Combining theoretical, methodological, and practical aspects, Latent Class Analysis of Survey Error successfully guides readers through the accurate interpretation of survey results for quality evaluation and improvement. This book is a comprehensive resource on the key statistical tools and techniques employed during the modeling and estimation of classification errors, featuring a special focus on both latent class analysis (LCA) techniques and models for categorical data from complex sample surveys. Drawing from his extensive experience in the field of survey methodology, the author examines early models for survey measurement error and identifies their similarities and differences as well as their strengths and weaknesses.
Subsequent chapters treat topics related to modeling, estimating, and reducing errors in surveys, including: * Measurement error modeling forcategorical data * The Hui-Walter model and othermethods for two indicators * The EM algorithm and its role in latentclass model parameter estimation * Latent class models for three ormore indicators * Techniques for interpretation of modelparameter estimates * Advanced topics in LCA, including sparse data, boundary values, unidentifiability, and local maxima * Special considerations for analyzing datafrom clustered and unequal probability samples with nonresponse * The current state of LCA and MLCA (multilevel latent class analysis), and an insightful discussion on areas for further research Throughout the book, more than 100 real-world examples describe the presented methods in detail, and readers are guided through the use of lEM software to replicate the presented analyses. Appendices supply a primer on categorical data analysis, and a related Web site houses the lEM software.
Extensively class-tested to ensure an accessible presentation, Latent Class Analysis of Survey Error is an excellent book for courses on measurement error and survey methodology at the graduate level. The book also serves as a valuable reference for researchers and practitioners working in business, government, and the social sciences who develop, implement, or evaluate surveys.
Paul P. Biemer, PhD, is Distinguished Fellow in Statistics at RTI International and Associate Director for Survey Research and Development at the Odum Institute for Research in Social Science at the University of North Carolina at Chapel Hill. An expert in the field of survey measurement error, Dr. Biemer has published extensively in his areas of research interest, which include survey design and analysis; general survey methodology; and nonsampling error modeling and evaluation. He is a coauthor of Introduction to Survey Quality and a coeditor of Telephone Survey Methodology, Survey Measurement and Process Quality, and Measurement Errors in Surveys, all published by Wiley.
Preface. Abbreviations. 1. Survey Error Evaluation. 1.1 Survey Error. 1.1.1 An Overview of Surveys. 1.1.2 Survey Quality and Accuracy and Total Survey Error. 1.1.3 Nonsampling Error. 1.2 Evaluating the Mean-Squared Error. 1.2.1 Purposes of MSE Evaluation. 1.2.2 Effects of Nonsampling Errors on Analysis. 1.2.3 Survey Error Evaluation Methods. 1.2.4 Latent Class Analysis. 1.3 About This Book. 2. A General Model for Measurement Error. 2.1 The Response Distribution. 2.1.1 A Simple Model of the Response Process. 2.1.2 The Reliability Ratio. 2.1.3 Effects of Response Variance on Statistical Inference. 2.2 Variance Estimation in the Presence of Measurement Error. 2.2.1 Binary Response Variables. 2.2.2 Special Case: Two Measurements. 2.2.3 Extension to Polytomous Response Variables. 2.3 Repeated Measurements. 2.3.1 Designs for Parallel Measurements. 2.3.2 Nonparallel Measurements. 2.3.3 Example: Reliability of Marijuana Use Questions. 2.3.4 Designs Based on a Subsample. 2.4 Reliability of Multiitem Scales. 2.4.1 Scale Score Measures. 2.4.2 Cronbach s Alpha. 2.5 True Values, Bias, and Validity. 2.5.1 A True Value Model. 2.5.2 Obtaining True Values. 2.5.3 Example: Poor- or Failing-Grade Data. 3. Response Probability Models for Two Measurements. 3.1 Response Probability Model. 3.1.1 Bross Model. 3.1.2 Implications for Survey Quality Investigations. 3.2 Estimating , , and . 3.2.1 Maximum-Likelihood Estimates of , , and . 3.2.2 The EM Algorithm for Two Measurements. 3.3 Hui Walter Model for Two Dichotomous Measurements. 3.3.1 Notation and Assumptions. 3.3.2 Example: Labor Force Misclassifi cations. 3.3.3 Example: Mode of Data Collection Bias. 3.4 Further Aspects of the Hui Walter Model. 3.4.1 Two Polytomous Measurements. 3.4.2 Example: Misclassifi cation with Three Categories. 3.4.3 Sensitivity of the Hui Walter Method to Violations in the Underlying Assumptions. 3.4.4 Hui Walter Estimates of Reliability. 3.5 Three or More Polytomous Measurements. 4. Latent Class Models for Evaluating Classifi cation Errors. 4.1 The Standard Latent Class Model. 4.1.1 Latent Variable Models. 4.1.2 An Example from Typology Analysis. 4.1.3 Latent Class Analysis Software. 4.2 Latent Class Modeling Basics. 4.2.1 Model Assumptions. 4.2.2 Probability Model Parameterization of the Standard LC Model. 4.2.3 Estimation of the LC Model Parameters. 4.2.4 Loglinear Model Parameterization. 4.2.5 Example: Computing Probabilities Using Loglinear Parameters. 4.2.6 Modifi ed Path Model Parameterization. 4.2.7 Recruitment Probabilities. 4.2.8 Example: Computing Probabilities Using Modified Path Model Parameters. 4.3 Incorporating Grouping Variables. 4.3.1 Example: Loglinear Parameterization of the Hui Walter Model. 4.3.2 Example: Analysis of Past-Year Marijuana Use with Grouping Variables. 4.4 Model Estimation and Evaluation. 4.4.1 EM Algorithm for the LL Parameterization. 4.4.2 Assessing Model Fit. 4.4.3 Model Selection. 4.4.4 Model-Building Strategies. 4.4.5 Model Restrictions. 4.4.6 Example: Continuation of Marijuana Use Analysis. 5. Further Aspects of Latent Class Modeling. 5.1 Parameter Estimation. 5.1.1 Simulation and "Expeculation". 5.1.2 Model Identifi ability. 5.1.3 Checking Identifi ability with Expeculation. 5.1.4 Data Sparseness. 5.1.5 Boundary Estimates. 5.1.6 Local Maxima. 5.1.7 Latent Class Flippage. 5.2 Local Dependence Models. 5.2.1 Unexplained Heterogeneity. 5.2.2 Correlated Errors. 5.2.3 Bivocality. 5.2.4 A Strategy for Modeling Local Dependence. 5.2.5 Example: Locally Dependent Measures of Sexual Assault. 5.3 Modeling Complex Survey Data. 5.3.1 Objectives of Survey Weighting. 5.3.2 LCA with Complex Survey Data. 5.3.3 Including Design Variables in the Fitted Model. 5.3.4 Weighted and Rescaled Frequencies. 5.3.5 Pseudo-Maximum-Likelihood Estimation. 5.3.6 Treating the Average Cell Weight as an Offset Parameter. 5.3.7 Two-Step Estimation. 5.3.8 Illustration of Weighted and Unweighted Analyses. 5.3.9 Conclusions and Recommendations. 6. Latent Class Models for Special Applications. 6.1 Models for Ordinal Data. 6.2 A Latent Class Model for Reliability. 6.2.1 Generalized Kappa Statistics. 6.2.2 Comparison of Error Model and Agreement Model Concepts of Reliability. 6.2.3 Reliability of Self-Reports of Race. 6.3 Capture Recapture Models. 6.3.1 Latent Class Capture Recapture Models. 6.3.2 Modeling Erroneous Enumerations. 6.3.3 Parameter Estimation. 6.3.4 Example: Evaluating the Census Undercount. 6.3.5 Example: Classifi cation Error in a PES. 7. Latent Class Models for Panel Data. 7.1 Markov Latent Class Models. 7.1.1 Manifest Markov Models. 7.1.2 Example: Application of the MM Model to Labor Force Data. 7.1.3 Markov Latent Class Models. 7.1.4 Example: Application of the MLC Model to Labor Force Data. 7.1.5 The EM Algorithm for MLC Models. 7.1.6 MLC Model with Grouping Variables. 7.1.7 Example: CPS Labor Force Status Classification Error. 7.1.8 Example: Underreporting in Consumer Expenditure Survey. 7.2 Some Nonstandard Markov Models. 7.2.1 Manifest Mover Stayer Model. 7.2.2 Latent Class Mover Stayer Model. 7.2.3 Second-Order MLC Model. 7.2.4 Example: CEIS Analysis with Four Timepoints. 7.2.5 MLC Model with Time-Varying Grouping Variables. 7.2.6 Example: Assessment of Subject Interests. 7.2.7 Multiple Indicators at One or More Waves. 7.3 Further Aspects of Markov Latent Class Analysis. 7.3.1 Estimation Issues with MLCA. 7.3.2 Methods for Panel Nonresponse. 7.3.3 Example: Assessment of Subject Interests with Nonresponse. 8. Survey Error Evaluation: Past, Present, and Future. 8.1 History of Survey Error Evaluation Methodology. 8.1.1 The US Census Bureau Model for Survey Error. 8.1.2 From Bross Model to the Standard LC and MLC Models. 8.1.3 Loglinear Models with Latent Variables. 8.2 Current State of the Art. 8.2.1 Criticisms of LCA for Survey Error Evaluation. 8.2.2 General Strategy for Applying LC and MLC Models. 8.3 Some Ideas for Future Directions. 8.4 Conclusions. Appendix A. Two-Stage Sampling Formulas. Appendix B. Loglinear Modeling Essentials. B.1 Loglinear versus ANOVA Models: Similarities and Differences. B.2 Modeling Cell and Other Conditional Probabilities. B.3 Generalization to Three Variables. B.4 Estimation of Loglinear and Logit Models. References. Index.