221 Pages 8 B/W Illustrations
    by Chapman & Hall

    Drawing on the authors’ substantial expertise in modeling longitudinal and clustered data, Quasi-Least Squares Regression provides a thorough treatment of quasi-least squares (QLS) regression—a computational approach for the estimation of correlation parameters within the framework of generalized estimating equations (GEEs). The authors present a detailed evaluation of QLS methodology, demonstrating the advantages of QLS in comparison with alternative methods. They describe how QLS can be used to extend the application of the traditional GEE approach to the analysis of unequally spaced longitudinal data, familial data, and data with multiple sources of correlation. In some settings, QLS also allows for improved analysis with an unstructured correlation matrix.

    Special focus is given to goodness-of-fit analysis as well as new strategies for selecting the appropriate working correlation structure for QLS and GEE. A chapter on longitudinal binary data tackles recent issues raised in the statistical literature regarding the appropriateness of semi-parametric methods, such as GEE and QLS, for the analysis of binary data; this chapter includes a comparison with the first-order Markov maximum-likelihood (MARK1ML) approach for binary data.

    Examples throughout the book demonstrate each topic of discussion. In particular, a fully worked out example leads readers from model building and interpretation to the planning stages for a future study (including sample size calculations). The code provided enables readers to replicate many of the examples in Stata, often with corresponding R, SAS, or MATLAB® code offered in the text or on the book’s website.

    When QLS Might Be Considered as an Alternative to GEE
    Motivating Studies

    Review of Generalized Linear Models
    Generalized Linear Models
    Generalized Estimating Equations
    Application for Obesity Study Provided in Chapter One

    Quasi-Least Squares Theory and Applications
    History and Theory of QLS Regression
    Why QLS Is a "Quasi" Least Squares Approach
    The Least-Squares Approach Employed in Stage One of QLS for Estimation of α
    Stage-Two QLS Estimates of the Correlation Parameter for the AR(1) Structure
    Algorithm for QLS
    Other Approaches That Are Based on GEE

    Mixed Linear Structures and Familial Data
    Notation for Data from Nuclear Families
    Familial Correlation Structures for Analysis of Data from Nuclear Families
    Other Work on Assessment of Familial Correlations with QLS
    Justification of Implementation of QLS for Familial Structures via Consideration of the Class of Mixed Linear Correlation Structures
    Demonstration of QLS for Analysis of Balanced Familial Data Using Stata Software
    Demonstration of QLS for Analysis of Unbalanced Familial Data Using R Software
    Simulations to Compare Implementation of QLS with Correct Specification of the Trio Structure versus Correct Specification with GEE and Incorrect Specification of the Exchangeable Working
    Structure with GEE
    Summary and Future Research Directions

    Correlation Structures for Clustered and Longitudinal Data
    Characteristics of Clustered and Longitudinal Data
    The Exchangeable Correlation Structure for Clustered Data
    The Tri-Diagonal Correlation Structure
    The AR(1) Structure for Analysis of (Planned) Equally Spaced Longitudinal Data
    The Markov Structure for Analysis of Unequally Spaced Longitudinal Data
    The Unstructured Matrix for Analysis of Balanced Data
    Other Structures
    Implementation of QLS for Patterned Correlation Structures

    Analysis of Data with Multiple Sources of Correlation
    Characteristics of Data with Multiple Sources of Correlation
    Multi-Source Correlated Data That Are Totally Balanced
    Multi-Source Correlated Data That Are Balanced within Clusters
    Multi-Source Correlated Data That Are Unbalanced
    Asymptotic Relative Efficiency Calculations

    Correlated Binary Data
    Additional Constraints for Binary Data
    When Violation of the Prentice Constraints for Binary Data Is Likely to Occur
    Implications of Violation of Constraints for Binary Data
    Comparison between GEE, QLS, and MARK1ML
    Prentice-Corrected QLS and GEE

    Assessing Goodness of Fit and Choice of Correlation Structure for QLS and GEE
    Simulation Scenarios
    Simulation Results
    Summary and Recommendations

    Sample Size and Demonstration
    Two-Group Comparisons
    More Complex Situations
    Worked Example
    Discussion and Summary



    Exercises appear at the end of each chapter.


    Justine Shults is an associate professor and co-director of the Pediatrics Section in the Division of Biostatistics in the Perelman School of Medicine at the University of Pennsylvania, where she is the principal investigator of the biostatistics training grant in renal and urologic diseases. She is the Statistical Editor of the Journal of the Pediatric Infectious Disease Society and the Statistical Section Editor of Springer Plus. Professor Shults (with N. Rao Chaganty) developed Quasi-Least Squares (QLS) and was funded by the National Science Foundation and the National Institutes of Health to extend QLS and develop user-friendly software for implementing her new methodology. She has authored or co-authored over 100 peer-reviewed publications, including the initial papers on QLS for unbalanced and unequally spaced longitudinal data and on MARK1ML and the choice of working correlation structure for GEE.

    Joseph M. Hilbe is a Solar System Ambassador with the Jet Propulsion Laboratory, an adjunct professor of statistics at Arizona State University, and an Emeritus Professor at the University of Hawaii. An elected fellow of the American Statistical Association and an elected member of the International Statistical Institute (ISI), Professor Hilbe is president of the International Astrostatistics Association as well as chair of the ISI Sports Statistics and Astrostatistics committees. He has authored two editions of the bestseller Negative Binomial Regression, Logistic Regression Models, and Astrostatistical Challenges for the New Astronomy. He also co-authored Methods of Statistical Model Estimation (with A. Robinson), Generalized Estimating Equations, Second Edition (with J. Hardin), and R for Stata Users (with R. Muenchen), as well as 17 encyclopedia articles and book chapters in the past five years.

    "The book does an excellent job of explaining basic concepts and techniques in the analysis of longitudinal and correlated data using QLS and GEE. Well-chosen data examples almost follow all the technical explanations, providing the readers a flavor on what problems QLS solves and how to solve those problems using software. Although the authors mainly use Stata to demonstrate the examples, they also provide web access to R, SAS, and MATLAB code and guidelines to replicate those examples, making the book appealing to a wide audience. The book also successfully incorporates some recent research work without raising its technical level. Therefore, the book will serve as a comprehensible guide to researchers who conduct analysis on correlated data. It would also be a good textbook for graduate students in statistics or biostatistics. Finally, I believe it would be a popular desk reference for methodology-oriented researchers who are interested in longitudinal studies and related fields."
    Journal of the American Statistical Association, March 2015

    "This book deals with the quasi-least squares (QLS) regression, presenting a computational approach for the estimation of correlation parameters in the context of the generalized estimating equations (GEEs). … The book is provided with illustrative examples for each topic."
    Zentralblatt MATH 1306