Speaker: Jared Murray, Duke University Department of Statistical Science
Title: Hierarchically Coupled Mixture Models with Local Dependence for Imputing Mixed Data
Abstract: We introduce a hierarchically coupled mixture model with local dependence (HCMM-LD) for modeling mixed continuous, ordered and unordered categorical variables ("mixed data"). Our proposal improves upon existing methods for mixed data by incorporating local dependence through a carefully structured hierarchical model. We illustrate its use in multiple imputation for heterogenous survey data. The MI setting has proven challenging for existing joint models, leading many researchers to recommend potentially improper sequential imputation techniques over proper joint modeling. We demonstrate that as a default imputation engine the HCMM-LD outperforms the most popular sequential imputation alternative on data from the Survey of Income and Program Participation.
º Duke University: contact Alan Karr
º Cornell University, Ithaca campus: Ives 381
º Census Bureau headquarters: Room 5K410, contact Dan Weinberg
- Live video conference. Please contact Lars Vilhuber (email@example.com) if you wish to participate by video conference, by Monday, Dec 2, 2013.