|
Abstract:
|
Multilevel Regression with Poststratification (MRP) is a powerful tool to handle differential nonresponse in survey samples. However, typically for a variable to be included in the regression it’s population distribution across the cells must be known so that in the poststratification step an appropriate cell weight could be calculated. While the US Census and other sources provide population level information on a variety of variables some variables that inform opinions such as political party do not have the necessary information to be used in the standard MRP framework. In this presentation we present an iterative approach based on Kastellec et al (2014) but extending to include multiple variables with unknown population distribtions. First, multilevel regressions for non-Census variables are fit. Then for each MCMC iteration a sample of the non-Census variable is drawn from the posterior for each poststratification cell. Finally, the samples of non-Census variable are included to estimate the main parameter of interest using standard MRP. This method is applied to a survey on public support of gene drive at the end of the survey given support at the beginning of a survey.
|
