Bayesian Penalized Spline Models for Statistical Process Monitoring of Survey Paradata Quality Indicators
Joseph L. Schafer · 2013
This chapter serves as an introduction to the use of penalized splines for describing and monitoring a paradata series. It presents flexible semiparametric models for paradata series that allow the process mean to vary. The models presented are designed for a single variable recorded over time at regular or irregular intervals. The chapter describes efficient Markov chain Monte Carlo strategies for simulating random draws of model parameters from the high-dimensional posterior distribution and produces graphical summaries for process monitoring. It illustrates these methods on monthly paradata series from the National Crime Victimization Survey (NCVS). The penalized splines in the chapter are intended to serve as a first generation of descriptive models for NCVS data-quality indicators. The authors experimented with Gibbs samplers for the NCVS screener data series and found that they converged very slowly, prompting the search for more efficient alternatives.