Responsive and adaptive survey designs can be used to reduce the risk of nonresponse bias through data collection. In responsive design, different protocols that appeal to prior nonrespondents can be introduced in phases. In adaptive survey design, particular nonrespondents can be targeted in these subsequent phases, based on predefined criteria. In the case of nonresponse bias, the criteria can be propensity models. Key, however, is the specification of these models.

We introduce the concept of bias propensity models—response propensity models that include covariates of key survey variables, but deliberately exclude strong predictors of only nonresponse. This approach is demonstrated in a longitudinal survey, while addressing two common limitations in the research literature on responsive and adaptive survey designs: such labor-intensive designs are often implemented in well-funded studies with rigorous designs that can have reduced risk of nonresponse bias, and large-scale studies do not always have the opportunity to include an experimental design. Both of these limitations call for analytic solutions.