|
|
|
|
|
Kathleen Welch, MPH, MS, Center for Statistical Consultation and Research, The University of Michigan, 3550 Rackham Building, 915 E. Washington St., Ann Arbor, MI 48109-1070, 734 647 4611, kwelch@umich.edu, Brady West, MS, Center for Statistical Consultation and Research, the University of Michigan, 3550 Rackham Bldg., 915 E. Washington St., Ann Arbor, MI 48109-1070, and Brenda Gillespie, PhD, Center for Statistical Consultation and Research, University of Michigan, 3550 Rackham Building, Ann Arbor, MI 48109-1070.
Public Health researchers often work with data having longitudinal or clustered measurements. For example, the effect of a drug may be measured over time for each patient, or patients may be clustered within a clinic or hospital. Linear mixed models are used to account for the correlated errors inherent in such data. In the past decade, advances in statistical software have resulted in several procedures capable of fitting linear mixed models with normally distributed errors. Currently, such procedures are available in SAS (Proc Mixed), SPSS (Mixed Models), S+/R, STATA and HLM (Hierarchical Linear Models), with each program offering a unique set of available models, options and defaults. The data analyst is faced with making a choice between these statistical software packages. We compare these packages in terms of their ability to fit a variety of models, their default settings, and their options. Some options of interest include types of tests for fixed effects (F-tests, Wald tests, likelihood ratio tests), estimates of degrees of freedom for F-tests (e.g. traditional analysis of variance degrees of freedom vs. adjusted degrees of freedom such as Satterthwaite, Kenward-Roger, or Huynh-Feldt), standard error estimates (model-based vs. robust or sandwich-type), and correlation structures available for repeated measures (e.g. compound symmetry, autoregressive, unstructured, toeplitz). Examples will be presented illustrating sample output for each package, with comparisons of output using the same data. We will make recommendations on appropriate software for various settings.
Learning Objectives: After reading this poster, the participant will be able to
Keywords: Biostatistics, Statistics
Presenting author's disclosure statement:
Organization/institution whose products or services will be discussed: SAS, SPSS, Stata, Splus, R, HLM statistical software packages.
I have a significant financial interest/arrangement or affiliation with any organization/institution whose products or services are being discussed in this session.
Relationship: SAS. I am an author of a book published by SAS Institute, "Selecting Statistical Techniques for Social Science Data: A Guide for SAS Users" by Frank M. Andrews, Laura Klem, Patrick M.. O'Malley, Willard L. Rodgers, Kathleen B. Welch, Terrence N. Davidson,