Estimating Mail or Web Survey Eligibility for Undeliverable Addresses: A Latent Class Analysis Approach
Biemer, P., Murphy, J., & Kott, P. (2016). Estimating Mail or Web Survey Eligibility for Undeliverable Addresses: A Latent Class Analysis Approach. In JSM Proceedings (pp. 1166-1172). American Statistical Association.
Mail surveys, as well as many web surveys, rely on mailings to the sample members inviting them to complete a paper or web questionnaire. Sample members are selected from a frame such as the address-based sampling (ABS) frame derived from the U.S. Postal Service’s (USPS) Computerized Delivery Sequence file. A well-known problem with such surveys is determining the eligibility of sample members who mailings are returned as “undeliverable.” The undeliverable codes provided by the USPS are often inconsistent across repeated mailings to the same address, yet they typically are treated as accurate in determining case eligibility. This paper describes how sample member eligibility was estimated using a latent class analysis of four indicators of eligibility. In our application, three indicators were based on the USPS codes from 3 mailings sent within a 12-day period to all sampled households and the fourth was the vacancy indicator on the ABS frame. This approach was applied to data from the Residential Energy Consumption Survey National Pilot – a sample survey of 9,650 households – in the calculation of response rates and survey weights.