508 – Leading the Estimates Towards Known Benchmarks
A Global Convergent Algorithm for Integer Calibration Weighting
Kelly Toppin
United States Department of Agriculture
Luca Sartore
United States Department of Agriculture
Clifford Spiegelman
Texas A&M University
The USDA's National Agricultural Statistics Service (NASS) conducts the U.S. Census of Agriculture in years ending in 2 and 7. The census describes the characteristics of U.S. farms and the people who operate them. To adjust for under-coverage, nonresponse and misclassification, NASS produces the weights on the responding records using a capture-recapture methodology. However, the weights need to be further refined through a calibration process so that the census estimates agree with known population values. The current algorithm (called INCA) was developed to provide integer calibrated weights per NASS requirements. In INCA, weights adjusted for undercoverage, nonresponse, and misclassification are first rounded using an optimal rounding procedure, and then integer programming using coordinate descent is performed on the integer weights. However, the existence of multiple local solutions makes the search of a global solution exponentially complex. This article describes a global optimization algorithm for integer calibration based on an L1-norm relative error. The results of a simulation study designed to investigate the properties of the estimator is presented. SID # 215233