Performance Metrics for Sample Selection Bias Correction
64th ISI World Statistics Congress - Ottawa, Canada
Format: CPS Abstract
Keywords: non-probability sample, performance metrics, selection bias
Session: CPS 01 - Statistical methodology II
Monday 17 July 8:30 a.m. - 9:40 a.m. (Canada/Eastern)
When estimating a population parameter by a non-probability sample, a sample without a known sampling mechanism, the estimate may suffer from sample selection bias. To correct selection bias, one of the often-used methods is assigning a set of unit pseudo-weights to the non-probability sample, and estimating the target parameter by the weighted sum. However, a tailor-made framework to evaluate the assigned weights is missing in the literature, and the evaluation framework for prediction problems may not be suitable for population parameter estimation. We try to fill in the gap by discussing several promising performance metrics, which are inspired by classical calibration and measures of selection bias. A simulation study and real data examples show that some performance metrics have a strong positive relationship with the mean squared error of the estimated population mean. These performance metrics may be helpful for model selection when correcting selection bias by logistic regression or machine learning algorithms.