203 – Section on Statistical Education P.M. Roundtable Discussion (Fee Event)
Improved Sampling Weight Calibration by Generalized Raking with Optimal Unbiased Modification
Avi Singh
NORC at the University of Chicago
Nadarajasundaram Ganesh
NORC at the University of Chicago
Yongheng Lin
NORC at the University of Chicago
In calibration methods for sampling weight adjustment, there is no built-in mechanism for ensuring variance reduction although typically it does lead to variance reduction. We introduce new stratum-specific scale parameters in the calibration or generalized raking model to capture possibly varying nonresponse bias, coverage bias, and design characteristics by strata or super-strata. Approximate unbiasedness of calibration estimators is still maintained in the presence of these extra parameters which are estimated outside the calibration equations by minimizing the generalized variance of key study variables or alternatively the unequal weighting effect for simplicity. Besides, instead of trimming potential extreme weights produced in calibration, we propose modeling to smooth extreme weights to avoid introducing bias. Using a simulation study, various calibration methods are compared in terms of bias and variance.