Bias in economic evaluation of variable rate application based on geographically weighted regression models with misspecified functional form

Abstract

Geographically weighted regression (GWR) has been presented as a valuable tool for estimating site-specific yield response functions to derive recommendations of variable rate input. This study employs Monte Carlo simulations to illustrate that if GWR assumes a quadratic yield response functional form while the actual yield-input relationship is quadratic-plateau, it can significantly overestimate the economic value of variable rate application compared to its true value. Practitioners in precision agriculture should exercise caution when utilizing GWR for site-specific input recommendations. Statistical community is also encouraged to develop tools in software packages providing GWR that allow more flexibility in functional form assumptions.