Modeling Heteroskedasticity of Crop Yield Distributions: Implications for Normality

The paper analyzes the extent to which ignorance of heteroskedasticity or its inadequate modeling would result in misleading statistical inferences about crop yield distribution. We follow the "detrending mean yield approach" in which we model the conditional mean yield using a panel data model. We assume alternative structures of variance-covariance matrix for the random component. Heteroskedasticity robust and non-robust estimation methods are used before performing a joint normality test on the random component of crop yield data. Our findings provide evidence against the claim that virtually all previous findings of non-normality in crop yields are infected because of the ignorance of heteroskedasticity or its inappropriate modeling. Accounting for heteroskedasticity in crop yield data would matter for validity of evidence against normality only to the extent that its proportion among the source of departure from normal distribution is relatively sizable.