Economists typically assume that risk compensation has uniformly negative or self-protective
effects – that people become more careful as the perceived health risks of their actions increase.
However, under certain conditions, rational responses can instead be positive, or fatalistic:
increased perceived risks can lead people to take fewer precautions. By extending the basic
model of rational responses to health risks, I show that fatalism will result whenever perceived
health risks are sufficiently high and people do not have perfect control over all possible
exposures. This result holds even for agents who do not understand probability in a sophisticated
way. I then test for the possibility that sufficiently high health risks induce fatalistic responses
using a randomized experiment that provided information on HIV transmission risks to people in
rural Malawi. On average, responses to the information are self-protective and fairly small in
magnitude, but statistically significant: the risk elasticity of sexual behavior is roughly -0.5.
Consistent with my theoretical predictions, these responses are strongly heterogeneous by
baseline risk beliefs. I develop a novel estimation strategy that extends the logic of
heterogeneous treatment effects analysis to instrumental-variables estimates of the marginal
effect of risk beliefs on sexual risk-taking. Using this approach, I find that over 10% of the
population has risk beliefs high enough that they have become fatalistic, so their risk elasticity of
sexual behavior is positive rather than negative. These individuals reduced their sexual risktaking
in response to the information treatment, consistent with the notion that HIV risk
messaging is backfiring for this subpopulation. They also have higher baseline indicators of
sexual activity, which suggests that the consequences of providing accurate risk information for
the overall prevalence of HIV are epidemiologically ambiguous.