356 – Contributed Poster Presentations: Survey Research Methods Section
Sample Design and Weighting for Estimating a Dose-Response Curve
Sharon Lohr
Westat
Pamela Broene
Westat
Eric Jodts
Westat
The Neighborhood Environment Survey was designed for the purpose of estimating the relationship between a household's aircraft noise exposure and the probability that the household reports being highly annoyed by aircraft noise. Unlike most surveys, in which at least some of the primary outcomes are means or totals, here the primary outcome is a logistic regression function that estimates the dose-response relationship. The first stage of the design uses balanced sampling to select airports. The household selection at the second stage of sampling takes a stratified sample of households from an address-based sampling frame. At most airports, there are relatively few households that are close to the airport and have high noise exposure levels, but there are many households that are farther away from the airport and have low noise exposure levels. We use optimal design results of Chaloner and Larntz (1989) to inform the second-stage sample allocation to noise exposure strata at each airport. The resulting sampling weights for households have high variation, and we discuss the choice of weights for estimating the dose-response curve and for nonresponse adjustments.