Estimating the impacts of climate change: reconciling disconnects between physical climate and statistical models

Climate impacts studies often rely on empirical statistical methods to isolate the effects of changing environmental factors on human outcomes of interest. However, this research may not always account for the ways in which the underlying structure of climate might influence estimates from such models. Here we show how the different characteristic spatial scales of temperature (T) and precipitation (P), as well as the physical relationship between T and P, lead to biased historical and projected impact estimates in standard statistical approaches. Furthermore, we demonstrate how contemporary statistical research designs that exploit local fluctuations may unintentionally mis-characterize the influence of shifts in mean climate on outcomes of interest. Drawing on data from published studies we demonstrate these three issues; we focus primarily on the economic growth literature, though our results also apply to other contexts. We also propose several avenues to correct and bound the magnitude of the identified biases.