Each year the United States Department of Agriculture’s (USDA’s) National Agriculture Statistics Service (NASS) conducts more than 100 surveys and produces more than 400 reports on all aspects of U.S. Agriculture, ranging from estimates of corn and cattle production to agricultural prices and expenses. For over a half century, sample surveys have been the foundation for producing these official statistics. In recent years, as is common with other National Statistical Agencies, response rates and the coverage of the NASS list frame (of all known U.S. farms) have continued to decline. Further, because U.S. agricultural production is increasingly concentrated in a few large farming operations, the quality of the official statistics produced relies heavily on obtaining responses from the producers of these large farms. Consequently, these producers may each receive more than 20 surveys each year, constituting a heavy reporting burden. At the same time, non-survey data, including administrative, remotely sensed, and weather data, have become increasing available. NASS is currently working to incorporate more non-survey data in the estimation process to improve the precision of the estimates while reducing reporting burden. Two approaches to integrating survey and non-survey data are (1) to combine survey estimates and non-survey information at some specified level of geography through modelling and (2) to link the survey and non-survey data at the farm level and then to produce estimates based on the linked data. To date, NASS has taken approach (1), using Bayesian models to combine survey and non-survey data at the county or regional levels in three major programs. These models incorporate the survey estimates, non-survey information, and known constraints on the estimates. This presentation focuses on the Bayesian models now being used to provide official statistics and describes the process of moving a new model into the production. Linking the survey and non-survey data at the farm level, i.e., approach (2), has been challenging because, while all non-survey data acquired thus far are georeferenced, the NASS list frame itself is not. Progress is being made and will be outlined. Current and future research directions will be highlighted.
22 February 2023
Location: Watch this webinar