Go backSearch Session number: 1157

Workshop title: Comparison of Software Packages and Analyses of VA Data with a Hierarchical Structure

Author(s):
CL Christiansen CHQOER and Boston University SPH
F Wang CHQOER and Boston University SPH
BH Chang CHQOER and Boston University SPH

Objective(s): The future of health care depends on quality research; quality research requires statistical models and methods that fit the data and analyses with translatable results. Most HSR questions are answered using hierarchically structured data. This workshop illustrates and compares analyses of hierarchically structured data and interprets results for research questions using commercial software packages. The objectives of the workshop are that the participants will 1) learn some of the statistical terminology for hierarchical modeling. We will explain concepts such as: clustered data, multi-level covariates, covariate effects, variation measure, random effects, Bayesian models, etc. 2) review examples of data and research questions in health services research that require the use of hierarchical statistical models. Typical goals of analyzing this kind of data include estimating the effect of patient-level, provider-level and facility-level characteristics and profiling facilities based upon utilization, costs and outcomes. 3) be introduced to different estimation methods used for hierarchical modeling, e.g., likelihood-based approach, generalized estimating equations (GEE), and Markov Chain Monte Carlo simulations (MCMC). 4) learn how to interpret results from different software packages. We will discuss and interpret output from two studies analyzed using different software. Output from SAS PROC mixed, PROC GENMOD, Macro GLIMMIX, MLwiN, WinBUGS, and Splus lme are compared. Participants will receive a short guidebook on using the hierarchical software discussed. Several easy-to-read articles on Bayesian models will be distributed and web sites that offer information on software will be provided.

Activities: We will use examples from health service research to demonstrate hierarchical modeling. Results from different software packages will be presented, interpreted, and compared.

Target audience: The workshop is designed for health service researchers who use statistical models in their work, for decision-makers who rely on output from statistical analyses, and for applied statisticians who are unfamiliar with this area of statistics. In particular, it will help researchers who are interested in hierarchical modeling to gain better understandings in this area.

Audience familiarity: Participants should be familiar with the term 'variation' and have an understanding of standard statistical methods such as t-tests and chi-square tests.