Abstract:
|
Efficient estimation of crop parameters at the county (small domain) level is an important priority for the USDA's National Agricultural Statistics Service (NASS). This paper focuses on three mixed modeling approaches to county-level estimation of crop planted or harvested area where survey reported values are fit to unit (farm) and area (county) level covariates: 1) an empirical best linear unbiased predictor (EBLUP) model, 2) an adaptive empirical best prediction (EBP) model, and 3) a log-transformed EBP model. In a simulation study involving corn and soybean planted area in Ohio and South Dakota for 2018, the three estimators are compared using data from NASS’s County Agricultural Production Survey (CAPS) and auxiliary data sources. Control data from NASS’s list sampling frame are used as the unit level covariate while two options are considered for the area level covariate: 1) Farm Service Agency (FSA) planted acreage, and 2) satellite-based pixel counts obtained from NASS’s Cropland Data Layer. Since the unit level covariate is missing for a subset of the list frame records, regression synthetic estimation is applied in that portion of the population to ensure complete coverage.
|