Advances in Business Statistics, Methods and Data Collection

29Variance Estimation for Probability and Nonprobability Establishment Surveys: An Overview

Jill A. Dever and Dan Liao

RTI International, Washington, DC, USA

The goal of variance estimation for finite population inference is to capture the variability in the characteristics of interest within the inferential population and the random attributes associated with the sample data. The sample attributes can include the mechanism by which sample members are “captured” for the study and random adjustments applied to create the final analysis weights (e.g. weight adjustments made to limit nonresponse bias), if weights are developed for analyses. Additionally, the analytic data file may contain imputed values for missing data arising from item nonresponse within an otherwise complete information set, or possibly unit nonresponse linked to nonparticipating sample members. Proper specification of these attributes is key to quantifying the precision of the survey estimates used for statistical tests, confidence intervals, and the like.

A sampling design specifies how participants are identified for a study. Probability sampling uses sourced information that (ideally) covers the entire population of interest with prespecified sample inclusion probabilities. For example, Dun & Bradstreet maintains a database of US businesses with many characteristics available for tailored sampling. However, when this approach is infeasible or inefficient for a study (e.g. a high‐quality sampling frame does ...