Online Program Home
My Program

Abstract Details

Activity Number: 259 - SPEED: Missing Data and Causal Inference Methods, Part 2
Type: Contributed
Date/Time: Monday, July 29, 2019 : 3:05 PM to 3:50 PM
Sponsor: Health Policy Statistics Section
Abstract #307654
Title: Bayesian Inference of Separable Covariance Models for Health Care Quality Measures
Author(s): Judith Law* and Laura A Hatfield and Alan M. Zaslavsky
Companies: Harvard Medical School and Harvard Medical School and Harvard Medical School
Keywords: hierarchical models; Bayesian methods; longitudinal data analysis; multivariate data; observational studies
Abstract:

Analysis of periodically administered health care surveys may reveal key patterns of variation in quality measures over time. Observations of multiple measures in multiple years yield an array-structured random variable. Its variance-covariance matrix may be constrained to be a Kronecker product of an across-measure matrix and across-year matrix. Imposing constraints such as these can aid interpretation of the patterns in correlations across dimensions (such as measure and year).

In this work, we take a Bayesian approach to estimation and inference of the covariance matrix implied by a multi-level model when the covariance is Kronecker-structured or a sum of Kronecker-structured matrices. We use a separation strategy, decomposing the covariance matrix into standard deviations and correlations. We consider shrinking the covariance matrix toward a Kronecker structure using informative priors as well as strictly enforcing the structure as a constraint. We use Markov chain Monte Carlo to make draws from the posterior distribution and explore graphical methods for displaying this posterior uncertainty about variance-covariance matrices. We illustrate with data from the CAHPS survey.


Authors who are presenting talks have a * after their name.

Back to the full JSM 2019 program