Analysing farmland rental rates using Bayesian geoadditive quantile regression

Empirical studies on farmland rental rates have predominantly concentrated on modelling conditional means using spatial autoregressive models, where a linear functional form between the response and the covariates is usually assumed. However, if it is in fact non-linear, misspecifying the functional form can adversely affect inference. While mean regression models only allow limited insights into the way covariates influence the response, extending the analysis to the modelling of conditional quantiles can give a more detailed picture of the conditional distribution. Based on data from the German agricultural census, this article contributes to the agricultural literature by modelling conditional quantiles of farmland rental rates semi-parametrically using Bayesian geoadditive quantile regression models. The flexibility of this model class overcomes the problems associated with functional form misspecifications and allows us to present a more detailed analysis. Our results stress the importance of making use of semi-parametric regression models as several covariates influence farmland rental rates in an explicit non-linear way.