Data were collected during the baseline examination of the German Heinz Nixdorf Recall (HNR) Study, an ongoing prospective population-based cohort study in Germany. The rationale, design and methods of this Study have been described in detail elsewhere 1415. The study base was the German population aged 45 to 74 years, living in three cities in an industrialized urban region (Ruhr Area). Participants were recruited from a random sample derived from mandatory citizen registries. 4814 men and women agreed to participate, which corresponds to a response proportion of 56% 15. Comprehensive baseline examinations were conducted from December 2000 to August 2003. A five-year follow up is currently under way. The main aim of the HNR study is to improve prediction of coronary heart disease by combining established with new cardiovascular risk factors. Socio-economic position and social relations were included into the baseline screening as part of a psychosocial risk factor assessment through face-to-face interviews and paper and pencil questionnaires.

The association of socio-economic position with social networks and support was tested in two ways: (a) mean values of close ties were compared across SEP categories by bivariate analyses; test for trend (p) is based on Jonckheere Terpstra test. Since we assume that age and gender interact with socio-economic position, we have also stratified our analyses accordingly in order to control respective confounding effects; (b) the indicators of poor social networks and support were regressed on education and income by multivariate logistic regression analysis. The reference category in these analyses was good social networks and support. Multivariate analyses were adjusted for age and sex. All statistical analyses were carried out using SPSS 12.0 programme.

Table 2 illustrates the association of socio-economic position with different indicators of social networks and support. In the respective left hand column, numbers (N) and percentages (%) of SEP groups reporting poor networks/support are represented; in the respective right hand column, odds ratios (OR) and 95% confidence intervals (CI) are represented, adjusted for age and sex. Lower SEP groups are compared to highest group.

Findings show that socially disadvantaged persons are more often exposed to poor social networks and social support. In bivariate analyses it becomes obvious that there is a higher percentage of having no confidant and no partner, of no participation, of being socially isolated and of lacking social support. In multivariate analyses, odds ratios are elevated in low SEP groups. Based on education, odds ratios range from 1.0 (highest education) to 4.9 (lowest education). Findings based on income show similar effects, ranging from 1.0 to 2.5. In some cases (no confidant, poor social support), the association of SEP with poor social networks and support gets stronger with decreasing SEP, but by far the strongest effects are observed in most disadvantaged groups.

Structural measures (SII) show effects comparable with qualitative support measures. If we decompose the Social Integration Index, we see that having no confidant shows the strongest effect in both SEP groups; the lowest effect was found for partnership in education and for participation in income groups.

Figures 1 and 2 depict the mean values of close ties compared across SEP categories education and income. Figure 1 shows that the total number of close ties (a) significantly diminishes with decreasing educational status (p for trend = .000). Concerning the number of those living close-by (b) no such relationship is observed. Here, we find an insignificant u-shaped association (p for trend = .105). As regards those regularly seen (c) we find the same u-shaped association slightly below significance level (p for trend = .047). It is of interest to note again a positive relationship of educational status with close ties via communication media (d; p for trend = .000). The same pattern is observed for the second socio-economic indicator income (figure 2). To rule out possible age and gender effects, we stratified the analyses accordingly and results remained basically unchanged (results not shown).

It becomes obvious that there is a difference between the number of reported ties and the number of ties with persons who are actually seen. Another interesting finding is that this difference widens with increasing socio-economic position.

At the methodological level it should be mentioned that standard deviations of means are rather large. Therefore, we repeated our analyses using the median number of close ties. Again, results remain basically unchanged.

A particular strength of the study is given by the fact that it is based on a large community sample rather than on a specific population such as an occupational cohort. Furthermore, within our sample we controlled for sample bias in terms of status-specific refusal rates. In fact, participation rates were lower among lower SEP people, but with the help of an additional subgroup analysis of non-responders we were able to identify the observed association with low social participation in this non-responder group as well 15.

Another strength concerns the measurement of variables of social relationships. Instead of using non-systematic or proximate indicators we applied original validated questionnaires assessing a comprehensive range of indicators of social relationships with internationally comparable items 171819. The Heinz-Nixdorf-Recall-Study has been externally certified, thus confirming a high quality of data collection and data handling.

These strengths are balanced by several methodological limitations. First, we cannot rule out a social desirability effect that may weaken the validity of some indicators, such as reported household income 2021, or the assessment of close social relationships. Moreover, the concept of 'friendship' may be more familiar to middle class compared to lower class people 2223 although we provided a clear description of the concept during the interview.

A second limitation concerns our choice of socio-economic indicators. Since only one third of the study population was working at the time of the interview, we excluded occupational status as an additional SEP indicator. In a subgroup analysis where occupation was classified according to ISCO-88 and, additionally, according to the Treiman-prestige-scale 24, a less consistent pattern was observed with regard to social networks and support. Low occupational status was significantly associated with having no confident and with lack of participation, but was unrelated to the remaining measures of social relationships. Moreover, important sociodemographic factors such as age and gender were analysed as confounders rather than variables subjected to indepth-subgroup analysis. Future studies should complement the current approach.

Finally, in the cross-sectional study, no information on the temporal relation between socio-economic position and social networks/support was available. While we hypothesize that socio-economic position to some extent determines the range and quality of social relationships, we cannot exclude a reverse association. The prospective design of this study will allow us to test the direction of associations in future analyses.

This rather consistent association of SEP with social relations is supported by another German study which recently identified 8 per cent of the German population as living and working under precarious conditions with accumulated material and psychosocial disadvantage 25. Interventions to tackle social isolation should be carried out in disadvantaged settings and in highly vulnerable groups, such as elementary schools or deprived urban settings, and in unemployed people or lone mothers. Some instructive examples can be found in 26.