How does bridging social capital relate to health-behavior, overweight and obesity among low and high educated groups? A cross-sectional analysis of GLOBE-2014

Educational level is an important indicator of social stratification in contemporary Dutch society [15, 17, 28]. Survey participants reported their highest attained educational level, which was classified according to the International Standard Classification of Education (ISCED): 1– high education (tertiary education (ISCED 5–7)); 2– mid education (upper secondary education (ISCED 3–4)); 3– low education (primary education and lower secondary education (ISCED 0–2)).

Education-specific bridging social capital was measured with the question: “How many of your close friends have the same educational level as you have?”, with five answering options: all, most of them, about half, some, and none. Bridging social capital was coded as ‘1 = bridging’ for those who answered ‘about half’, ‘some’, or ‘none of my friends’, and ‘0 = bonding’ for those who answered ‘all’ or ‘most of my friends’.

Sports participation was measured by means of the validated Short QUestionnaire to ASsess Health-enhancing physical activity (SQUASH) [29]. Participants were asked to think about their sports participation over the past month and to write down up to four different types of sports they had participated in on a weekly basis. For each type of sport they reported frequency (days per week), duration (hours and minutes per day) and intensity (light, moderate, intense). Self-reported intensity and activity-specific intensity Metabolic Equivalents (METs) were used to calculate the numbers of days participants participated in sports for at least 30 min at moderate or vigorous intensity (moderate intensity = 4–6 MET for 18–55 years and 3–5 MET for 55+ years). The variable was dichotomized into 1) no sports participation at least once per week for > 30 min at moderate intensity, vs. 0) sports participation at least once per week for > 30 min at moderate intensity (reference group).

Walking and cycling in leisure time were also measured as part of the SQUASH. Participants reported frequency (days per week), duration (hours and minutes per day) and intensity (light, moderate, intense) for walking and cycling in their leisure time. For both activities separately, we calculated how many days per week the participant walked and cycled for at least 30 min at moderate intensity. Both variables were dichotomized: 1) no walking at least once per week for > 30 min at moderate intensity, vs. 0) walking at least once per week for > 30 min at moderate intensity (reference group); and 1) no cycling at least once per week for > 30 min at moderate intensity, vs. 0) cycling at least once per week for > 30 min at moderate intensity (reference group).

Fruit and vegetable consumption were measured as part of a food frequency questionnaire [30]. Participants reported frequency (days per week) over the past month, for fruit and vegetable consumption separately. They also indicated the portion size (i.e. number of pieces of fruit, and number of serving spoons [=50 g] of vegetables) on a typical occasion. Based on this, we calculated whether participants did or did not meet recommendations for fruit and vegetable consumption. Those who did not consume two pieces of fruit every day were coded as “not meeting recommended fruit intakes” (and those eating two or more pieces of fruit every day were used as reference group). Those who not consumed 200 g of vegetables every day were coded as “not meeting recommended vegetable intakes” (with eating > 200 g of vegetables every day as reference group).

Water intake was also part of the food frequency questionnaire. Participants reported how many days per week, over the past month, they consumed tap water or spring water. Water consumption less than daily was coded as ‘no daily water intake’ (with ‘daily water intake’ as reference group). Further, participants reported how many days per week, over the past month, they consumed meat. Meat intake on 5 till 7 days per week was coded as ‘high meat intake’, whereas meat intake on less than 5 days per week was coded ‘no high meat intake’ (reference group).

Body mass index (BMI) was calculated by self-reported height and weight. Participants with a BMI higher than 25 were categorized as overweight (and BMI < 25 as no overweight, reference group). A BMI higher than 30 was categorized as obesity (and BMI < 30 as no obesity, reference group). Overweight and obesity were applied as outcomes since these are related to two types of health-behaviour: diet and physical activity.

Potential confounders were included in all analyses: sex (male, female), age (in 10-year age groups), country of birth (Netherlands, other), living together with a partner (yes, no), children living in your household (yes, no), employment status (employed, unemployed, retired, or other (e.g. homemaker, student)), father’s highest attained educational level, and mother’s highest attained educational level. The latter two were included as confounders, since parental education level could influence the likelihood that low educated become friends with higher educated, and the likelihood to behave healthily (and the opposite for high educated becoming friends with lower educated). Low educated who were raised by high-educated parents may be more likely to get in touch with higher educated (e.g. directly, via their parents connections, or indirectly, as they have learned to speak the ‘appropriate’ language to connect with higher educated [31]). Also, low educated with high educated parents may be raised in a sociocultural environment in which healthy behavior was more prevalent.

Participants with missing values for educational level or the confounders sex, age, country of birth, living with a partner, children, or employment status, were excluded from the analyses (n = 110). This resulted in an analytic sample of N = 2702. To estimate the main effects of bridging social capital and educational level on the outcomes, a logistic regression model was run for each outcome separately, including bridging social capital, educational level, and confounders (results in Additional file 1). To examine whether the association between bridging social capital and health-behavior was modified by educational level, Knol and VanderWeele’s template for reporting analysis investigating effect modifications was used [32]. Their recommendations include running three different logistic regression models for each outcome (referred to as model A, B and C below), in order to provide readers with sufficient information to draw conclusions on the size and statistical significance of the effect modification [32]. Model A is most commonly used in the field of public health for calculating effect modification (or interaction), namely a model with educational level, bridging social capital and confounders, plus a multiplicative interaction term between educational level and bridging social capital. The results from model A are presented in the footnotes of Tables 2-4, and used as an overall test of whether education significantly modifies the association between bridging social capital and outcomes. In model B, odds ratios (ORs) for each combination of educational level and bridging capital were calculated. Thereto, we first composed a combined ‘education*capital’-variable, resulting in a variable with six categories (1 = high education with bonding capital (reference group); 2 = high education with bridging capital; 3 = mid-education with bonding capital; 4 = mid education with bridging capital; 5 = low education with bonding capital; 6 = low education with bridging capital). In model B, this combined education*capital-variable and confounders were included. In model C, bridging social capital and confounders were included in a regression model, and results were stratified by educational level. In this way, model C produced ORs for the association between bridging social capital and the outcomes for each educational group separately (low, mid and high educated). All statistical analyses were conducted in SPSS 23.0.

Different relations of bridging social capital with health-behavior, overweight and obesity for low and high educated were found. This partly confirmed our hypothesis that bridging social capital would have beneficial relations with health-behavioral outcomes for low educated, but not for high educated. Indeed, among low educated, bridging social capital reduced the likelihood of overweight and obesity, but did not reduce the likelihood of unhealthy behavior. Among high educated, bridging social capital increased the likelihood of some unhealthy behaviors: smoking, low cycling levels, low vegetable intake, and high meat intake.

Research into bridging social capital suffers from the lack of a standardized measurement approach [13], which makes it difficult to compare our results to previous studies. A recent review found evidence that collective social capital had a stronger positive association with good health for people with a lower socioeconomic status, and may buffer against the negative health effects of a low socioeconomic status [3]. This parallels our findings with respect to overweight and obesity. Two Japanese studies and a British study were somewhat similar to our study in the sense that these measured individual-level bridging social as some degree of heterogeneity in social contacts [12, 33, 34]. Where we found that bridging social capital increased the odds of some types of unhealthy behavior (but had no relations with other), these studies showed that bridging social capital had beneficial relations with several outcomes: a lower odds of physical inactivity [33], inverse associations with depressive mood [12], and positive associations with self-rated health [34]. Differences in the measurement of bridging social capital likely contribute to the contrasting results: whereas we measured bridging social capital referring to (dis) similarities in the educational level of close friends, other studies referred to more general connections (network connections, participants of community activities), and (dis) similarities in ‘social characteristics’ in general (e.g. age, sex) [12, 33] or ethnicity and income [34].

Our study extends previous research by investigating whether associations of bridging social capital with health-behavior differ for educational groups. Indeed, bridging social capital increased the likelihood of smoking, low cycling levels, low vegetable intake, and high meat intake among high educated groups, but was unrelated to health-behavior of low educated. An explanation for this may be that healthy behavior, for most people, requires more effort and constraint than unhealthy behavior (e.g. it is often easier to be inactive than sufficiently active, and it is easier to eat few than recommended amounts of vegetables). It seems as if behaving healthily is most ‘doable’ for high educated with high educated friends (i.e. with bonding social capital), possibly because they experience stronger descriptive norms, and more social support and role-modelling to adopt and maintain healthy behavior, than high educated with low educated friends. High educated with bridging social capital (i.e. with lower educated friends) may perceive less social support, or less strict descriptive norms, making them more likely to adopt unhealthy behavior.

Our results for overweight and obesity seem in contrast to this explanation: bridging social capital did reduce the likelihood to be overweight and obese among low educated, although one could similarly argue for overweight/obesity that the unhealthy option (putting on weight) is easier than the healthy option (weight control). However, there is an important difference between overweight/obesity and health-behavior that could play a role here: there is a strong slimness ideal in Western societies, and such a strong ideal is lacking for most types of health-behavior. Although this slimness ideal may be strongest among high educated, qualitative studies show that weight control and bodily appearance also matter to low educated [35, 36], whereas health-behavior on the other hand is more often rejected or opposed by low educated (e.g. ‘healthy food is tasteless, boring and insufficiently satisfying’) [36]. So, as interpretation of our findings for obesity, we surmise that these may show that the healthy norm outweighs the unhealthy norm. For this reason, it may be that low educated with higher educated friends have a reduced risk of obesity (instead of high educated with low educated friends having an increased risk of obesity - what we observed for health behavior).

The previous interpretations assume a causal relation between bridging capital and health-behavior. However, as the cross-sectional design of our study does not give insight in the direction of the observed relations, another plausible explanation may be that ‘like attracts like’: low educated with a healthy lifestyle may 'select' friends with a similar healthy lifestyle, who are more likely to be higher educated. Studies from sociology and social psychology show that, before all, the composition and structure of personal networks is affected by the social contexts that a person enters during daily life, such as the work place, school and voluntary associations [37]. Who then, of all people you meet within these contexts, become your friends, is further determined by similarities in age, sex, ethnicity, educational level [38], and personality [39]. The role of people’s health-behavior in the process of who becomes friends with whom, is less well known.

Since, in the Netherlands, educational inequalities in health-behavior are larger than those by income (and occupation), we expected education-specific bridging links to matter more for health-behavior than income-specific bridging links. We tested this assumption in additional analyses. In our survey, respondents indicated their income level, and how many close friends had a similar income level. In additional analyses, we tested whether income level modified the association between income-specific bridging capital and health-behavior. No meaningful effect modification was found (results in Additional file 2). Apparently, in the Netherlands, educational level is the crucial dimension of bridging social capital in relation to health-behavior inequalities. In other societies, in which race and income level are important dimensions of societal segregation, bridging social capital with respect to race and income level may be just as important.

An important strength of this study is that we conducted a more detailed measurement of bridging social capital compared to previous studies. Also, we tested whether associations of bridging social capital differed for educational groups. Thereby, this study took a differentiated approach to bridging capital, which has often been called for [3, 7, 13], providing more insight in possible underlying pathways between socioeconomic position, bridging social capital and health-behavior. Further, we measured multiple types of health-behavior, as well as overweight and obesity, which allowed us to rigorously test our hypothesis. Lastly, we controlled our cross-sectional analyses for a wide range of potential confounders, including parental education. In this way, we tried to rule out confounding effects as much as possible, by controlling for factors that may affected the respondents’ educational level, bridging social capital and/or health-behavior.