Incorporating Sampling Designs Into a Grade of Membership (GoM) Model

• Activity Number: 186
• Type: Contributed
• Date/Time: Monday, August 2, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Section on Survey Research Methods
• Author(s): Marnie Bertolet*+
• Companies: University of Pittsburgh
• Keywords: Sampling Weights ; Grade of Membership Model ; Bayesian Mixed-Membership

Abstract

The Grade of Membership (GoM) model is a heirarchical Bayesian mixed-membership model used recently to analyze latent class disability profiles in the National Long Term Care Survey (NLTCS). Survey data often includes complexities such as informative sampling and dependencies induced by clustering. Two modifications of the GoM model are introduced to incorporate these complexities. The first changes the Dirichlet prior to a polytomous logistic mixed-effects prior to model the stratification and clustering in the sampling design. The second adds probability weights using a method called Weighting based on the Estimated Parameter, or wEP. To demonstrate the effectiveness of these modifications, we use simulated data to compare the Dirichlet prior to the mixed-effects and then apply the method to the NLTCS data.