Analyzing whether physicians use cesarean sections (c-sections) as defensive medicine (DM) has proven difficult. Using natural experiments arising out of Oregon court decisions overturning a state legislative cap on non-economic damages in tort cases, we analyze the impact of patient conditions on estimates of DM. Consistent with theory, we find heterogeneous impacts of tort laws across patient conditions. When medical exigencies dictate a c-section, tort laws have no impact on physician decisions. When physicians have latitude in their decision making, we find evidence of DM. When we estimate a model combining all women and not accounting for patient conditions (such as models estimated in previous studies) we obtain a result which is the opposite of DM, which we call offensive medicine (OM). The OM result appears to arise out of a bias in the difference-in-differences estimator associated with changes in the marginal distributions of patient conditions in control and treatment groups. The changes in the marginal distributions appear to arise from the impact of tort law on the market for midwives (substitutes for physicians for low-risk women). Our analysis suggests that not accounting for theoretically expected heterogeneity in physician reactions to changes in tort laws may produce biased estimates of DM.

We will state a general version of Simpson's paradox, which corresponds to the loss of some dependence properties under marginalization. We will then provide conditions under which the paradox is avoided. Finally we will relate these Simpson-type paradoxes to some well-known paradoxes concerning the loss of ageing properties when the level of information changes.