Background It is well known that individual-level social capital is positively associated with health, but most preceding studies have not fully controlled for an individual's time-invariant attributes, especially unobserved ones. The current study attempted to address how the association between individual-level social capital and health is confounded by an individual's unobserved time-invariant attributes.

Methods Data were collected from six-wave nationwide panel surveys conducted from 2005 to 2010, with 162 720 observations from 30 590 individuals. Individual-level bonding and bridging social capital, as well as their associations with self-rated health (SRH) and psychological distress (measured by Kessler 6 scores), were considered. Estimation results of cross-sectional, prospective cohort and fixed-effects logistic models were compared.

Results The OR of reporting poor SRH responding to high bonding social capital rose from 0.64 (95% CI 0.62 to 0.65) in the pooled cross-sectional model to 0.77 (0.75 to 0.80) in the prospective cohort model and 0.87 (0.82 to 0.92) in the fixed-effects model. Similar patterns were observed for bridging social capital, but the OR of reporting poor SRH became non-significant in the fixed-effects model. Similar results were obtained for psychological distress.

Conclusions The results suggest that the association between individual-level social capital and health is overstated by an individual's unobserved time-invariant attributes. The relevance of health in individual-level social capital should be assessed cautiously.