Online Program

Complex Weighting and Disposition Coding in a Survey of Establishments

Keywords: Phone, weighting, CHAID, disposition codes

Analytic weights were created in steps to account for unequal selection probabilities and varying response rates among population subgroups. Sample records were classified according to eligibility status (known or unknown) based on two disposition codes (phone and postal); then the sampling weights were calculated as the inverse of selection probability. Non-response adjustments were made in two stages and derived based on Logistic Regression Models using the CHAID procedure: first, sampling weights were adjusted to account for members with unknown eligibility; then the weights were adjusted to account for eligible sample members who did not return complete questionnaires. The weights were then post-stratified to match population totals and reduce bias unaccounted for by the previous steps. Finally, sampling strata were collapsed to create strata for variance estimation using Taylor series linearization. In addition to weighting, this article details the process of merging phone and postal dispositions to derive the final disposition codes meeting AAPOR standards, as well as procedures to roll up nested domain estimates to the national level.