Soft calibration for selection bias problems under mixed-effects models

Abstract Calibration weighting has been widely used to correct selection biases in nonprobability sampling, missing data and causal inference. The main idea is calibrate the biased sample benchmark by adjusting subject weights. However, hard calibration can produce enormous weights when an exact enforced on a large set of extraneous covariates. This article proposes soft scheme, where outcome indicator follow mixed-effect models. scheme imposes fixed effects approximate random effects. On one hand, our intrinsic connection with best linear unbiased prediction, which results more efficient estimation compared calibration. other be envisioned as penalized propensity score weight estimation, penalty term motivated structure. asymptotic distribution valid variance estimator are derived for We demonstrate superiority proposed over competitors simulation studies using real-world application effect BMI screening childhood obesity.