The Use of Panel Models in Assessments of Climate Impacts on Agriculture
There is a growing literature on the effects of changing climate conditions on various sectors of the economy. Agriculture is one of the sectors that is most susceptible to changes in climate since most farmland is directly exposed to changing weather conditions. Scientific studies examining how weather affects plant growth and agricultural output have a long history. Agronomists have studied the biophysical mechanism between crop growth and various inputs, including weather. In 2010, the Agricultural Model Intercomparison and Improvement Project (AgMIP) was launched with the aim of creating comparable model outputs of various biophysical models by running them with comparable input assumptions. Statistical studies linking agricultural outcomes to weather have a long history of their own—we discuss one approach: panel studies.
The main strength of a panel model is that it allows for the identification of a causal relationship. As emphasized by many statisticians, correlation does not imply causation. To show how panel models are able to uncover a causal relationship, we discuss how fixed effects overcome the omitted variable problem by relying on exogenous weather shocks. Due to the short-term nature of the weather shocks, panel models include short-term adaptation. These deviations are random and exogenous, which means that unlike average weather, they are uncorrelated with other production inputs (land allocation, capital inputs, etc.) that are chosen before the weather is observed. This means that the omission of these other production inputs will not bias the coefficients on the weather variables.
One of the assumptions underlying panel models is that the responsiveness to fluctuations in the exogenous variables is the same for all entities (location or group). This translates into assuming that weather shocks (deviations from the mean) have the same effects in all places. This assumption can be tested by allowing for heterogeneity in the response function through an interaction term. In fact, some studies have tested whether the response function is the same for countries with favorable versus unfavorable farming conditions.
Farmers do adapt to new climate conditions, and there is an emerging literature using panel models to estimate this adaptation. We describe ways to use nonlinear specifications in panel models to estimate costly adaptation measures that involve growing different varieties of the same crop, for example, short-season versus long-season corn. Then we examine adaption through switching between crops.