Operationalisation matters: Weather extremes as noisy natural experiment show no influence on political attitudes

Abstract

Weather extremes are of (in)direct interest for a large literature. However, measuring "weather" is theoretically ambiguous, with a wide variety of potential operationalisations. Also, the theoretical link between exposure and preferences/behaviour is structured by intermediate steps with multiple potential dependent variables, usually reported on in separate projects. In combination, this leads to vast researcher degrees of freedom increasing the chance of false positive results. We develop a template to caution against false positives, and apply it to the much-researched question whether personal experiences of weather extremes "green" citizens’ preferences. From a large literature with inconclusive results, we derive 34 plausible operationalisations of extreme weather, and 4 core steps investigated. Drawing on high-quality geo-coded survey and weather data for Switzerland, we show that singular specifications support arbitrary conclusions, while the whole set indicates robust null results. This indicates a note of caution for all research with latent variables and spatial clustering.