A1105
Title: A novel continuum-of-resistance model and doubly robust for nonresponse adjustment with callback data
Authors: Kendrick Li - University of Michigan (United States) [presenting]
Xu Shi - University of Michigan (United States)
Wang Miao - Peking University (China)
Abstract: Callback data design is a powerful tool to address missingness not at random in survey sampling. Although the notion of a continuum of resistance on the callback mechanism is widely used in social surveys for leveraging callback data to make nonresponse adjustments, the corresponding statistical analysis literature is relatively sparse. A novel model is proposed that clarifies the underlying assumptions for leveraging the continuum of resistance to make nonresponse adjustments. The proposed model assumes only that the odds ratio functions of the response propensity in the last two captures are equal, conditioning on the other variables. Under this model, the identification was established, and a suite of estimators was developed for the estimation of a general function of the full data law, including a doubly robust one that is consistent if either the callback mechanism or the outcome distribution is specified correctly. The doubly robust estimator with a double machine learning estimator was further extended, which estimates the callback mechanism and the outcome data distribution with flexible machine learning methods. The performance of the proposed estimators with comprehensive simulation studies and an application to ConsumerExpenditure Survey data are demonstrated.
