《纵向数据分析方法与应用(英文版)》旨在系统地介绍纵向数据分析的基本概念、理论设定和应用步骤,重点通过SAS计算机程序对实际数据进行分析,从而深入浅出地描述纵向数据分析的各类模型。书中涉及的统计方法包括各类描述性估算法、线性混合效应模型、随机效应的统计推断及估计、残差协方差结构类型、广义线性混合效应模型的理论性描述、二分组结局混合效应模型、多结局混合效应模型、各类潜变量发展模型、缺损数据分类及分析方法以及一些纵向分析方面的专题研究。
本书着重于各类纵向分析模型的实际运用,而不拘泥于模型的纯理论推论,从而使对纵向数据分析有兴趣的科研人员以及大学生、研究生从中受益。
- Biography
- Preface
- CHAPTER 1 Introduction
- 1.1 What is Longitudinal Data Analysis?
- 1.2 History of Longitudinal Analysis and its Progress
- 1.3 Longitudinal Data Structures
- 1.3.1 Multivariate Data Structure
- 1.3.2 Univariate Data Structure
- 1.3.3 Balanced and Unbalanced Longitudinal Data
- 1.4 Missing Data Patterns and Mechanisms.
- 1.5 Sources of Correlation in Longitudinal Processes
- 1.6 Time Scale and the Number of Time Points
- 1.7 Basic Expressions of Longitudinal Modeling
- 1.8 Organization of the Book and Data Used for Illustrations.
- 1.8.1 Randomized Controlled Clinical Trial on the Effectiveness of Acupuncture Treatment on PTSD
- 1.8.2 Asset and Health Dynamics Among the Oldest Old (AHEAD)
- CHAPTER 2 Traditional Methods of Longitudinal Data Analysis
- 2.1 Descriptive Approaches
- 2.1.1 Time Plots of Trends
- 2.1.2 Paired t-Test
- 2.1.3 Effect Size Between Two Means and its Confi dence Interval
- 2.1.4 Empirical Illustration: Descriptive Analysis on the Effectiveness of Acupuncture Treatment in Reduction of PTSD Symptom Severity
- 2.2 Repeated Measures ANOVA
- 2.2.1 Specifi cations of One-Factor ANOVA
- 2.2.2 One-Factor Repeated Measures ANOVA
- 2.2.3 Specifi cations of Two-Factor Repeated Measures ANOVA
- 2.2.4 Empirical Illustration: A Two-Factor Repeated Measures ANOVA – The Effectiveness of Acupuncture Treatment on PCL Revisited
- 2.3 Repeated Measures MANOVA
- 2.3.1 General MANOVA
- 2.3.2 Hypothesis Testing on Effects in MANOVA
- 2.3.3 Repeated Measures MANOVA
- 2.3.4 Empirical Illustration: A Two-Factor Repeated Measures MANOVA on the Effectiveness of Acupuncture Treatment on Two Psychiatric Disorders
- 2.4 Summary
- CHAPTER 3 Linear Mixed-Effects Models
- 3.1 Introduction of Linear Mixed Models: Three Cases
- 3.1.1 Case I: One-Factor Linear Mixed Model with Random Intercept
- 3.1.2 Case II: linear Mixed Model with Random Intercept and Random Slope.
- 3.1.3 Case III: Linear Mixed Model with Random Effects and Three Covariates
- 3.2 Formalization of Linear Mixed Models
- 3.2.1 General Specifi cation of Linear Mixed Models
- 3.2.2 Variance–Covariance Matrix and Intraindividual Correlation
- 3.2.3 Formalization of Variance–Covariance Components
- 3.3 Inference and Estimation of Fixed Effects In Linear Mixed Models
- 3.3.1 Maximum Likelihood Methods
- 3.3.2 Statistical Inference and Hypothesis Testing on Fixed Effects
- 3.3.3 Missing Data
- 3.4 Trend Analysis
- 3.4.1 Polynomial Time Functions
- 3.4.2 Methods to Reduce Collinearity in Polynomial Time Terms
- 3.4.3 Numeric Checks on Polynomial Time Functions
- 3.5 Empirical Illustrations: Application of Two Linear Mixed Models
- 3.5.1 Linear Mixed Model on Effectiveness of Acupuncture Treatment on PCL Score
- 3.5.2 Linear Mixed Model on Marital Status and Disability Severity in Older Americans
- 3.6 Summary
- CHAPTER 4 Restricted Maximum Likelihood and Inference of Random Effects in Linear Mixed Models
- 4.1 Overview of Bayesian Inference
- 4.2 Restricted Maximum Likelihood Estimator
- 4.2.1 MLE Bias in Variance Estimate in General Linear Models
- 4.2.2 Specifi cation of REML in General Linear Models
- 4.2.3 REML Estimator in Linear Mixed Models
- 4.2.4 Justifi cation of the Restricted Maximum Likelihood Method
- 4.2.5 Comparison Between ML and REML Estimators
- 4.3 Computational Procedures
- 4.3.1 Newton–Raphson Algorithm
- 4.3.2 Expectation–Maximization Algorithm
- 4.4 Approximation of Random Effects in Linear Mixed Models
- 4.4.1 Best Linear Unbiased Prediction
- 4.4.2 Shrinkage and Reliability
- 4.5 Hypothesis Testing on Variance Component G
- 4.6 Empirical Illustrations: Linear Mixed Models with REML
- 4.6.1 Linear Mixed Model on Effectiveness of Acupuncture Treatment on PCL Score
- 4.6.2 Linear Mixed Model on Marital Status and Disability Among Older Americans
- 4.7 Summary
- CHAPTER 5 Patterns of Residual Covariance Structure
- 5.1 Residual Covariance Pattern Models with Equal Spacing
- 5.1.1 Compound Symmetry (CS)
- 5.1.2 Unstructured Pattern (UN)
- 5.1.3 Autoregressive Structures – AR(1) and ARH(1)
- 5.1.4 Toeplitz Structures – TOEP and TOEPH
- 5.2 Residual Covariance Pattern Models with Unequal Time Intervals
- 5.2.1 Spatial Power Model – SP(POW)
- 5.2.2 Spatial Exponential Model – SP(EXP)
- 5.2.3 Spatial Gaussian Model – SP(GAU)
- 5.2.4 Hybrid Residual Covariance Model
- 5.3 Comparison of Covariance Structures
- 5.4 Scaling of Time as a Classifi cation Factor
- 5.4.1 Scaling Approaches for Classifi cation Factors.
- 5.4.2 Coding Schemes of Time as a Classifi cation Factor.
- 5.5 Least Squares Means, Local Contrasts, and Local Tests
- 5.5.1 Least Squares Means
- 5.5.2 Local Contrasts and Local Tests
- 5.6 Empirical Illustrations: Estimation of Two Linear Regression Models
- 5.6.1 Linear Regression Model on Effectiveness of Acupuncture Treatment on PCL Score
- 5.6.2 Linear Regression Model on Marital Status and Disability Severity Among Older Americans
- 5.7 Summary
- CHAPTER 6 Residual and Infl uence Diagnostics
- 6.1 Residual Diagnostics
- 6.1.1 Types of Residuals in Linear Regression Models
- 6.1.2 Semivariogram in Random Intercept Linear Models
- 6.1.3 Semivariogram in the Linear Random Coeffi cient Model
- 6.2 Infl uence Diagnostics
- 6.2.1 Cook’s D and Related Infl uence Diagnostics
- 6.2.2 Leverage
- 6.2.3 DFFITS, MDFFITS, COVTRACE, and COVRATIO Statistics
- 6.2.4 Likelihood Displacement Statistic Approximation
- 6.2.5 LMAX Statistic for Identifi cation of Infl uential Observations
- 6.3 Empirical Illustrations on Infl uence Diagnostics
- 6.3.1 Infl uence Checks on Linear Mixed Model Concerning Effectiveness of Acupuncture Treatment on PCLScore
- 6.3.2 Infl uence Diagnostics on Linear Mixed Model Concerning Marital Status and Disability Severity Among Older Americans
- 6.4 Summary
- CHAPTER 7 Special Topics on Linear Mixed Models
- 7.1 Adjustment of Baseline Response in Longitudinal Data Analysis
- 7.1.1 Adjustment of Baseline Score and the Lord’s Paradox
- 7.1.2 Adjustment of Baseline Score in Longitudinal Data Analysis
- 7.1.3 Empirical Illustrations on Adjustment of Baseline Score
- 7.2 Misspecifi cation of the Assumed Distribution of Random Effects
- 7.2.1 Heterogeneity Linear Mixed Model
- 7.2.2 Nonnormal Random Effect Distribution in Linear Mixed Models
- 7.2.3 Best Predicted Random Effects in Different Distributions
- 7.2.4 Empirical Illustration: Comparison Between Blup and Least Squares Means.
- 7.3 Pattern-Mixture Modeling
- 7.3.1 Classifi cation of Heterogeneous Groups
- 7.3.2 Basic Theory of Pattern-Mixture Modeling
- 7.3.3 Pattern-Mixture Model.
- 7.3.4 Empirical Illustration of Pattern-Mixture Modeling
- 7.4 Summary
- CHAPTER 8 Generalized Linear Mixed Models on Nonlinear Longitudinal Data
- 8.1 A Brief Overview of Generalized Linear Models
- 8.2 Generalized Linear Mixed Models and Statistical Inferences
- 8.2.1 Basic Specifi cations of Generalized Linear Mixed Models
- 8.2.2 Statistical Inference and Likelihood Functions
- 8.2.3 Procedures of Maximization and Hypothesis Testing on Fixed Effects
- 8.2.4 Hypothesis Testing on Variance Components
- 8.3 Methods of Estimating Parameters in Generalized Linear Mixed Models
- 8.3.1 Penalized Quasi-Likelihood Method
- 8.3.2 Marginal Quasi-Likelihood Method
- 8.3.3 The Laplace Method
- 8.3.4 Gaussian Quadrature and Adaptive Gaussian Quadrature Methods
- 8.3.5 Markov Chain Monte Carlo Methods
- 8.4 Nonlinear Predictions and Retransformation of Random Components
- 8.4.1 Best Linear Unbiased Prediction Based on Linearization
- 8.4.2 Empirical Bayes BLUP
- 8.4.3 Retransformation Method
- 8.5 Some Popular Specifi c Generalized Linear Mixed Models
- 8.5.1 Mixed-Effects Logistic Regression Model
- 8.5.2 Mixed-Effects Ordered Logistic Model
- 8.5.3 Mixed-Effects Multinomial Logit Regression Models
- 8.5.4 Mixed-Effects Poisson Regression Models
- 8.5.5 Survival Models
- 8.6 Summary
- CHAPTER 9 Generalized Estimating Equations (GEEs) Models
- 9.1 Basic Specifi cations and Inferences of GEEs
- 9.1.1 Specifi cations of “Naïve” Model with Independence Hypothesis
- 9.1.2 Basic Specifi cations of GEEs
- 9.1.3 Specifi cations of Working Correlation Matrix
- 9.1.4 Quasi-Likelihood Information Criteria for GEEs
- 9.2 Other GEE Approaches
- 9.2.1 Prentice’s GEE Approach
- 9.2.2 Zhao and Prentice’s GEE Method (GEE2).
- 9.2.3 GEE Models on Odds Ratios
- 9.3 Relationship Between Marginal and Random-Effects Models
- 9.3.1 Comparison Between the Two Approaches
- 9.3.2 Use of GEEs to Fit a Conditional Model
- 9.4 Empirical Illustration: Effect of Marital Status on Disability Severity in Older Americans
- 9.5 Summary
- CHAPTER 10 Mixed-Effects Regression Model for Binary Longitudinal Data
- 10.1 Overview of Conventional Logistic and Probit Regression Models
- 10.2 Specifi cation of Random Intercept Logistic Regression Model
- 10.3 Specifi cation of Random Coeffi cient Logistic Regression Model
- 10.4 Inference of Mixed-Effects Logit Model
- 10.5 Approximation of Variance for Predicted Response Probability
- 10.6 Interpretability of Regression Coeffi cients and Odds Ratios
- 10.7 Computation of Conditional Effect and Conditional Odds Ratio for a Covariate
- 10.8 Empirical Illustration: Effect of Marital Status on Probability of Disability Among Older Americans
- 10.8.1 Analytic Plan
- 10.8.2 Analytic Steps with SAS Programs
- 10.8.3 Analytic Results
- 10.8.4 Nonlinear Predictions
- 10.8.5 Graphical Results
- 10.9 Summary.
- CHAPTER 11 Mixed-Effects Multinomial Logit Model for Nominal Outcomes
- 11.1 Overview of Multinomial Logistic Regression Model
- 11.2 Mixed-Effects Multinomial Logit Models and Nonlinear Predictions
- 11.3 Estimation of Fixed and Random Effects
- 11.4 Approximation of Variance–Covariance Matrix on Probabilities
- 11.5 Conditional Effects of Covariates on Probability Scale.
- 11.6 Empirical Illustration: Marital Status and Longitudinal Trajectories of Disability and Mortality AmongOlder Americans.
- 11.6.1 Data, Measures, and Models
- 11.6.2 Analytic Steps with SAS Programs
- 11.6.3 Analytic Results and Nonlinear Predictions
- 11.6.4 Conditional Effects of Marital Status
- 11.6.5 Graphical Analysis on Nonlinear Predictions
- 11.7 Summary
- CHAPTER 12 Longitudinal Transition Models for Categorical Response Data
- 12.1 Overview of Two-Time Multinomial Transition Modeling
- 12.2 Longitudinal Transition Models with Only Fixed Effects
- 12.3 Mixed-Effects Multinomial Logit Transition Models
- 12.3.1 Random Intercept Multinomial Logit Transition Model
- 12.3.2 Random Coeffi cient Multinomial Logit Transition Model
- 12.3.3 Statistical Inference of Mixed-Effects Multinomial Logit Transition Model
- 12.3.4 Approximation of Variance–Covariance Matrix for Transition Probabilities
- 12.3.5 Creation of Separate Multinomial Logit Transition Models
- 12.4 Empirical Illustration: Predicted Transition Probabilities in Functional Status and Marital Status Among Older Americans
- 12.4.1 Measures, Models, and SAS Programs
- 12.4.2 Prediction of Transition Probabilities
- 12.4.3 Effects of Marital Status on Transition Probabilities
- 12.5 Summary
- CHAPTER 13 Latent Growth, Latent Growth Mixture,and Group-Based Models
- 13.1 Overview of Structural Equation Modeling
- 13.2 Latent Growth Model
- 13.3 Latent Growth Mixture Model
- 13.4 Group-Based Model
- 13.5 Empirical Illustration: Effect of Marital Status on ADL Count Among Older Americans Revisited
- 13.6 Summary
- CHAPTER 14 Methods for Handling Missing Data
- 14.1 Mathematical Defi nitions of MCAR, MAR, and MNAR
- 14.2 Methods Handling Missing at Random
- 14.2.1 Simple Approaches
- 14.2.2 Last Observation Carried Forward
- 14.2.3 Multiple Imputations.
- 14.2.4 Comparison Between MI and Shrinkage
- 14.2.5 Empirical Illustration: Comparison of AnalyticResults with and without Multiple Imputationson Missing Data
- 14.3 Methods Handling Missing Not at Random.
- 14.3.1 Impact of Nonignorable Missing Data in Longitudinal Data Analysis
- 14.3.2 Selection Model on MNAR
- 14.3.3 Pattern Mixture Model on MNAR.
- 14.3.4 Nonparametric Regression Model on NonignorableMissing Data
- 14.3.5 Empirical Illustration: Comparison of Different Methods for Handling Nonignorable Missing Data
- 14.4 Summary
- Appendix A Orthogonal Polynomials
- Appendix B The Delta Method
- Appendix C Quasi-Likelihood Functions and Properties
- Appendix D Model Specifi cation and SAS Program for Random Coeffi cient Multinomial Logit Model on Health State Among Older Americans
- References
- Subject Index
- 版权