|
Abstract:
|
To reduce survey costs while maintaining desired sample sizes, survey designers blend probability samples with nonprobability samples; e.g., to increase the nominal sample size of a hard-to-reach population, a sample design may include a convenience sample of the hard-to-reach population along with a traditional, random probability sample. However, nonprobability samples are likely biased & therefore not representative. Using a nonprobability sample for inference therefore poses statistical challenges which must be addressed. How can the nonprobability sample be corrected to represent part or all the population of interest? What are the most appropriate analytic & variance estimation methods for blended samples? While several statisticians have offered suggestions for how to address these questions – and other related questions – no single methodology has been adopted by the survey community. Participants will discuss the different aspects for blending probability & nonprobability samples & different solutions to achieve a combined sample which is generalizable. Statisticians & methodologists with experience in blending surveys as well as researchers new to this area are welcome.
|
