Improved Small Domain Estimation via Compromise Regression Weights

Shrinkage estimates of small domain parameters typically use a combination noisy “direct” estimate that only uses data from specific and more stable regression estimate. When the model is misspecified, estimation performance for noisier domains can suffer due to substantial shrinkage toward poorly estimated surface. In this article, we introduce new class robust, empirically-driven weights target means under potential misspecification global model. Our are convex model-based associated with best linear unbiased predictor (BLUP) those observed (OBP). The mixing parameter in found by minimizing novel, mean-squared prediction error means, label “compromise predictor,” or CBP. Using data-adaptive mixture enables CBP preserve robustness OBP while retaining main advantages EBLUP whenever correct. We demonstrate an application estimating gait speed older adults. Supplementary materials article available online.