Foreword
I’m delighted to see this new book on multiple imputation by Stef van Buuren for several reasons. First, to me at least, having another book devoted to multiple imputation marks the maturing of the topic after an admittedly somewhat shaky initiation. Stef is certainly correct when he states in Section 2.1.2: “The idea to create multiple versions must have seemed outrageous at that time (late 1970’s). Drawing imputations from a distribution, instead of estimating the ‘best’ value, was a severe breach with everything that had been done before.” I remember how this idea of multiple imputation was even ridiculed by some more traditional statisticians, sometimes for just being “silly” and sometimes for being hopelessly inefficient with respect to storage demands and outrageously expensive with respect to computational requirements.
Some others of us foresaw what was happening to both (a) computational storage (I just acquired a 64 GB flash drive the size of a small finger for under $60, whereas only a couple of decades ago I paid over $2500 for a 120 KB hard drive larger than a shoebox weighing about 10 kilos), and (b) computational speed and flexibility. To develop statistical methods for the future while being bound by computational limitations of the past was clearly inapposite. Multiple imputation’s early survival was clearly due to the insight of a younger generation of statisticians, including many colleagues and former students, who realized future possibilities.
A second reason for my delight at the publication of this book is more personal and concerns the maturing of the author, Stef van Buuren. As he mentions, we first met through Jan van Rijckevorsel at TNO. Stef was a young and enthusiastic researcher there, who knew little about the kind of statistics that I felt was essential for making progress on the topic of dealing with missing data. But consider the progress over the decades starting with his earlier work on MICE! Stef has matured into an independent researcher making important and original contributions to the continued development of multiple imputation.
This book represents a ‘no nonsense’ straightforward approach to the application of multiple imputation. I particularly like Stef’s use of graphical displays, which are badly needed in practice to supplement the more theoretical discussions of the general validity of multiple imputation methods. As I have said elsewhere, and as implied by much of what is written by Stef, “It’s not that multiple imputation is so good; it’s really that other methods for addressing missing data are so bad.” It’s great to have Stef’s book on multiple imputation, and I look forward to seeing more editions as this rapidly developing methodology continues to become even more effective at handling missing data problems in practice.
Finally, I would like to say that this book reinforces the pride of an academic father who has watched one of his children grow and develop. This book is a step in the growing list of contributions that Stef has made, and, I am confident, will continue to make, in methodology, computational approaches and application of multiple imputation.
I am very pleased for the opportunity to add this short addendum to my preface to the first edition of Stef’s wonderfully readable book on multiple imputation. Over the past few years, I’ve recommended that many check out Stef’s first edition for excellent advice to practitioners of multiple imputation. The increased appreciation and use of multiple imputation between these two editions reflects, not only the growing maturity of computational statistics, but also the growing acceptance of multiple imputation as essentially being “the only game in town” for dealing generally with the problem of missing data, especially because it leads so naturally to visual displays of sensitivity of conclusions to differing assumptions about the reasons for the missing data. I am enthusiastic about the growing role of sensitivity analysis using visual displays, always prominent in Stef’s contributions, but also now in other places, such as Liublinska and Rubin (2014) — some of us can be slow learners!