Multilevel and Longitudinal Modeling with IBM SPSS

462 Pages·2013·18.588 MB·English

by  Ronald H. Heck, Scott L. Thomas, Lynn N. Tabata

Multilevel and
Longitudinal
Modeling with
IBM SPSS
Th is book demonstrates how to use multilevel- and longitudinal-modeling techniques avail-
able in the IBM SPSS mixed-eff ects program (MIXED). Annotated screen shots provide read-
ers with a step-by-step understanding of each technique and navigating the program. Readers
learn how to set up, run, and interpret a variety of models. Diagnostic tools, data management
issues, and related graphics are introduced throughout. Annotated syntax is also available for
those who prefer this approach. Extended examples illustrate the logic of model development
to show readers the rationale of the research questions and the steps around which the analyses
are structured. Th e data used in the text and syntax examples are available at www.routledge.
com/9780415817110.
Highlights of the new edition include the following:
• Updated throughout to refl ect IBM SPSS Version 21.
• Further coverage of growth trajectories, coding time-related variables, covariance structures,
individual change, and longitudinal experimental designs.
• Extended discussion of other types of research designs for examining change (e.g., regression
discontinuity, quasi-experimental) over time.
• New examples specifying multiple latent constructs and parallel growth processes.
• Discussion of alternatives for dealing with missing data and the use of sample weights within
multilevel data structures.
Th e book opens with the conceptual and methodological issues associated with multilevel and
longitudinal modeling, followed by a discussion of SPSS data management techniques that fa-
cilitate working with multilevel, longitudinal, and cross-classifi ed data sets. Chapters 3 and 4 in-
troduce the basics of multilevel modeling: developing a multilevel model, interpreting output,
and trouble-shooting common programming and modeling problems. Models for investigating
individual and organizational change are presented in Chapters 5 and 6, followed by models with
multivariate outcomes in Chapter 7. Chapter 8 provides an illustration of multilevel models with
cross-classifi ed data structures. Th e book concludes with ways to expand on the various multilevel-
and longitudinal-modeling techniques and issues when conducting multilevel analyses.
Ideal as a supplementary text for graduate courses on multilevel and longitudinal modeling,
multivariate statistics, and research design taught in education, psychology, business, and sociol-
ogy, this book’s practical approach also appeals to researchers in these fi elds. Th e book provides
an excellent supplement to Heck and Th omas’s A n Introduction to Multilevel Modeling Techniques
(2nd ed.); however, it can also be used with any multilevel- and/or longitudinal-modeling book
or as a stand-alone text.
Ronald H. Heck is professor of education at the University of Hawai‘i at Mānoa. His areas of
interest include organizational theory, leadership, policy, and quantitative research methods.
Scott L. Th omas is professor and dean of the School of Educational Studies at Claremont Grad-
uate University. His specialties include sociology of education, policy, and quantitative research
methods.
Lynn N. Tabata is an affi liate graduate faculty member and research consultant at the University
of Hawai‘i at Mānoa. Her research interests focus on faculty, distance learning, and technology
issues in higher education.
Quantitative Methodology Series
George A. Marcoulides, Series Editor
This series presents methodological techniques to investigators and students. Th e goal is to
provide an understanding and working knowledge of each method with a minimum of
mathematical derivations. Each volume focuses on a specifi c method (e.g. Factor Analysis, Mul-
tilevel Analysis, Structural Equation Modeling).
Proposals are invited from interested authors. Each proposal should consist of: a brief de-
scription of the volume’s focus and intended market; a table of contents with an outline of each
chapter; and a curriculum vita. Materials may be sent to Dr. George A. Marcoulides, University
of California – Riverside, [e-mail].
Marcoulides • Modern Methods for Business Research
Marcoulides/Moustaki • Latent Variable and Latent Structure Models
Heck • Studying Educational and Social Policy: Th eoretical Concepts and Research Methods
Van der Ark/Croon/Sijtsma • New Developments in Categorical Data Analysis for the Social
and Behavioral Sciences
Duncan/Duncan/Strycker • An Introduction to Latent Variable Growth Curve Modeling:
Concepts, Issues, and Applications, Second Edition
Heck/Th omas • An Introduction to Multilevel Modeling Techniques, Second Edition
Cardinet/Johnson/Pini • Applying Generalizability Th eory Using EduG
Creemers/Kyriakides/Sammons • Methodological Advances in Educational Eff ectiveness
Research
Hox • Multilevel Analysis: Techniques and Applications, Second Edition
Heck/Th omas/Tabata • Multilevel Modeling of Categorical Outcomes Using IBM SPSS
Heck/Th omas/Tabata • Multilevel and Longitudinal Modeling with IBM SPSS, Second
Edition
McArdle/Ritschard • Contemporary Issues in Exploratory Data Mining in the Behavioral
Sciences
This page intentionally left blank
Multilevel and
Longitudinal
Modeling with
IBM SPSS
Second Edition
Ronald H. Heck
-
University of Hawai‘i at Manoa
Scott L. Thomas
Claremont Graduate University
Lynn N. Tabata
-
University of Hawai‘i at Manoa
Please visit the eResources site at
www.routledge.com/9780415817110
First published 2014
by Routledge
711 Th ird Avenue, New York, NY 10017
Simultaneously published in the UK
by Routledge
27 Church Road, Hove, East Sussex BN3 2FA
Routledge is an imprint of the Taylor & Francis Group, an informa business
© 2014 Taylor & Francis
Th e right of Ronald H. Heck, Scott L. Th omas, and Lynn N. Tabata to be identifi ed as authors of this work has been
asserted by them in accordance with sections 77 and 78 of the Copyright, Designs and Patents Act 1988.
All rights reserved. No part of this book may be reprinted or reproduced or utilised in any form or by any electronic,
mechanical, or other means, now known or hereafter invented, including photocopying and recording, or in any
information storage or retrieval system, without permission in writing from the publishers.
Trademark notice : Product or corporate names may be trademarks or registered trademarks, and are used only for
identifi cation and explanation without intent to infringe.
Reprinted IBM SPSS Screenshots Courtesy of International Business Machines Corporation, © SPSS, Inc., an
IBM Company*
Library of Congress Cataloging in Publication Data
Heck, Ronald H.
Multilevel and longitudinal modeling with IBM SPSS / Ronald H. Heck, Th e University of Hawaii at Manoa, Scott
L. Th omas, Claremont Graduate University, Lynn N. Tabata, Th e University of Hawaii at Manoa.—Second edition.
pages cm. — (Quantitave methodology series)
Includes bibliographical references and index.
1. Social sciences—Longitudinal studies. 2. Social sciences—Statistical methods. 3. PASW (Computer fi le)
4. SPSS (Computer fi le) I. Th omas, Scott L. II. Tabata, Lynn Naomi. III. Title.
HA32.H39 2013
005.5'5—dc23 2013004161
ISBN: 978-0-415-81710-3 (hbk)
ISBN: 978-0-415-81711-0 (pbk)
ISBN: 978-0-203-70124-9 (ebk)
Typeset in ACaslon Pro
by Apex CoVantage, LLC
*SPSS was acquired by IBM in October 2009.
Contents
Preface xv
Chapter 1 Introduction to Multilevel Modeling with IBM SPSS 1
Our Intent 2
Overview of Topics 4
Analysis of Multilevel Data Structures 4
Partitioning Variation in an Outcome 8
Developing a General Multilevel-Modeling Strategy 9
Illustrating the Steps in Investigating a Proposed Model 10
1. One-Way ANOVA (No Predictors) Model 11
2. Analyze a Level 1 Model with Fixed Predictors 12
3. Add the Level 2 Explanatory Variables 13
4. Examine Whether a Particular Slope Coeffi cient Varies Between Groups 14
5. Adding Cross-Level Interactions to Explain Variation in the Slope 15
Syntax Versus IBM SPSS Menu Command Formulation 16
Model Estimation and Other Typical Multilevel-Modeling Issues 18
Sample Size 20
Power 20
Diff erences Between Multilevel Software Programs 21
Standardized and Unstandardized Coeffi cients 21
Missing Data 22
Missing Data at Level 2 26
Missing Data in Vertical Format in IBM SPSS MIXED 28
Design Eff ects, Sample Weights, and the Complex Samples Routine
in IBM SPSS 30
An Example Using Multilevel Weights 32
Summary 34
Chapter 2 Preparing and Examining the Data for Multilevel Analyses 35
Data Requirements 35
File Layout 36
Getting Familiar with Basic IBM SPSS Data Commands 38
Recode: Creating a New Variable Th rough Recoding 39
Recoding Old Values to New Values 39
Recoding Old Values to New Values Using “Range” 41
Compute: Creating a New Variable Th at Is a Function of Some Other Variable 44
Match Files: Combining Data From Separate IBM SPSS Files 46
Aggregate: Collapsing Data Within Level 2 Units 52
VARSTOCASES: Vertical Versus Horizontal Data Structures 53
Using “Compute” and “Rank” to Recode the Level 1 or Level 2 Data
for Nested Models 59
Creating an Identifi er Variable 59
Creating an Individual-Level Identifi er Using “Compute” 60
Creating a Group-Level Identifi er Using “Rank Cases” 61
Creating a Within-Group-Level Identifi er Using “Rank Cases” 63
Centering 65
vii
viii (cid:2) Contents
Grand-Mean Centering 67
Group-Mean Centering 69
Checking the Data 72
A Note About Model Building 73
Summary 73
Chapter 3 Defi ning a Basic Two-Level Multilevel Regression Model 75
From Single-Level to Multilevel Analysis 75
Building a Two-Level Model 77
Research Questions 78
Th e Data 78
Specifying the Model 78
Graphing the Relationship Between SES and Math Test Scores
with IBM SPSS Menu Commands 80
Graphing the Subgroup Relationships Between SES and Math Test Scores
with IBM SPSS Menu Commands 86
Building a Multilevel Model with IBM SPSS MIXED 88
Step 1: Examining Variance Components Using the Null Model 89
Defi ning Model 1 (Null) with IBM SPSS Menu Commands 90
Interpreting the Output From Model 1 (Null) 93
Step 2: Building the Individual-Level (or Level 1) Random Intercept Model 95
Defi ning Model 2 with IBM SPSS Menu Commands 96
Interpreting the Output From Model 2 98
Step 3: Building the Group-Level (or Level 2) Random Intercept Model 101
Defi ning Model 3 with IBM SPSS Menu Commands 102
Interpreting the Output From Model 3 104
Defi ning Model 3A ( Public as Covariate) with IBM SPSS
Menu Commands 108
Step 4: Adding a Randomly Varying Slope (the Random Slope and
Intercept Model) 110
Defi ning Model 4 with IBM SPSS Menu Commands 111
Interpreting the Output From Model 4 113
Step 5: Explaining Variability in the Random Slope (More Complex
Random Slopes and Intercept Models) 115
Defi ning Model 5 with IBM SPSS Menu Commands 116
Add First Interaction to Model 5: ses_mean*ses 118
Add Second Interaction to Model 5: p ro4yrc*ses 118
Add Th ird Interaction to Model 5: p ublic*ses 118
Interpreting the Output From Model 5 119
Defi ning Model 5A with IBM SPSS Menu Commands 121
Graphing a Cross-Level Interaction (SES-Achievement Relationships
in High- and Low-Achieving Schools) with IBM SPSS Menu Commands 123
Centering Predictors 126
Centering Predictors in Models with Random Slopes 128
Summary 130
Chapter 4 Th ree-Level Univariate Regression Models 131
Th ree-Level Univariate Model 131
Research Questions 131
Th e Data 132
Contents (cid:2) ix
Defi ning the Th ree-Level Multilevel Model 133
Th e Null Model (No Predictors) 134
Defi ning Model 1 (Null) with IBM SPSS Menu Commands 135
Interpreting the Output From Model 1 (Null) 138
Model 2: Defi ning Predictors at Each Level 139
Defi ning Model 2 with IBM SPSS Menu Commands 142
Interpreting the Output From Model 2 144
Model 3: Group-Mean Centering 145
Defi ning Model 3 with IBM SPSS Menu Commands 145
Interpreting the Output From Model 3 147
Covariance Estimates 148
Model 4: Does the Slope Vary Randomly Across Schools? 149
Defi ning Model 4 with IBM SPSS Menu Commands 150
Interpreting the Output From Model 4 153
Developing an Interaction Term 154
Preliminary Investigation of the Interaction 155
Defi ning Models A and B (Preliminary Testing of Interactions)
with IBM SPSS Menu Commands 156
Model A Test Interaction: t eacheff ect*classlowses_mean 158
Model B Test Interaction: g mteacheff ect*gmclasslowses_mean 159
Model 5: Examining a Level 2 Interaction 161
Defi ning Model 5 with IBM SPSS Menu Commands 161
Add Interaction to Model 5: g mclasslowses_mean*gmteacheff ect 163
Interpreting the Output From Model 5 163
Comparing the Fit of Successive Models 164
Summary 166
Chapter 5 Examining Individual Change with Repeated Measures Data 167
Ways to Examine Repeated Observations on Individuals 167
Considerations in Specifying a Linear Mixed Model 168
An Example Study 171
Research Questions 171
Th e Data 171
Examining the Shape of Students’ Growth Trajectories 173
Graphing the Linear and Nonlinear Growth Trajectories
with IBM SPSS Menu Commands 175
Select Subset of Individuals 176
Generate Figure 5.3 (Linear Trajectory) 178
Generate Figure 5.4 (Nonlinear Quadratic Trajectory) 180
Coding the Time-Related Variables 181
Coding Time Interval Variables ( time to quadtime )
with IBM SPSS Menu Commands 182
Coding Time Interval Variables ( time to orthtime , orthquad )
with IBM SPSS Menu Commands 184
Specifying the Two-Level Model of Individual Change 186
Level 1 Covariance Structure 188
Repeated Covariance Dialog Box 188
Model 1.1: Model with No Predictors 191
Defi ning Model 1.1 (Null) with IBM SPSS Menu Commands 192
Interpreting the Output From Model 1.1 (Null) 195
Model 1.1A: What Is the Shape of the Trajectory? 196