Flexible Imputation of Missing Data, 2nd ed.

By Shu Yang. Journal of the American Statistical Association, 2019

527 citations recorded by the source corpus.

Missing data are frequently encountered in practice. A broader class of missing data is called incomplete data, which includes data with measurement error, multilevel data with latent variables, and potential outcomes in causal inference. Due to its intuitive appeal, multiple imputation has been the most popular method for handling incomplete data: it multiply fills in the missing values and pools analyses based on straightforward rules. The book focuses mainly on the multiple imputation approach to incomplete data but not on alternative approaches, such as weighting procedures and likelihood-based approaches. The main objective of the book is to provide a tool kit for practitioners to execute multiple imputation. The first edition of this book (van Buuren 2012) has been popular and wellreceived in the statistics and applied research communities. The main text explains the basic and key ideas underpinning multiple imputation, how to implement it in practice, and how to report and interpret the results. Moreover, it uses a readerfriendly style with lots of worked-out examples for illustration based on the MICE package. Therefore, the book can also be viewed as an extended MICE tutorial. Drawing from the author’s own work and from the most recent developments in the field, the new edition expands discussions on important topics or incorporates new topics. This edition features new chapters on multiple imputation of multilevel data, causal inference by multiple imputation of the potential outcomes, and new dataadaptive algorithms for imputation such as predictive mean matching, trees, and many other machine learning tools. The aspect I like most about this book is that it is well-balanced between practicality and technicality. On the one hand, it avoids too much mathematical and technical details and uses graphical tools and visual displays to aid understanding. Practitioners are able to understand the big picture and follow the code provided in the book. On the other hand, theoreticians are able to find technical materials and references to journal articles for in-depth investigation. Divided into three parts, this book begins by an introduction of useful background material and an overview. Chapter 2 reviews the history of multiple imputation and introduces the general taxonomy of missing data mechanisms to set the stage for the whole book. Chapters 3 and 4 cover imputation methods for univariate missing data and multivariate missing data, respectively. Chapter 5 reviews post-imputation statistical analyses of the multiply imputed data. Dedicated statistical tests and model selection techniques are also included. The second part covers various advances of multiple imputation. Chapter 6 covers practical issues such as model specification and diagnosis for multivariate missing data. Chapter 7 is a new chapter discussing multiple imputation for nested data accounting for the multilevel data structure. Chapter 8 is also a new chapter, discussing multiple imputation for causal inference of individualized treatment effects. Chapters 9–11 use case studies to illustrate applications of multiple imputation with item non-response, measurement error, and drop out. The last part discusses limitations, reporting steps, and various extensions. Multiple imputation was originated for survey data, which often contain design weights (or sample weights) to account for sample selection. However, there is no coverage of such topics in the current book. It would be useful to include a new chapter devoted to multiple imputation with survey data. There are many important topics that are worth discussion, such as how to incorporate design weights in the imputation model, weighted or unweighted estimation, and valid inference procedures. In summary, the author has succeeded in achieving his objective of providing a practical introduction to multiple imputation for incomplete data. The book is particularly appropriate for substantive researchers and graduate students. For educators, the book would also make a reference for a course on the introduction and practical implementation of multiple imputation. Anyone who often utilizes complete-case analyses for handling missing data will find this book a good alternative to solve their problems.

Subfields: Artificial Intelligence, Statistics and Probability

View Paper: https://www.semanticscholar.org/paper/11daacd80d2d23d02dd5e3fc7bcc27fc92d66483

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Statistics and Probability

Citations in the corpus

A Comparison of Undersampling, Oversampling, and SMOTE Methods for Dealing with Imbalanced Classification in Educational Data Mining (2023, methodology): “Next, variables with more than a 30% rate of missing data were dropped, and the rest was imputed with multiple imputations by chained equations [41]; this step dropped 16 variables.”
Maximum Likelihood Multiple Imputation: Faster Imputations and Consistent Standard Errors Without Posterior Draws (2012, unknown): “A faster and stabler way to get PD estimates is to boot-strap the incomplete data and calculate a ML estimate from each bootstrapped sample [15, 22, 31].”
The Role of Personality in COVID-19-Related Perceptions, Evaluations, and Behaviors: Findings Across Five Samples, Nine Traits, and 17 Criteria (2021, methodology): “We assumed missing-at-random and consequently conducted (partly passive; Van Buuren, 2018) multiple imputation analyses with 20 imputations using predictive mean matching for continuous and logistic regression for categorical outcomes with the MICE package (Van Buuren & Groothuis-Oudshoorn, 2011), i”