Comparison of Methods for Updating Census Based Estimates of Number of Farms to Non-Census Years

The National Agricultural Statistics Service (NASS) conducts the census of agriculture (a complete count of US farms and ranches) every five years. Sample surveys, including the June area survey (JAS), are carried out annually to obtain estimates of many of the same agricultural quantities as the census. Due to the large number of operators surveyed and the complete coverage provided by the census, its numbers are considered more accurate than those derived from the much smaller scale sample surveys. An interesting question is whether census figures for specific survey items can be used in conjunction with survey data to improve estimation accuracy for non-census years. Because of its relative stability over time, the survey item considered most likely to benefit from such an approach is number of farms in a state. Two proposed methods for projecting census counts of number of farms to subsequent non-census years are evaluated. The first method updates the census figure to the current year using JAS data only, while the second makes additional use of official NASS state level estimates of number of farms for the previous year (if it wasn’t a census year). The two methods are identical for the first post-census year. The proposed estimators are compared with area frame based and hybrid operational estimators for the years 2003-06 in a study covering most of the lower 48 states, both at the state level and within categories defined by farm value of sales. Variances are estimated using an extended delete-a-group jackknife method.