“Storm autocracies”: Islands as natural experiments

We exploit the exogenous variation in the timing and intensity of storms in island countries to estimate the storms' effect on the extent of democracy. Using a rich panel dataset spanning the period 1950–2020, our difference-in-differences estimations, which allow multiple treatments over time, indicate that storms trigger autocratic tendencies in island countries by reducing the Polity2 score by about four percent in the following year. These findings resonate with our simple dynamic game-theoretical model, which predicts that governments move towards autocracy by placating citizens with post-disaster assistance in response to citizens’ insurgency threat in the absence of relief, giving rise to the political regime of “storm autocracies”. Our results survive a battery of robustness analyses, randomization tests, potential spatial biases, and other falsification and placebo checks.