|
Abstract:
|
Adaptive survey designs are emerging more frequently in practice as a means of achieving higher quality data per unit cost. These designs tend to apply differential data collection protocols to the sample at various stages of collection. Protocols may be determined a priori or identified during data collection (as is the case for a responsive design) and are allocated to the sample based on paradata and other auxiliary information. The manner in how protocols are identified and allocated have implications for weighting and variance estimation. Since information at early stages of collection may be used to determine later stage protocols, variance computations for final combined-stages estimators must reflect the variation in the realized information at earlier stages. Thus, traditional variance estimators must be extended so as to fully reflect the variance contributions from each stage. Following a review of concepts, we outline a mathematical framework for weighting and variance estimation that incorporates the aforementioned variance contributions. We also present a simulation study to illustrate this framework and identify practical considerations for survey practitioners.
|
