Meta-analysis : a structural equation modeling approach /
Presents a novel approach to conducting meta-analysis using structural equation modeling. Structural equation modeling (SEM) and meta-analysis are two powerful statistical methods in the educational, social, behavioral, and medical sciences. They are often treated as two unrelated topics in the lite...
Main Author: | |
---|---|
Corporate Authors: | , |
Format: | Book |
Language: | English |
Published: |
Chichester, England ; West Sussex, England :
Wiley,
2015
Chichester, West Sussex, United Kingdom : 2015 |
Subjects: |
Table of Contents:
- 1 Introduction 1
- 1.1 What is meta-analysis? 1
- 1.2 What is structural equation modeling? 2
- 1.3 Reasons for writing a book on meta-analysis and structural equation modeling 3
- 1.3.1 Benefits to users of structural equation modeling and meta-analysis 6
- 1.4 Outline of the following chapters 6
- 1.4.1 Computer examples and data sets used in this book 8
- 1.5 Concluding remarks and further readings 8
- References 9
- 2 Brief review of structural equation modeling 13
- 2.1 Introduction 13
- 2.2 Model specification 14
- 2.2.1 Equations 14
- 2.2.2 Path diagram 15
- 2.2.3 Matrix representation 15
- 2.3 Common structural equation models 18
- 2.3.1 Path analysis 18
- 2.3.2 Confirmatory factor analysis 19
- 2.3.3 Structural equation model 21
- 2.3.4 Latent growth model 22
- 2.3.5 Multiple-group analysis 23
- 2.4 Estimation methods, test statistics, and goodness-of-fit indices 25
- 2.4.1 Maximum Likelihood estimation 25
- 2.4.2 Weighted least squares 26
- 2.4.3 Multiple-group analysis 28
- 2.4.4 Likelihood ratio test and Wald test 28
- 2.4.5 Confidence intervals on parameter estimates 29
- 2.4.6 Test statistics versus goodness-of-fit indices 34
- 2.5 Extensions on structural equation modeling 38
- 2.5.1 Phantom variables 38
- 2.5.2 Definition variables 39
- 2.5.3 Full information maximum likelihood estimation 41
- 2.6 Concluding remarks and further readings 42
- References 42
- 3 Computing effect sizes for meta-analysis 48
- 3.1 Introduction 48
- 3.2 Effect sizes for univariate meta-analysis 50
- 3.2.1 Mean differences 50
- 3.2.2 Correlation coefficient and its Fisher's z transformation 55
- 3.2.3 Binary variables 56
- 3.3 Effect sizes for multivariate meta-analysis 57
- 3.3.1 Mean differences 57
- 3.3.2 Correlation matrix and its Fisher's z transformation 59
- 3.3.3 Odds ratio 60
- 3.4 General approach to estimating the sampling variances and covariances 60
- 3.4.1 Delta method 61
- 3.4.2 Computation with structural equation modeling 64
- 3.5 Illustrations Using R 68
- 3.5.1 Repeated measures 69
- 3.5.2 Multiple treatment studies 71
- 3.5.3 Multiple-endpoint studies 73
- 3.5.4 Multiple treatment with multiple-endpoint studies 75
- 3.5.5 Correlation matrix 77
- 3.6 Concluding remarks and further readings 78
- References 78
- 4 Univariate meta-analysis 81
- 4.1 Introduction 81
- 4.2 Fixed-effects model 83
- 4.2.1 Estimation and hypotheses testing 83
- 4.2.2 Testing the homogeneity of effect sizes 85
- 4.2.3 Treating the sampling variance as known versus as estimated 85
- 4.3 Random-effects model 87
- 4.3.1 Estimation and hypothesis testing 88
- 4.3.2 Testing the variance component 90
- 4.3.3 Quantifying the degree of the heterogeneity of effect sizes 92
- 4.4 Comparisons between the fixed- and the random-effects models 93
- 4.4.1 Conceptual differences 93
- 4.4.2 Statistical differences 94
- 4.5 Mixed-effects model 96
- 4.5.1 Estimation and hypotheses testing 97
- 4.5.2 Explained variance 98
- 4.5.3 A cautionary note 99
- 4.6 Structural equation modeling approach 100
- 4.6.1 Fixed-effects model 100
- 4.6.2 Random-effects model 101
- 4.6.3 Mixed-effects model 102
- 4.7 Illustrations using R 105
- 4.7.1 Odds ratio of atrial fibrillation between bisphosphonate and non-bisphosphonate users 105
- 4.7.2 Correlation between organizational commitment and salesperson job performance 108
- 4.8 Concluding remarks and further readings 116
- References 117
- 5 Multivariate meta-analysis 121
- 5.1 Introduction 121
- 5.1.1 Types of dependence 121
- 5.1.2 Univariate meta-analysis versus multivariate meta-analysis 122
- 5.2 Fixed-effects model 124
- 5.2.1 Testing the homogeneity of effect sizes 125
- 5.2.2 Estimation and hypotheses testing 126
- 5.3 Random-effects model 127
- 5.3.1 Structure of the variance component of random effects 128
- 5.3.2 Nonnegative definite of the variance component of random effects 129
- 5.3.3 Estimation and hypotheses testing 131
- 5.3.4 Quantifying the degree of heterogeneity of effect sizes 132
- 5.3.5 When the sampling covariances are not known 133
- 5.4 Mixed-effects model 134
- 5.4.1 Explained variance 135
- 5.5 Structural equation modeling approach 136
- 5.5.1 Fixed-effects model 136
- 5.5.2 Random-effects model 137
- 5.5.3 Mixed-effects model 138
- 5.6 Extensions: mediation and moderation models on the effect sizes 140
- 5.6.1 Regression model 141
- 5.6.2 Mediating model 143
- 5.6.3 Moderating model 144
- 5.7 Illustrations using R 145
- 5.7.1 BCG vaccine for preventing tuberculosis 146
- 5.7.2 Standardized mean differences between males and females on life satisfaction and life control 156
- 5.7.3 Mediation and moderation models 161
- 5.8 Concluding remarks and further readings 174
- References 174
- 6 Three-level meta-analysis 179
- 6.1 Introduction 179
- 6.1.1 Examples of dependent effect sizes with unknown degree of dependence 180
- 6.1.2 Common methods to handling dependent effect sizes 180
- 6.2 Three-level model 183
- 6.2.1 Random-effects model 183
- 6.2.2 Mixed-effects model 187
- 6.3 Structural equation modeling approach 188
- 6.3.1 Two representations of the same model 189
- 6.3.2 Random-effects model 191
- 6.3.3 Mixed-effects model 193
- 6.4 Relationship between the multivariate and the three-level meta-analyses 195
- 6.4.1 Three-level meta-analysis as a special case of the multivariate meta-analysis 195
- 6.4.2 Approximating a multivariate meta-analysis with a three-level meta-analysis 196
- 6.4.3 Three-level multivariate meta-analysis 198
- 6.5 Illustrations using R 200
- 6.5.1 Inspecting the data 201
- 6.5.2 Fitting a random-effects model 202
- 6.5.3 Obtaining the likelihood-based confidence interval 203
- 6.5.4 Testing τ²₍₃₎ = 0 204
- 6.5.5 Testing τ²₍₂₎ = 0 205
- 6.5.6 Testing τ²₍₂₎ = &tau²₍₃₎ 205
- 6.5.7 Testing types of proposals (grant versus fellowship) 206
- 6.5.8 Testing the effect of the year of application 207
- 6.5.9 Testing the country effect 209
- 6.6 Concluding remarks and further readings 210
- References 211
- 7 Meta-analytic structural equation modeling 214
- 7.1 Introduction 214
- 7.1.1 Meta-analytic structural equation modeling as a possible solution for conflicting research findings 215
- 7.1.2 Basic steps for conducting a meta-analytic structural equation modeling 217
- 7.2 Conventional approaches 218
- 7.2.1 Univariate approaches 218
- 7.2.2 Generalized least squares approach 221
- 7.3 Two-stage structural equation modeling: fixed-effects models 223
- 7.3.1 Stage 1 of the analysis: pooling correlation matrices 224
- 7.3.2 Stage 2 of the analysis: fitting structural models 227
- 7.3.3 Subgroup analysis 233
- 7.4 Two-stage structural equation modeling: random-effects models 233
- 7.4.1 Stage 1 of the analysis: pooling correlation matrices 234
- 7.4.2 Stage 2 of the analysis: fitting structural models 235
- 7.5 Related issues 235
- 7.5.1 Multiple-group structural equation modeling versus meta-analytic structural equation modeling 236
- 7.5.2 Fixed-effects model: two-stage structural equation modeling versus generalized least squares 237
- 7.5.3 Alternative random-effects models 239
- 7.5.4 Maximum likelihood estimation versus restricted (or residual) maximum likelihood estimation 242
- 7.5.5 Correlation coefficient versus Fisher's z score 242
- 7.5.6 Correction for unreliability 243
- 7.6 Illustrations using R 244
- 7.6.1 A higher-order confirmatory factor analytic model for the Big Five model 244
- 7.6.2 A regression model on SAT (Math) 258
- 7.6.3 A path model for cognitive ability to supervisor rating 266
- 7.7 Concluding remarks and further readings 273
- References 274
- 8 Advanced topics in SEM-based meta-analysis 279
- 8.1 Restricted (or residual) maximum likelihood estimation 279
- 8.1.1 Reasons for and against the maximum likelihood estimation 280
- 8.1.2 Applying the restricted (or residual) maximum likelihood estimation in SEM-based meta-analysis 281
- 8.1.3 Implementation in structural equation modeling 283
- 8.2 Missing values in the moderators 289
- 8.2.1 Types of missing mechanisms 289
- 8.2.2 Common methods to handling missing data 290
- 8.2.3 Maximum likelihood estimation 291
- 8.3 Illustrations using R 294
- 8.3.1 Restricted (or residual) maximum likelihood estimation 295
- 8.3.2 Missing values in the moderators 300
- 8.4 Concluding remarks and further readings 309
- References 310
- 9 Conducting meta-analysis with Mplus 313
- 9.1 Introduction 313
- 9.2 Univariate meta-analysis 314
- 9.2.1 Fixed-effects model 314
- 9.2.2 Random-effects model 317
- 9.2.3 Mixed-effects model 322
- 9.2.4 Handling missing values in moderators 325
- 9.3 Multivariate meta-analysis 327
- 9.3.1 Fixed-effects model 328
- 9.3.2 Random-effects model 333
- 9.3.3 Mixed-effects model 337
- 9.3.4 Mediation and moderation models on the effect sizes 340
- 9.4 Three-level meta-analysis 346
- 9.4.1 Random-effects model 346
- 9.4.2 Mixed-effects model 351
- 9.5 Concluding remarks and further readings 353
- References 354