Social factors and health: the causation-selection issue revisited.

Since social scientists rarely have access to experimental data, they rely heavily upon observational studies. As a consequence, their attempts to make causal inferences about the effects of social factors--such as occupation or marital status--on health are plagued by potential selection problems. Some researchers have addressed this selection-causation problem on the basis of the presence or absence of a particular aggregate pattern of health status. The rationale underlying this approach derives from the investigators' hypotheses that the presence of selection would lead to a particular type of pattern that is distinct from the pattern that would result in the absence of selection. Although intuitively appealing, this strategy appears to be seriously flawed. The essential weakness is that the range of patterns that can result from selection is often much broader than researchers have speculated.