Data
Our study is a secondary analysis of data collected by the FFCWB study. We used publicly available data and registered for its use through Princeton University’s Office of Population Research. The FFCWB study follows a cohort of about 5,000 children born to low-income parents in 20 major U.S. cities (in 15 states), during the years 1998–2000 and who have been surveyed at regular intervals since birth. By design approximately three-quarters of the mothers were unmarried. Face-to-face interviews were conducted with 4898 mothers shortly after giving birth. Both parents were interviewed at the time of their infant’s birth and then again one, three, five, and nine years later. The study was designed to provide information on the conditions and capabilities of new parents, the determinants and trajectories of parental relationships, and the consequences for parents and their children of health, child welfare, and social service policies as well as other aspects of their environment [31].
We use data from the five and nine-year post-baseline follow-up interviews to assess the relationship between social capital and health outcomes. It is important to note that the questions related to social capital were not consistently measured across all waves of the FFCWB study which limits our ability to assess changes in social capital over time in relationship to health. We use data from only the mothers with regard to social capital and health, as there was a disproportionate response rate for the fathers beyond the initial wave of the FFCWB study.
Measures
Dependent variable - self-rated health
We measured our dependent variable, SRH, using the data obtained during the nine-year wave of the survey. Mothers were asked to rate their overall general health on a five-point scale ranging from excellent to poor. Self-reported health has been shown to be a valid measure of general health and its use is consistent across studies examining the relationship between social capital and health. The five-point scale was recoded into a dichotomous variable ‘favorable’ (excellent, very good, good) and ‘unfavorable’ (fair, poor) health.
Measures of social capital
Based on individual’s responses to questions from the five-year follow up survey of mothers, four indices of social capital were created. The inclusion of questions within each index was informed by the 2006 Social Capital Community Survey [32]. Each index is described below.
Social support and trust
Mothers were asked about whether they had others in their life to provide them with emotional and tangible supports when needed. Specifically, mothers were asked whether they had someone that 1) they could trust to look after their child if they were away, 2) would loan them $200, 3) could provide them a place to live 4) could provide them with emergency child care, 5) co-sign for a bank loan ($1,000) and 6) they could share confidence with. Response options for each question were dichotomous (yes/no). The constructed measure of social support and trust is a summation of the “yes” (=1) responses to each question.
Social participation
This measure counts the number of “yes” responses by mothers to questions related to participation in various community entities. The three questions included participation in activities at the child’s school, community groups, and religious services. For the question on religious services, responses of once a week or more were characterized as participation for this index. A response of yes to any question indicated involvement or participation in the entity during the 12 months prior to completion of the survey.
Perceptions of neighborhood social cohesion
The FFCWB uses a validated measure of social cohesion [33]. The five questions comprising this index ask for the respondent’s level of agreement with the following statements: 1) willingness to help a fellow neighbor, 2) if neighborhood was viewed as a close-knit community, 3) if people generally get along with each other, 4) if they share the same values and, 5) if gangs were a problem in the neighborhood. The question regarding gangs being a problem was recoded so that lower levels of agreement were coded as being more positive. After recoding, the mean response to the Likert-type questions expressing level of agreement was used to construct the index, with higher scores indicating higher levels of neighborhood social cohesion.
Perceived neighborhood social control
The measure of perceived neighborhood social control also uses a validated measure within the collective efficacy construct and is constructed from five questions [33]. Using a Likert-type scale, mothers were asked about the likelihood of neighbors to intervene if children are 1) skipping school, 2) spray painting a building and, 3) showing disrespect to an adult. Additionally, the likelihood of a neighbor intervening to diffuse a fight, as well as a neighbor’s willingness to intervene to save a local firehouse were assessed and utilized. The index was constructed as a mean, based on an individual’s response to the above questions, with higher scores indicating higher levels of perceived neighborhood social control. While the four constructed measures of social capital are related, the pairwise correlation values are muted; the highest degree of correlation (0.48) is found between Perceived Neighborhood Social Cohesion and Perceived Neighborhood Social Control, while the remaining correlations are below 0.24.
Explanatory variables
Several socioeconomic, demographic, and behavioral variables are known to influence self-rated health. Such variables, which were controlled for, included level of education, age, race, income, relationship status, employment status, number of children, poverty, smoking behavior, and SRH at (t-1). Education was measured as highest educational level obtained. Age was categorized into four categories <25 years, 25–34 years, 35–44 years and 45 years and above. Race is categorized by the FFCWB as “black” “Hispanic” “white” or “other.” Income was categorized into three categories of less than $30,000, $30,000–$59,999, and $60,000 and above. Since income was missing a fourth category “missing income” was included. Relationship status was measured as two separate binary variables as to married or not married, cohabitating or not cohabitating to get at the diversity of relationship types among women in the sample. Employment status was a binary measure “yes” or “no.” The number of children the woman has was a continuous measure. Poverty status was a binary measure equaling one if the individual has received welfare or food stamps in the previous 12 months. Whether the person has smoked within the last 30 days was a binary “yes” or “no” measure. SRH at (t-1) was categorized the same way the dependent variable was categorized. All measures, with the exception of education which comes from the baseline interview, come from the five-year follow-up interview (t-1).
Analysis
Our empirical model is adapted from Bolin et al.’s theoretical model with the family as producer of social capital and health, and with the amount of social capital being positively related to level of health [34]. We account for the temporal ordering using measures of social capital from an earlier time point in reference to health measures; interest lies in examining the temporal relationship over a four-year period between the explanatory variables, lagged one period (t-1), and the current period for self-reported health. In particular, we tested the hypothesis that social capital at time (t-1) is positively associated with SRH at time (t). To examine this hypothesis a logistic regression model is estimated, including socio-economic and demographic variables. The equation for the logistic model is given as:
$$ \mathrm{Logit}\left({H}_t\right)=\beta {X}_{t-1}+\varphi {S}_{t-1}+{\varepsilon}_t $$
(1)
where H
t
is SRH in period t, X
t-1
includes socio-economic and demographic variables in period t-1, and S
t-1
represents the individual measure of social capital, in period t −1. The hypothesis is represented by φ, which is expected to be positive. Each dimension of social capital was included individually in a model examining its impact on SRH, while controlling for the covariates (models 1–4). A fifth model which includes all measures of social capital was examined. In a similar vein, ordered logistic regression models were estimated as a robustness check. Marginal effects were calculated for each social capital construct on SRH.