|
Abstract:
|
Current solutions for unit nonresponse focus on improving nonresponse follow-up and enhancing post-survey weighting adjustment. We propose an alternative inferential paradigm to adjust for unit nonresponse using micro-level auxiliary data that captures the same features of the target population, referred to as 'benchmark'. We describe a benchmark-driven mitigation and imputation (M&I) strategy, in the context of a multi-phase survey, that sequentially guides the sampling and estimation to improve survey inferences regardless of the nonresponse mechanism. The M&I strategy employs a high quality benchmark to mitigate undesirable nonresponse patterns through a benchmarked sequential sampling (BSS) design; and to impute population information through benchmarked multivariate imputation (B-MI) by chained equations. The performance of the M&I strategy will be evaluated by simulation experiments to mimic adaptive design under various nonresponse mechanisms including missing not at random (MNAR). We report on the preservation of marginal and joint distribution for population estimates of the M&I strategy from respondent data and completed data.
|
