Multilevel Structural Equation Modeling. Bruno Castanho Silva
China Square Central
Singapore 048423
Printed in the United States of America
This book is printed on acid-free paper.
Library of Congress Cataloging-in-Publication Data
Names: Silva, Bruno Castanho, author. | Bosancianu, Constantin Manuel, author. | Littvay, Levente, author.
Title: Multilevel structural equation modeling / Bruno Castanho Silva (University of Cologne), Constantin Manuel Bosancianu (Wissenschaftszentrum Berlin für Sozialforschung), Levente Littvay (Central European University).
Description: Thousand Oaks, California : SAGE Publications, Inc., 2019. | Includes bibliographical references and index.
Identifiers: LCCN 2018057333 | ISBN 978-1-5443-2305-3 (pbk. : alk. paper)
Subjects: LCSH: Path analysis (Statistics) | Structural equation modeling. | Multilevel models (Statistics) | Regression analysis.
Classification: LCC QA278.3 .S55 2019 | DDC 519.5/3–dc23 LC record available at https://lccn.loc.gov/2018057333
19 20 21 22 23 10 9 8 7 6 5 4 3 2 1
Acquisitions Editor: Helen Salmon
Editorial Assistant: Megan O’Heffernan
Production Editor: Rebecca Lee
Copy Editor: Gillian Dickens
Typesetter: Integra
Proofreader: Wendy Jo Dymond
Indexer: Marilyn Augst
Cover Designer: Candice Harman
Marketing Manager: Shari Countryman
Series Editor’s Introduction
I am pleased to introduce Multilevel Structural Equation Modeling, by Bruno Castanho Silva, Constantin Manuel Cosancianu, and Levente Littvay. Multilevel structural equation models (MSEMs) combine the study of relationships between variables measured with error central to structural equation modeling (SEM) with an interest in macro-micro relationships central to multilevel models (MLM). This volume is well-organized with a clear progression of topics, starting with SEMs of observed variables, proceeding to confirmatory factor analysis (CFA), and then the full model, adding a multilevel component to each along the way. In each chapter, the authors proceed systematically from simpler to more complex model specifications, using examples to illustrate each step. Readers can practice by replicating these examples using materials available in the online appendix.
An innovation of the volume is the notation. SEMs and MLMs each have their own conventions; the authors blend them. They maintain notation used in standard multilevel texts but introduce superscripts to keep track of the outcome variable associated with particular coefficients. This approach makes it possible to write MSEM models as a series of equations, making the volume broadly accessible, even to readers not well-versed in matrix algebra. To further reinforce the reader understanding, they show most models both as a set of equations and also as graphical presentations that build on SEM traditions.
In terms of the preparation needed, readers having experience with structural equations models (SEM) or multilevel models (MLM) are in the best position to benefit from this volume. Chapter 1 provides a very helpful review of both, and then shows how the model and notation can be organized into a single framework for MSEM. Chapter 2 introduces multilevel path models, considering both the random intercept model and the random slopes model. It uses World Values Survey data, Wave 4, from 55 countries to explore the individual-level and country-level factors associated with high self-expression values (importance of civic activism, subjective well-being, tolerance and trust, personal autonomy and choice). Multilevel factor models are the focus of Chapter 3. This chapter uses data from the 2015 wave of the Program for International Student Assessment (PISA) on the use of digital devices in the Dominican Republic to build a two-level CFA, starting with a comparison of the multilevel CFA to the multiple group CFA, then building in random latent variable intercepts and finishing with a multilevel CFA with random loadings. It also includes a useful discussion of measurement invariance. Chapter 4 merges the subject matter of chapters 2 and 3 together into the full multilevel structural equation model (MSEM). The example for this chapter, based on the 2004 Workplace Employment Relations Survey teaching dataset, explores whether employees consider themselves to be under- or over-qualified for their jobs, as affected by their perception of how demanding their jobs are, how responsive their managers, their pay, and—at the company level—number of employees. Chapter 5 concludes the text by addressing some advanced topics such as categorical dependent variables, sampling weights, and missing data, pointing to references where the interested reader can learn more and providing advice on how to approach the technical literature.
Multilevel structural equation models can be quite complex. Indeed, as the authors say, the complexity of the models to be investigated is only limited by the imagination of the investigators (and of course, the data, software, etc.). Given this, readers will especially appreciate this hands-on introduction and the lengths to which the authors have gone to make the material accessible to researchers from a variety of backgrounds.
—Barbara Entwisle
Series Editor
About the Authors
Bruno Castanho Silvais a postdoctoral researcher at the Cologne Center for Comparative Politics (CCCP), University of Cologne. Bruno received his PhD from the Department of Political Science at Central European University and teaches introductory and advanced quantitative methods courses, including multilevel structural equation modeling and machine learning, at the European Consortium for Political Research Methods Schools. His methodological interests are on applications of structural equation models for scale development and causal analysis, as well as statistical methods of causal inference with observational and experimental data.Constantin Manuel Bosancianuis a postdoctoral researcher at the WZB Berlin Social Science Center, in the Institutions and Political Inequality research unit. He received his PhD from the Department of Political Science at Central European University in Budapest, Hungary, and has been an instructor for multiple statistics courses and workshops at the European Consortium for Political Research Methods Schools, at the Universities of Heidelberg, Giessen, and Zagreb, as well as at the Institute of Sociology of the Czech Academy of Sciences. Manuel’s methodological focus is on practical applications of multilevel models, Bayesian analysis, and the analysis of time-series cross-sectional data sets.Levente Littvayis an associate professor at Central European University’s Department of Political Science. He is a recipient of the institution’s Distinguished Teaching Award for graduate courses in research methods and applied statistics with a topical emphasis on political psychology, experiments, and American politics. He received an MA and a PhD in Political Science and an MS in Survey Research and Methodology from the University of Nebraska–Lincoln, has taught numerous methods workshops, and is an academic co-convenor of the European Consortium for Political Research Methods Schools. His research interests include populism, political socialization, and biological explanations of social and political attitudes and behaviors. He often works as a methodologist with medical researchers and policy analysts, co-runs the Hungarian Twin Registry, is an associate editor for social sciences of Twin Research and Human Genetics, and publishes in both social science and medical journals.
Acknowledgments
Throughout the three years in which