The adequate representation of crop response functions is crucial for agricultural modeling and analysis. So far, the evaluation of such functions focused on the comparison of different functional forms. In this article, the perspective is expanded by also considering an alternative regression method. This is motivated by the fact that extreme climatic events can result in crop yield observations that cause misleading results if Least Squares regression is applied. We show that such outliers are adequately treated if and only if robust regression or robust diagnostics are applied. The example of simulated Swiss corn yields shows that the application of robust instead of Least Squares regression causes reasonable shifts in coefficient estimates and their level of significance, and results in higher levels of goodness of fit. Furthermore, the costs of misspecification decrease remarkably if optimal input recommendations are based on results of robust regression. We therefore recommend the application of the latter instead of Least Squares regression for agricultural and environmental production function estimation.
展开▼
