|
Longitudinal Data
Analysis: Featuring the QIF Approach
Peter Song (University of Waterloo) and Annie Qu (Oregon State University) |
Abstract:
Our presentation at the workshop is intended to introduce the quadratic
inference function (QIF), a new powerful statistical method for
longitudinal data analysis, via real-world data examples and the use of
the newly developed SAS MACRO QIF by Song and Jiang (2006).
The QIF, first proposed by Qu et al. (2000), is getting increasingly
popular in the analysis of repeated measurements. The QIF approach can
easily take into account correlation within subjects, and deals
directly with both continuous and discrete response longitudinal data
under the framework of generalized linear models. To date, it has been
shown that the QIF has the following important features:
(1) The QIF yields a more efficient estimator than the currently
popular generalized estimation equation (GEE) approach (available on
SAS PROC GENMOD) when the working correlation is misspecified, and is
as efficient as the GEE when the working correlation is correctly
specified;
(2) The QIF is advantageous over the GEE because it does not involve
the estimation of the correlation parameters;
(3) The QIF is robust against a small portion of outliers/contaminated
data (Qu and Song, 2004), as opposed to the GEE being highly sensitive
to even one outlier; and,
(4) The QIF is analogous to twice the negative loglikelihood, so it
naturally provides for hypothesis testing and goodness-of-fit tests for
model assumptions, as well as defining model selection criteria, such
as the Akaike Information Criterion (AIC) and Bayesian Information
Criterion (BIC).
Instructors:
Peter Song is an Associate Professor in
the Biostatistics Research Unit, Department of Statistics and Actuarial Science,
University of Waterloo. He received his PhD in Statistics from the University of
British Columbia, 1996. Prior to the appointment at Waterloo, he was a visiting
Associate Professor at Department of Biostatistics, University of Michigan
School of Public Health (2002-2003), and Associate/Assistant Professor at
Department of Mathematics and Statistics (1996-2004), York University.
Peter Song's research interests include
bioinformatics, biostatistics, generalized linear models, missing data problems
in clinical trials, time series analysis, and longitudinal data analysis. He is
eager to work both applied and theoretical problems in modelling and
inference, with strong motivation from real world data analysis. He has
published over 30 statistical methodology papers in various statistical
journals. For more details, visit his personal webpage:
www.stats.uwaterloo.ca/~song
Annie Qu has been an Associate
Professor in the Statistics Department at Oregon State University since
2004. She was a visiting Associate Professor at the University of
Washington (2005-2006), the MD Anderson Cancer Center (2005-2006), and
was also Assistant Staff at the Cleveland Clinic Foundation (1999).
Dr. Qu's research interests include
Longitudinal Data Analysis, Estimating Equations, Missing Data,
Robustness, Nonparametric Models, Mixed Effects Models, Model Selection
and Diagnostic Tests, Cell Cycle Microarray Data, and Biostatistics. She
has published papers in Biometrika, JASA, Journal of Royal Statistical
Society-B, Statistical Science, Biometrics, Statistica Sinica and other
scientific journals. Dr. Qu's research has been supported by the
National Science Foundation (NSF). She has received an NSF Career Award
and the Thomas T. Sugihara Young Faculty Research Award at Oregon State
University. Dr. Qu's website is:
www.stat.oregonstate.edu/people/qu