Abstract:
|
A Bayesian analysis of survey data collection may be profitable when expert knowledge and/or historic survey data from the same or similar surveys are available. This knowledge and data may then be employed to set informative prior distributions to coefficients in regression models for survey design parameters, e.g. contact propensities, eligibility propensities, participation propensities, costs per sample unit and survey variable outcomes. During or after data collection posterior distributions may be derived for the same parameters, but also for overall quality and cost measures. Even when survey design parameters change gradually in time or change from one survey to the other, the posterior distributions during or after data collection may be more informative than without the prior knowledge. In earlier papers, we demonstrated how a Bayesian analysis may be implemented and analyzed in monitoring survey data collection. In the current paper, we discuss the optimization and adaptation of survey design using the posterior distributions for survey design parameters, and quality and cost measures. We do so using two case studies.
|