Statistical Analysis With Missing Data

Rating:
Summary: the bible on missing data
Review: Don Rubin developed much of the current theory on missing data. He and Rod Little have written eloquently on an important and difficult topic. This is the best available reference on missing data. Multiple imputation has been a highly successful technique. It was developed by Rubin and it gets good coverage here. If you are particularly interested in multiple imputation Rubin has another text devoted solely to it. The only drawback to the text is that standard software to handle the new methods was not available in 1987. So there is no coverage of software packages. However, Rubin has worked with Statistical Solutions to get imputation and particularly multiple imputation techniques into a software package called Solas. With Rubin's books and the Solas manual you will be ready to do imputation and more importantly you will understand the modeling assumptions that the methods hinge on.

Rating:
Summary: Cautious and applicable
Review: I'm working with data sets where up to 15% of measurements are unusable. If I'm going to get any results at all, I have to get them despite the lost values.

This book provides a huge library of techniques for working around the holes, as well as techniques for filling them in. This is not a cut-and-paste text for programmers - it gives the basic theory and algorithms for each technique. Still, the presentation is quite readable and fairly easy to put into practice.

The book's emphasis is on imputation - filling in values so that analysis can move forward. This is something to approach with real caution, though. The imputed (synthesized) values must not perturb the analysis, so the imputation must differ according to the analysis being performed. The authors present a variety of imputation techniques, as well as bootstrap, jacknife, and other techniques for measuring the quality of the results.

The authors also dedicate chapters to approaches that work only with available data, and to cases where missing data can not simply be ignored.

This is the most thorough and practical guide I know to handling missing data. In an ideal world, experiments would all produce usable results and surveys would all have every question answered. When you have to deal with reality, though, this is the book.

Rating:
Summary: Classic Text on Missing Data
Review: This is the standard reference for statistics of missing data. Anyone working in the field will find it indispensable. The new edition is updated to cover a number of recent developments in the field.