Bayesian Statistics for the Social Sciences, Second Edition

David Kaplan, PhD, is the Patricia Busk Professor of Quantitative Methods in the Department of Educational Psychology at the University of Wisconsin–Madison and holds affiliate appointments in the University of Wisconsin’s Department of Population Health Sciences, the Center for Demography and Ecology, and the Nelson Institute for Environmental Studies. Dr. Kaplan’s research focuses on the development of Bayesian statistical methods for education research. His work on these topics is directed toward applications to large-scale cross-sectional and longitudinal survey designs. He has been actively involved in the OECD Program for International Student Assessment (PISA), serving on its Technical Advisory Group from 2005 to 2009 and its Questionnaire Expert Group from 2004 to the present, and chairing the Questionnaire Expert Group for PISA 2015. He also serves on the Design and Analysis Committee and the Questionnaire Standing Committee for the National Assessment of Educational Progress. Dr. Kaplan is an elected member of the National Academy of Education and former chair of its Research Advisory Committee, president (2023–2024) of the Psychometric Society, and past president of the Society for Multivariate Experimental Psychology. He is a fellow of the American Psychological Association (Division 5), a former visiting fellow at the Luxembourg Institute for Social and Economic Research, a former Jeanne Griffith Fellow at the National Center for Education Statistics, and a current fellow at the Leibniz Institute for Educational Trajectories in Bamberg, Germany. He is a recipient of the Samuel J. Messick Distinguished Scientific Contributions Award from the American Psychological Association (Division 5), the Alexander von Humboldt Research Award, and the Hilldale Award for the Social Sciences from the University of Wisconsin–Madison. Dr. Kaplan was the Johann von Spix International Visiting Professor at the Universität Bamberg and the Max Kade Visiting Professor at the Universität Heidelberg, both in Germany, and is currently International Guest Professor at the Universität Heidelberg.

"This very practical book is well suited to social science students because of the examples used (large-scale surveys) and the coverage of methods that social scientists often need (latent variables, variable selection, and dealing with missing data). The book also covers some topics readers may not know they need--Bayesian model averaging and workflow, for example. Illustrations use RStan, perhaps the most flexible of programs for Bayesian modeling. Full integration of RStan input and output is provided in the text."--David Rindskopf, PhD, Distinguished Professor of Educational Psychology and Psychology, The Graduate Center, The City University of New York"Kaplan's book is the perfect follow-up for those whose curiosity has been piqued about Bayesian statistics. The many code examples will give users a head start for applying Bayes' theorem to their data. I highly appreciate that the author uses open-source software for all models. The topics are introduced with a rich amount of background information, some equations (but never too many), detailed explanations, and code examples. Empirical results are used to illustrate each topic."--Rens van de Schoot, PhD, Department of Methodology and Statistics, Utrecht University, Netherlands"An excellent resource for researchers at the graduate level or above with an interest in Bayesian statistics. Readers are skillfully guided through the process of statistical reasoning from a Bayesian perspective. This book is practical and minimally technical while also introducing readers to interesting historical and philosophical issues. What makes the book especially helpful is Kaplan’s careful balance of breadth and depth of coverage of key topics. In this timely second edition, important recent advances in Bayesian statistics are distilled and disseminated for researchers in the social sciences."--Sierra A. Bainter, PhD, Department of Psychology, University of Miami"This book has all the essential components to help readers, especially quantitative researchers in social sciences, understand and conduct Bayesian modeling. The second edition includes new material on recent Markov chain Monte Carlo (MCMC) methods, such as Hamiltonian MC, in addition to a range of other updates."--Insu Paek, PhD, Senior Scientist, Human Resources Research Organization"I recommend this book for providing a careful overview of the Bayesian framework, at a level accessible to a wide audience, with examples, code, and key references. Kaplan does a great job of covering so many different aspects of Bayesian modeling in a coherent way and presenting a number of substantive methods for analyzing complex data. I liked the comparisons and analogies to the frequentist approach."--Irini Moustaki, PhD, Department of Statistics, London School of Economics and Political Science, United Kingdom-A valuable read for researchers, practitioners, teachers, and graduate students in the field of social sciences….Extremely accessible and incredibly delightful….The wide breadth of topics covered, along with the author's clear and engaging style of writing and inclusion of numerous examples, should provide an adequate foundation for any psychologist wishing to take a leap into Bayesian thinking. Furthermore, the technical details and analytic aspects provided in all chapters should equip readers with enough knowledge to embark on Bayesian analysis with their own research data. (on the first edition)--Psychometrika, 3/1/2017

• I. Foundations1. Probability Concepts and Bayes' Theorem1.1 Relevant Probability Axioms1.1.1 The Kolmogorov Axioms of Probability1.1.2 The Rényi Axioms of Probability1.2 Frequentist Probability1.3 Epistemic Probability1.3.1 Coherence and the Dutch Book1.3.2 Calibrating Epistemic Probability Assessment1.4 Bayes' Theorem1.4.1 The Monty Hall Problem1.5 Summary2. Statistical Elements of Bayes' Theorem2.1 Bayes' Theorem Revisited2.2. Hierarchical Models and Pooling2.3 The Assumption of Exchangeability2.4 The Prior Distribution2.4.1 Non-informative Priors2.4.2 Jeffreys' Prior2.4.3 Weakly Informative Priors2.4.4 Informative Priors2.4.5 An Aside: Cromwell's Rule2.5 Likelihood2.5.1 The Law of Likelihood2.6 The Posterior Distribution2.7 The Bayesian Central Limit Theorem and Bayesian Shrinkage2.8 Summary3. Common Probability Distributions and Their Priors3.1 The Gaussian Distribution3.1.1 Mean Unknown, Variance Known: The Gaussian Prior3.1.2 The Uniform Distribution as a Non-informative Prior3.1.3 Mean Known, Variance Unknown: The Inverse-Gamma Prior3.1.4 Mean Known, Variance Unknown: The Half-Cauchy Prior3.1.5 Jeffreys' Prior for the Gaussian Distribution3.2 The Poisson Distribution3.2.1 The Gamma Prior3.2.2 Jeffreys' Prior for the Poisson Distribution3.3 The Binomial Distribution3.3.1 The Beta Prior3.3.2 Jeffreys' Prior for the Binomial Distribution3.4 The Multinomial Distribution3.4.1 The Dirichlet Prior3.4.2 Jeffreys' Prior for the Multinomial Distribution3.5 The Inverse-Wishart Distribution3.6 The LKJ Prior for Correlation Matrices3.7 Summary4. Obtaining and Summarizing the Posterior Distribution4.1 Basic Ideas of Markov Chain Monte Carlo Sampling4.2 The Random Walk Metropolis–Hastings Algorithm4.3 The Gibbs Sampler4.4 Hamiltonian Monte Carlo4.4.1 No-U-Turn (NUTS) Sampler4.5 Convergence Diagnostics4.5.1 Trace Plots4.5.2 Posterior Density Plots4.5.3 Auto-Correction Plots4.5.4 Effective Sample Size4.5.5 Potential Scale Reduction Factor4.5.6 Possible Error Messages When Using HMC/NUTS4.6 Summarizing the Posterior Distribution4.6.1 Point Estimates of the Posterior Distribution4.6.2 Interval Summaries of the Posterior Distribution4.7 Introduction to Stan and Example4.8 An Alternative Algorithm: Variational Bayes4.8.1 Evidence Lower Bound (ELBO)4.8.2 Variational Bayes Diagnostics4.9 SummaryII. Bayesian Model Building5. Bayesian Linear and Generalized Models5.1 The Bayesian Linear Regression Model5.1.1 Non-informative Priors in the Linear Regression Model5.2 Bayesian Generalized Linear Models5.2.1 The Link Function5.3 Bayesian Logistic Regression5.4 Bayesian Multinomial Regression5.5 Bayesian Poisson Regression5.6 Bayesian Negative Binomial Regression5.7 Summary6. Model Evaluation and Comparison6.1 The Classical Approach to Hypothesis Testing and Its Limitations6.2 Model Assessment6.2.1 Prior Predictive Checking6.2.2 Posterior Predictive Checking6.3 Model Comparison6.3.1 Bayes Factors6.3.2 The Deviance Information Criterion (DIC)6.3.3 Widely Applicable Information Criterion (WAIC)6.3.4 Leave-One-Out Cross-Validation6.3.5 A Comparison of WAIC and LOO6.4 Summary7. Bayesian Multilevel Modeling7.1 Revisiting Exchangeability7.2 Bayesian Random Effects Analysis of Variance7.3 Bayesian Intercepts as Outcomes Model7.4 Bayesian Intercepts and Slopes as Outcomes Model7.5 Summary8. Bayesian Latent Variable Modeling8.1 Bayesian Estimation for the CFA8.1.1 Priors for CFA Model Parameters8.2 Bayesian Latent Class Analysis8.2.1 The Problem of Label-Switching and a Possible Solution8.2.2 Comparison of VB to the EM Algorithm8.3 SummaryIII. Advanced Topics and Methods9. Missing Data From a Bayesian Perspective9.1 A Nomenclature for Missing Data9.2 Ad Hoc Deletion Methods for Handling Missing Data9.2.1 Listwise Deletion9.2.2 Pairwise Deletion9.3 Single Imputation Methods9.3.1 Mean Imputation9.3.2 Regression Imputation9.3.3 Stochastic Regression Imputation9.3.4 Hot Deck Imputation9.3.5 Predictive Mean Matching9.4 Bayesian Methods for Multiple Imputation9.4.1 Data Augmentation9.4.2 Chained Equations9.4.3 EM Bootstrap: A Hybrid Bayesian/Frequentist Methods9.4.4 Bayesian Bootstrap Predictive Mean Matching9.4.5 Accounting for Imputation Model Uncertainty9.5 Summary10. Bayesian Variable Selection and Sparsity10.1 Introduction10.2 The Ridge Prior10.3 The Lasso Prior10.4 The Horseshoe Prior10.5 Regularized Horseshoe Prior10.6 Comparison of Regularization Methods10.6.1 An Aside: The Spike-and-Slab Prior10.7 Summary11. Model Uncertainty11.1 Introduction11.2 Elements of Predictive Modeling11.2.1 Fixing Notation and Concepts11.2.2 Utility Functions for Evaluating Predictions11.3 Bayesian Model Averaging11.3.1 Statistical Specification of BMA11.3.2 Computational Considerations11.3.3 Markov Chain Monte Carlo Model Composition11.3.4 Parameter and Model Priors11.3.5 Evaluating BMA Results: Revisiting Scoring Rules11.4 True Models, Belief Models, and M-Frameworks11.4.1 Model Averaging in the M-Closed Framework11.4.2 Model Averaging in the M-Complete Framework11.4.3 Model Averaging in the M-Open Framework11.5 Bayesian Stacking11.5.1 Choice of Stacking Weights11.6 Summary12. Closing Thoughts12.1 A Bayesian Workflow for the Social Sciences12.2 Summarizing the Bayesian Advantage12.2.1 Coherence12.2.2 Conditioning on Observed Data12.2.3 Quantifying Evidence12.2.4 Validity12.2.5 Flexibility in Handling Complex Data Structures12.2.6 Formally Quantifying UncertaintyList of Abbreviations and AcronymsReferencesAuthor IndexSubject Index