Interest in adaptive survey designs (ASDs)has been driven by increasing data collection costs and deteriorating survey data collection conditions. ASDs allow tailoring of data collection features to particular cases or groups of cases in an effort to reach data collection goals in an efficient way.
In order to make effective interventions during data collection, it is necessary to understand the implications of potential interventions on response propensity, cost, and a quality metric of interest, in this case root mean squared error (RMSE) of survey estimates. This understanding is then translated into predictive models which are estimated throughout the data collection process in order to drive interventions.
Using the National Survey of College Graduates, this paper introduces a Bayesian method for estimating these predictive models and compares this method to standard methods currently used in adaptive designs. By targeting the RMSE, this work expands upon prior experiments in the NSCG by using adaptive survey design to improve a direct measure of the quality of survey estimates, rather than a proxy (such as the R-indicator or sample balance) of quality.