Hofer TP (VA Ann Arbor Center for Clinical Management Research), Chistiansen CL
(Center for Health Quality, Outcomes, & Economic Research (CHQOER) and Boston University)
Increasingly, health services research incorporates multilevel designs and data analysis. Multi-level models are well suited to investigations of how the organization of healthcare affects its delivery given the hierarchical structure of health care organizations with patients cared for by providers within clinics and hospitals and temporally with observations repeated over time. Furthermore, multi-level models also allow for the construction of measurement models when variables are measured with error, as is so often the case with clinical variables. This workshop will build on an introductory workshop at the 2007 HSR&D meeting where an overview of multi-level models was presented. Two somewhat more advanced topics will be presented in this workshop. First, we will address issues in multi-level modeling with dichotomous and ordinal outcomes. Objectives will include understanding the advantages and disadvantages of different possible estimation methods, ways of quantifying variance and explained variance at different levels, as well as the different types of predicted probabilities that can be obtained, including marginal and cluster specific probabilities. A second focus will be on multi-level longitudinal models where participants will understand how to choose different covariance structures, model change, and build growth models.
We will use examples from health services research for all of the objectives. Practical focus will be on STATA with some cross-references to SAS, although other packages will also be referenced. Participants are encouraged to talk and ask questions. One-third of the time in the session is reserved for questions about multi-level modeling raised by the participants, and where answers and discussion will hopefully come from members of the audience with experience in multi-level modeling as well as the panelists.
The workshop is designed for health services researchers who use statistical models in their work, as well as decision-makers who rely on output from statistical analyses, and for applied statisticians who may be unfamiliar with some of these topics or who have advice and specific experience or examples to contribute to this session.
Assumed Audience Familiarity with Topic:
Participants should be familiar with the basic concepts of multi-level modeling either by attending the session last year, some equivalent introduction to multi-level modeling, or reading in the early chapters of Multilevel Analysis: An Introduction to Basic and Advanced Multi-level Modeling" by Tom Snijders and Roel Bosker.