APPLIED MISSING DATA

ANALYSIS EXAMPLES

ANALYSIS EXAMPLES

The examples from Chapter 1 illustrate steps to prepare for a missing data analysis, including identifying correlates of missingness and effective auxiliary variables. The examples also illustrate simulation-based power analyses for planned missing data.

- Example 1.6 – Identifying Auxiliary Variables
- Example 1.10 – Power Analyses for Planned Missingness Designs

The examples from Chapter 2 illustrate complete-data maximum likelihood estimation for different regression models and multivariate normal data. The files also include custom R scripts that hand-code FIML estimators for multivariate normal data and a single-mediator model.

- Example 2.10 – Multiple Regression
- Example 2.12 – Means, Covariances, and Correlations
- Example 2.13 – Probit and Logistic Regression
- Custom R programs illustrating FIML coding

Chapter 3 illustrates maximum likelihood estimation with incomplete data. The chapter begins with estimation for multivariate normal data and progresses to newer factored regression approaches that disassemble a model into multiple parts that leverage different types of distributions. Factorization paves the way for estimating interactions and nonlinear effects with missing data, and the analysis examples illustrate this approach. The files also include custom R scripts that hand-code FIML estimators for bivariate normal data and a regression model.

- Example 3.2 – Means, Covariances, and Correlations
- Example 3.6 – Multiple Regression
- Example 3.8 – Moderated Regression with an Interaction
- Example 3.9 – Curvilinear Regression
- Example 3.10 – FIML with Auxiliary Variables
- Example 3.11 – Probit and Logistic Regression
- Custom R programs illustrating FIML coding

The examples from Chapter 4 illustrate complete-data Bayesian estimation for linear regression models and multivariate normal data. The files also include custom R scripts that hand-code MCMC estimators for univariate normal data and a regression model.

- Example 4.8 – Multiple Regression
- Example 4.10 – Means, Covariances, and Correlations
- Custom R programs illustrating MCMC coding

Chapter 5 illustrates Bayesian estimation with incomplete data. The chapter begins with newer factored regression approaches that disassemble a model into multiple parts that leverage different types of distributions. Factorization paves the way for estimating interactions and nonlinear effects with missing data, and the analysis examples illustrate this approach. The chapter concludes with Bayesian missing data handling for multivariate normal data. The files also include custom R scripts that hand-code MCMC estimators for a regression model with missing data.

- Example 5.3 – Multiple Regression
- Example 5.4 – Moderated Regression with an Interaction
- Example 5.7 – Curvilinear Regression
- Example 5.8 – Bayes Estimation with Auxiliary Variables
- Example 5.9 – Means, Covariances, and Correlations
- Custom R programs illustrating MCMC coding

Factored regression specifications are especially well-suited for variables with mixed response types, and Chapter 6 describes Bayesian estimation and missing data handling for binary, ordinal, and multicategorical nominal variables. The examples illustrate analyses with categorical predictors and outcomes, and the files also include custom R scripts that hand-code MCMC estimators for binary and ordinal probit models.

- Example 6.3 – Binary Probit Regression
- Example 6.4 – Ordinal Probit Regression
- Example 6.5 – Regression with Binary Predictors
- Example 6.7 – Regression with a Multicategorical Outcome
- Example 6.8 – Regression with Multicategorical Predictors
- Example 6.9 – Binary Logistic Regression
- Custom R programs illustrating MCMC coding

The examples in Chapter 7 describe two predominant and classic multiple imputation frameworks, joint model imputation and fully conditional specification. The chapter also illustrates newer model-based imputation methods based on factored regression specifications, highlighting multipleimputation for a moderated regression analysis with an interaction effect.

- Example 7.3 – Joint Model Multiple Imputation
- Example 7.4 – Fully Conditional Specification Multiple Imputation
- Example 7.11 – Model–Based Imputation for an Interaction Effect

The emergence of missing data handling methods for multilevel models is an important recent development. The examples in Chapter 8 illustrate Bayesian estimation and multiple imputation for two- and three-level regression models with various real-world complexities. The chapter concludes with FIML estimation for a two-level random intercept model.

- Example 8.2 – Regression with Random Intercepts
- Example 8.3 – Regression with Random Slopes
- Example 8.4 – Regression with a Cross-Level Interaction
- Example 8.5 – Three–Level Regression with an Interaction
- Example 8.7a – Joint Model Multiple Imputation
- Example 8.7b – Joint Model Imputation w Random Covariances
- Example 8.8 – Fully Conditional Specification Multiple Imputation
- Example 8.9 – FIML Regression with Random Intercepts

Chapter 9 illustrates two major modeling frameworks for missing not at random processes, selection models and pattern mixture models. Both approaches introduce a model that describes the occurrence of missing data, albeit in different ways. The analysis examples apply these frameworks to regression models and single-level and multilevel longitudinal growth models.

- Example 9.6 – Selection Model for Regression
- Example 9.8 – Pattern Mixture Models for Regression
- Example 9.13a – Diggle–Kenward Selection Growth Model
- Example 9.13b – Shared Parameter Growth Model
- Example 9.13c – Hedeker–Gibbons Pattern Mixture Growth Model

The data analysis examples in Chapter 10 illustrate specialized topics, advanced applications, and practical issues. Examples in this section highlight use cases that differentiate FIML, Bayesian estimation, and multiple imputation.

- Example 10.2 – Descriptives, Correlations, and Subgroups
- Example 10.3 – Non-Normal Predictor Variables
- Example 10.4 – Non-Normal Outcome Variables
- Example 10.5 – Mediation and Indirect Effects
- Example 10.6 – Structural Equation Models
- Example 10.7 – Scale Scores and Missing Questionnaire Items
- Example 10.8 – Interactions with Scales
- Example 10.9 – Longitudinal Data Analyses
- Example 10.10 – Regression with a Count Outcome
- Example 10.11 – Growth Model Power Analyses with Missing Data

Download an archive of all data sets and analysis scripts from the first edition of Applied Missing Data Analysis.

The examples from Chapter 1 illustrate steps to prepare for a missing data analysis, including identifying correlates of missingness and effective auxiliary variables. The examples also illustrate simulation-based power analyses for planned missing data.

- Example 1.6 – Identifying Auxiliary Variables
- Example 1.10 – Power Analyses for Planned Missingness Designs

The examples from Chapter 2 illustrate complete-data maximum likelihood estimation for different regression models and multivariate normal data. The files also include custom R scripts that hand code FIML estimators for multivariate normal data and a single-mediator model.

- Example 2.10 – Multiple Regression
- Example 2.12 – Means, Covariances, and Correlations
- Example 2.13 – Probit and Logistic Regression
- Custom R programs illustrating FIML coding

Chapter 3 illustrates maximum likelihood estimation with incomplete data. The chapter begins with estimation for multivariate normal data and progresses to newer factored regression approaches that disassemble a model into multiple parts that leverage different types of distributions. Factorization paves the way for estimating interactions and nonlinear effects with missing data, and the analysis examples illustrate this approach. A custom R program that hand codes FIML estimation is also available.

- Example 3.2 – Means, Covariances, and Correlations
- Example 3.6 – Multiple Regression
- Example 3.8 – Moderated Regression with an Interaction
- Example 3.9 – Curvilinear Regression
- Example 3.10 – FIML with Auxiliary Variables
- Example 3.11 – Probit and Logistic Regression
- Custom R programs illustrating FIML coding

The examples from Chapter 4 illustrate complete-data Bayesian estimation for linear regression models and multivariate normal data. A custom R program that hand codes MCMC estimation is also available.

- Example 4.8 – Multiple Regression
- Example 4.10 – Means, Covariances, and Correlations

Chapter 5 illustrates Bayesian estimation with incomplete data. The chapter begins with newer factored regression approaches that disassemble a model into multiple parts that leverage different types of distributions. Factorization paves the way for estimating interactions and nonlinear effects with missing data, and the analysis examples illustrate this approach. The chapter concludes with Bayesian missing data handling for multivariate normal data. A custom R program that hand codes MCMC estimation and imputation is also available.

- Example 5.3 – Multiple Regression
- Example 5.4 – Moderated Regression with an Interaction
- Example 5.7 – Curvilinear Regression
- Example 5.8 – Bayes Estimation with Auxiliary Variables
- Example 5.9 – Means, Covariances, and Correlations

Factored regression specifications are especially well-suited for variables with mixed response types, and Chapter 6 describes Bayesian estimation and missing data handling for binary, ordinal, and multicategorical nominal variables. The examples illustrate analyses with categorical predictors and outcomes, and a custom R program is available that hand codes MCMC estimation and imputation is also available.

- Example 6.3 – Binary Probit Regression
- Example 6.4 – Ordinal Probit Regression
- Example 6.5 – Regression with Binary Predictors
- Example 6.7 – Regression with a Multicategorical Outcome
- Example 6.8 – Regression with Multicategorical Predictors
- Example 6.9 – Binary Logistic Regression

The examples in Chapter 7 describe two predominant and classic multiple imputation frameworks, joint model imputation and fully conditional specification. The chapter also illustrates newer model-based imputation methods based on factored regression specifications, highlighting multipleimputation for a moderated regression analysis with an interaction effect.

- Example 7.3 – Joint Model Multiple Imputation
- Example 7.4 – Fully Conditional Specification Multiple Imputation
- Example 7.11 – Model–Based Imputation for an Interaction Effect

The emergence of missing data handling methods for multilevel models is an important recent development. The examples in Chapter 8 illustrate Bayesian estimation and multiple imputation for two- and three-level regression models with various real-world complexities. The chapter concludes with FIML estimation for a two-level random intercept model.

- Example 8.2 – Regression with Random Intercepts
- Example 8.3 – Regression with Random Slopes
- Example 8.4 – Regression with a Cross-Level Interaction
- Example 8.5 – Three–Level Regression with an Interaction
- Example 8.7a – Joint Model Multiple Imputation
- Example 8.7b – Joint Model Imputation w Random Covariances
- Example 8.8 – Fully Conditional Specification Multiple Imputation
- Example 8.9 – FIML Regression with Random Intercepts

Chapter 9 illustrates two major modeling frameworks for missing not at random processes, selection models and pattern mixture models. Both approaches introduce a model that describes the occurrence of missing data, albeit in different ways. The analysis examples apply these frameworks to regression models and single-level and multilevel longitudinal growth models.

- Example 9.6 – Selection Model for Regression
- Example 9.8 – Pattern Mixture Models for Regression
- Example 9.13a – Diggle–Kenward Selection Growth Model
- Example 9.13b – Shared Parameter Growth Model
- Example 9.13c – Hedeker–Gibbons Pattern Mixture Growth Model

The data analysis examples in Chapter 10 illustrate specialized topics, advanced applications, and practical issues. Examples in this section highlight use cases that differentiate FIML, Bayesian estimation, and multiple imputation.

- Example 10.2 – Descriptives, Correlations, and Subgroups
- Example 10.3 – Non-Normal Predictor Variables
- Example 10.4 – Non-Normal Outcome Variables
- Example 10.5 – Mediation and Indirect Effects
- Example 10.6 – Structural Equation Models
- Example 10.7 – Scale Scores and Missing Questionnaire Items
- Example 10.8 – Interactions with Scales
- Example 10.9 – Longitudinal Data Analyses
- Example 10.10 – Regression with a Count Outcome
- Example 10.11 – Power Analyses with Missing Data

Download an archive of all data sets and analysis scripts from the first edition of Applied Missing Data Analysis.