| |
| |
Preface | |
| |
| |
Data acknowledgements | |
| |
| |
Acknowledgements | |
| |
| |
Glossary | |
| |
| |
| |
Foundations | |
| |
| |
| |
Introduction | |
| |
| |
| |
Reasons for missing data | |
| |
| |
| |
Examples | |
| |
| |
| |
Patterns of missing data | |
| |
| |
| |
Consequences of missing data | |
| |
| |
| |
Inferential framework and notation | |
| |
| |
| |
Missing Completely At Random (MCAR) | |
| |
| |
| |
Missing At Random (MAR) | |
| |
| |
| |
Missing Not At Random (MNAR) | |
| |
| |
| |
Ignorability | |
| |
| |
| |
Using observed data to inform assumptions about the missingness mechanism | |
| |
| |
| |
Implications of missing data mechanisms for regression analyses | |
| |
| |
| |
Partially observed response | |
| |
| |
| |
Missing covariates | |
| |
| |
| |
Missing covariates and response | |
| |
| |
| |
Subtle issues I: The odds ratio | |
| |
| |
| |
Implication for linear regression | |
| |
| |
| |
Subtle issues II: Subsample ignorability | |
| |
| |
| |
Summary: When restricting to complete records is valid | |
| |
| |
| |
Summary | |
| |
| |
| |
The multiple imputation procedure and its justification | |
| |
| |
| |
Introduction | |
| |
| |
| |
Intuitive outline of the MI procedure | |
| |
| |
| |
The generic MI procedure | |
| |
| |
| |
Bayesian justification of MI | |
| |
| |
| |
Frequentist inference | |
| |
| |
| |
Large number of imputations | |
| |
| |
| |
Small number of imputations | |
| |
| |
| |
Choosing the number of imputations | |
| |
| |
| |
Some simple examples | |
| |
| |
| |
MI in more general settings | |
| |
| |
| |
Survey sample settings | |
| |
| |
| |
Constructing congenial imputation models | |
| |
| |
| |
Practical considerations for choosing imputation models | |
| |
| |
| |
Discussion | |
| |
| |
| |
Multiple Imputation for Cross Sectional Data | |
| |
| |
| |
Multiple imputation of quantitative data | |
| |
| |
| |
Regression imputation with a monotone missingness pattern | |
| |
| |
| |
MAR mechanisms consistent with a monotone pattern | |
| |
| |
| |
Justification | |
| |
| |
| |
Joint modelling | |
| |
| |
| |
Fitting the imputation model | |
| |
| |
| |
Full conditional specification | |
| |
| |
| |
Justification | |
| |
| |
| |
Full conditional specification versus joint modelling | |
| |
| |
| |
Software for multivariate normal imputation | |
| |
| |
| |
Discussion | |
| |
| |
| |
Multiple imputation of binary and ordinal data | |
| |
| |
| |
Sequential imputation with monotone missingness pattern | |
| |
| |
| |
Joint modelling with the multivariate normal distribution | |
| |
| |
| |
Modelling binary data using latent normal variables | |
| |
| |
| |
Latent normal model for ordinal data | |
| |
| |
| |
General location model | |
| |
| |
| |
Full conditional specification | |
| |
| |
| |
Justification | |
| |
| |
| |
Issues with over-fitting | |
| |
| |
| |
Pros and cons of the various approaches | |
| |
| |
| |
Software | |
| |
| |
| |
Discussion | |
| |
| |
| |
Multiple imputation of unordered categorical data | |
| |
| |
| |
Monotone missing data | |
| |
| |
| |
Multivariate normal imputation for categorical data | |
| |
| |
| |
Maximum indicant model | |
| |
| |
| |
Continuous and categorical variable | |
| |
| |
| |
Imputing missing data | |
| |
| |
| |
More than one categorical variable | |
| |
| |
| |
General location model | |
| |
| |
| |
FCS with categorical data | |
| |
| |
| |
Perfect prediction issues with categorical data | |
| |
| |
| |
Software | |
| |
| |
| |
Discussion | |
| |
| |
| |
Nonlinear relationships | |
| |
| |
| |
Passive imputation | |
| |
| |
| |
No missing data in nonlinear relationships | |
| |
| |
| |
Missing data in nonlinear relationships | |
| |
| |
| |
Predictive Mean Matching (PMM) | |
| |
| |
| |
Just Another Variable (JAV) | |
| |
| |
| |
Joint modelling approach | |
| |
| |
| |
Extension to more general models and missing data patterns | |
| |
| |
| |
Metropolis-Hastings sampling | |
| |
| |
| |
Rejection sampling | |
| |
| |
| |
FCS approach | |
| |
| |
| |
Discussion | |
| |
| |
| |
Interactions | |
| |
| |
| |
Interaction variables rally observed | |
| |
| |
| |
Interactions of categorical variables | |
| |
| |
| |
General nonlinear relationships | |
| |
| |
| |
Software | |
| |
| |
| |
Discussion | |
| |
| |
| |
Advanced Topics | |
| |
| |
| |
Survival data, skips and large datasets | |
| |
| |
| |
Time-to-event data | |
| |
| |
| |
Imputing missing covariate values | |
| |
| |
| |
Survival data as categorical | |
| |
| |
| |
Imputing censored survival times | |
| |
| |
| |
Nonparametric, or 'hot deck' imputation | |
| |
| |
| |
Nonparametric imputation for survival data | |
| |
| |
| |
Multiple imputation for skips | |
| |
| |
| |
Two-stage MI | |
| |
| |
| |
Large datasets | |
| |
| |
| |
Large datasets and joint modelling | |
| |
| |
| |
Shrinkage by constraining parameters | |
| |
| |
| |
Comparison of the two approaches | |
| |
| |
| |
Multiple imputation and record linkage | |
| |
| |
| |
Measurement error | |
| |
| |
| |
Multiple imputation for aggregated scores | |
| |
| |
| |
Discussion | |
| |
| |
| |
Multilevel multiple imputation | |
| |
| |
| |
Multilevel imputation model | |
| |
| |
| |
MCMC algorithm for imputation model | |
| |
| |
| |
Imputing level-2 covariates using FCS | |
| |
| |
| |
Individual patient meta-analysis | |
| |
| |
| |
When to apply Rubin's rules | |
| |
| |
| |
Extensions | |
| |
| |
| |
Random level-1 covariance matrices | |
| |
| |
| |
Model fit | |
| |
| |
| |
Discussion | |
| |
| |
| |
Sensitivity analysis: MI unleashed | |
| |
| |
| |
Review of MNAR modelling | |
| |
| |
| |
Framing sensitivity analysis | |
| |
| |
| |
Pattern mixture modelling with MI | |
| |
| |
| |
Missing covariates | |
| |
| |
| |
Application to survival analysis | |
| |
| |
| |
Pattern mixture approach with longitudinal data via MI | |
| |
| |
| |
Change in slope post-deviation | |
| |
| |
| |
Piecing together post-deviation distributions from other trial arms | |
| |
| |
| |
Approximating a selection model by importance weighting | |
| |
| |
| |
Algorithm for approximate sensitivity analysis by re-weighting | |
| |
| |
| |
Discussion | |
| |
| |
| |
Including survey weights | |
| |
| |
| |
Using model based predictions | |
| |
| |
| |
Bias in the MI variance estimator | |
| |
| |
| |
MI with weights | |
| |
| |
| |
Estimation in domains | |
| |
| |
| |
A multilevel approach | |
| |
| |
| |
Further developments | |
| |
| |
| |
Discussion | |
| |
| |
| |
Robust multiple imputation | |
| |
| |
| |
Introduction | |
| |
| |
| |
Theoretical background | |
| |
| |
| |
Simple estimating equations | |
| |
| |
| |
The Probability Of Missingness (POM) model | |
| |
| |
| |
Augmented inverse probability weighted estimating equation | |
| |
| |
| |
Robust multiple imputation | |
| |
| |
| |
Univariate MAR missing data | |
| |
| |
| |
Longitudinal MAR missing data | |
| |
| |
| |
Simulation studies | |
| |
| |
| |
Univariate MAR missing data | |
| |
| |
| |
Longitudinal monotone MAR missing data | |
| |
| |
| |
Longitudinal nonmonotone MAR missing data | |
| |
| |
| |
Nonlongitudinal nonmonotone MAR missing data | |
| |
| |
| |
Results and discussion | |
| |
| |
| |
The RECORD study | |
| |
| |
| |
Discussion | |
| |
| |
| |
Markov Chain Monte Carlo | |
| |
| |
| |
Probability distributions | |
| |
| |
| |
Posterior for the multivariate normal distribution | |
| |
| |
Bibliography | |
| |
| |
Index of Authors | |
| |
| |
Index of Examples | |
| |
| |
Index | |