Electoral Consequences of Rainfall on the Election Day: Compositional Data Analysis

When multiple elections are held concurrently, voter turnout is often boosted. Although this spillover effect of concurrent elections has been studied extensively, there remain two important identification problems: a failure to disentangle competing mechanisms underlying the effect (rational choice, psychological stimulus, and mobilization mechanisms) and a failure to estimate the size of the effect without omitted variable bias. To address these problems, we leverage a natural experimental setup in Japan. We show that voter turnout in a high-stimulus prefectural election is raised by one to two percentage points when a low-stimulus municipal election is scheduled in two weeks. We argue that our design enables us to attribute the estimated effect only to mobilization mechanism. Furthermore, blocking by districts for prefectural elections and the as-if random timing of municipal elections mitigate omitted variable bias. This study exemplifies a design-based solution to identify causal effects, while addressing the problem of bundling treatments.