Bayesian Factor Analysis for Multilevel Survey Data with Structured Nonresponse

• Activity Number: 397
• Type: Topic Contributed
• Date/Time: Thursday, August 12, 2004 : 8:30 AM to 10:20 AM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Alistair J. O'Malley*+ and Alan Zaslavsky
• Companies: Harvard Medical School and Harvard Medical School
• Keywords: Bayesian ; factor analysis ; multilevel modeling ; multivariate ; nonresponse ; survey data

Abstract

Health care quality surveys in the U.S. are administered to individual respondents (hospital patients, health plan members) to evaluate performance of health care units (hospitals, health plans). Due to both planned item nonresponse (caused by screener items and associated skip patterns) and unplanned nonresponse, quality measures, such as item means, are based on different subsets of the survey respondents. For better understanding and more parsimonious reporting of dimensions of quality, we analyze relationships between quality measures at the unit level, by applying techniques such as factor analysis to covariance structure estimated at the unit level in a hierarchical model. At the lower (patient) level we first fit generalized variance-covariance functions that take into account the nonresponse patterns in the survey responses. A between unit covariance matrix is then estimated using a hierarchical model, which evaluates the fitted generalized variance-covariance functions to account for sampling variation. Bayesian methods are used for model-fitting. An important advantage of this approach is that it allows inferences about the number of factors, the factor loadings, and more.