of accessibility of support and its effect on health it is important to distinguish between perceived and actually received support [44
]. Stansfeld and Fuhrer have developed several models to show how different facets of social relations may influence population health [11
]. In a meta-analytic review Holt-Lunstad and colleagues showed that especially a multidimensional assessment of social-relations led to strongest associations with mortality-risks, as they included the different pathways by which social relations influence health and mortality [21
Regarding the three SES indicators, similar results in the strength of the associations can be observed. Generally, the lowest SES groups have the highest risks of reporting poor subjective health. Moreover, in this group percentage reductions are largest when social relations are introduced into the logistic regression models. This is true for income, education as well as for occupational status.
In interpreting the presented results, several methodological aspects should be considered. It is a strength of our analyses that they are based on a cohort study. So far no study has been able to draw conclusions on the mediating effect of social relations on the association of SES and subjective health in a longitudinal perspective. Furthermore, special emphasis was put on quality control of data collection and data handling in the HNR study, as evidenced by external certification [33
]. Complex measures of social support indicators were used. When constructing the items for measuring support, both availability as well as adequacy of support were considered, as proposed in earlier research [18
On the other hand our results are limited as they are all based on self-reported data and do not include objectively measured health indicators. Therefore, a possible bias can not be ruled out. Also the longitudinal design is limited as we refer to one 5-year follow-up and include only two measurement points. As expected, there is only little variation in subjective health in a 5-year period. Moreover, we did not calculate significances of mediator effects by using the Sobel-Test as multiple mediators were included in our model. Another restriction is the high number of missing information on occupational status, especially in women (N
669). For those cases, no information on actual job status or occupational status before retirement was available. This might lead to bias in the results for this SES category. Furthermore, analyses draw on a specific study population, namely a sample of residents of the Ruhr Area aged 45 to 75 years. As noted earlier, the Ruhr Area is a region in transition. It is therefore possible, that in such uncertain times of change, social relationships play a particularly important role in protecting individuals from negative health effects of socioeconomic hardship. This may be one reason why mediating effects of social relationships found in our study are more consistent than in former studies. Therefore, our results cannot be generalised to the overall German population but maybe to populations living under similar conditions.