477 – Calibration Estimation and Weighting in Sample Surveys
Practical Issues When Calibrating Weights for Multiple Skewed Variables
Kimberly Henry
IRS/SOI
Yan K. Liu
IRS/SOI
Michael Strudler
IRS/SOI
The Statistics of Income Division of the IRS started a panel sample of individual returns in 2007 for longitudinal analyses. This panel sample has also been used for crosssectional estimations of multiple variables that are skewed and weakly correlated. Therefore, cross-sectional weights are needed to refer to the out-year population and multiple variables of interest. Calibration method is therefore applied to adjust weights such that sample estimates are close to population benchmarks. In this paper, we look at issues such as calibration cells, initial weights, prediction models and weight bounding. We share some experience in calibrating weights using R. We aim to produce final weights that are reasonable for multiple variables of interest.