A1330
Title: A unified two-step estimation approach for semiparametric models under two-phase sampling
Authors: Hoi Min Ng - The Hang Seng University of Hong Kong (Hong Kong) [presenting]
Qingning Zhou - University of North Carolina at Charlotte (United States)
Kin Yau Wong - Hong Kong Polytechnic University (Hong Kong)
Abstract: The two-phase sampling design provides a cost-effective strategy for studying associations between disease outcomes and relevant risk factors. In the first phase, inexpensive variables are collected on a large cohort, and in the second phase, a subsample is selected for more detailed and costly measurements. This design retains a broad population view while requiring only a representative subset for expensive covariate acquisition. Existing update estimation methods construct a consistent estimator using complete cases and then refine it using information from a working model built on auxiliary variables observed in the full cohort. Although this approach can substantially improve efficiency, most existing implementations focus on updating Euclidean parameters. A more general setting is considered in which the working model involves functional parameters. An update estimation framework is developed that accommodates both Euclidean and functional parameters, thereby improving efficiency across a broader class of targets. The proposed method supports estimation and inference for Euclidean parameters, functional parameters, and their combinations. This is an important advantage because many quantities of interest in biomedical studies are inherently functional. The proposed methodology is illustrated using semiparametric models, including the transformation model and partially linear model.
