Improving Predictions of Response Propensities for Effective Adaptive Survey Design (ASD)

Survey practitioners keep steadily searching for methods to improve effectiveness of adaptive survey design. The adaptation performance depends heavily on precise survey parameter estimates, such as response propensity. Recently, making precise estimates becomes increasingly difficult. The existing methods most often come in conflict with the rare historic data sets for running an infrequent or new survey. Also, methods most often ignore the timeliness of historic data of an ongoing survey. Therefore, this dissertation focuses on developing and applying Bayesian methods in adaptive survey design, both for precise and reliable predictions to make about survey design parameters and for ensuring timeliness of scarce survey resources to allocate. I discuss the Bayesian framework for its ability to include external data through prior distributions and to learn how responses vary in time in order to improve prediction precision. I also discuss effective adaptive survey designs that timely tailor the follow-up strategy to approach nonrespondents in order to enhance the obtained response. The proposed methods in this dissertation are applied to some case studies.