Online Program

Evaluation of Methods for County-Level Estimation of Crop Harvested Area that Employ Mixed Models

Keywords: small area estimation, empirical best prediction, regression synthetic estimation

The County Agricultural Production Survey (CAPS) is an annual survey conducted by the USDA’s National Agricultural Statistics Service (NASS) with the objective of providing accurate county level area and production estimates for approved federal and state crop commodities. Of special interest is the development of improved methodology for small area estimation of crop harvested area using CAPS data in conjunction with auxiliary information. We consider two empirical best prediction methods that employ mixed models to relate sample survey data to one unit (farm) level variable and one area (county) level variable: 1) an adaptive empirical best prediction (EBP) approach inspired by the Battese-Harter-Fuller (BHF) model, and 2) an enhanced EBP approach where modeling is applied on the logarithmic scale (i.e., involving logs of model variables). Regression synthetic estimation is used to compensate for undercoverage due to the unit level covariate being missing for a subset of farms in the population, meaning that the final estimates are of the hybrid (combined) variety. The proposed estimators were compared with two other estimators using CAPS survey data over a two-year period (2009-10) for corn and soybeans in three states.