The implications of aggregate measures of exposure to violence for the estimated impacts on individual risk preferences

A rapidly expanding literature causally links exposure to violence to changes in a variety of behavioral parameters. The estimated coefficients, however, vary greatly across studies in both magnitude and sign. Using original panel data and disaggregated measures of exposure to plausibly exogenous violence in northern Uganda, we investigate the effect of aggregating exposure to violence at the individual and geographical levels. We demonstrate that exposure to violence affects individual risk preferences in strikingly heterogeneous ways depending on the nature of the individual's exposure. Consequently, estimates based on aggregate measures - whether across types of violence within individuals or across individuals within a location - necessarily depend on the underlying distributions of exposure to violence. Simple sampling differences can thereby generate the sort of variability of estimated effects that has been reported in the literature to date.