502 – Innovative Statistical Methods for Complex Survey Data
A Bayesian Hierarchical Model for Combining Several Crop Yield Indications
Nathan B. Cruze
USDA National Agricultural Statistics Service
USDA's National Agricultural Statistics Service (NASS) conducts multiple surveys over the course of a growing season. Each of these surveys reflects current growing conditions and provides a prediction of end-of-season crop yield. In particular, NASS conducts two interview-based surveys and one field measurement survey from which indications of crop yield may be obtained. It is also known that a number of weather conditions during the growing season may contribute to changes in crop yield. This talk describes a Bayesian hierarchical model that improves end-of-season crop yield predictions by combining these several disparate sources of information. The model incorporates benchmarking of state-level forecasts with regional forecasts of crop yield and gives rise to rigorous measures of uncertainty. It also permits a useful decomposition with respect to the emphasis placed on each information source. Several aspects of covariates selection and model performance are discussed.