Social Science Statistics Blog

As is well known in the statistcial literature on the missing data problem, MI is not the silver bullet for dealing with missing values. In some cases, better (primarily more efficent) estimates can be obtained using weighted estimation procedures or specialized numerical methods (EM, etc.) Yet, these methods are often complicated and problem specific and thus not for the faint of heart applied researcher. MI in contrast is relatively easy to implement and works well in most instances. Want to know how to MI? I suggest you take a look at www.multiple-imputation.com, a website that brings together various ressources regaring the method, software, and literature citations that will help you to add MI to your toolkit. A nice (non-technical) introduction is also provided on Joseph Schafer's multiple imputation FAQ page. Gary and co-authors have also written extensivley on this subject offering lots of practical advice for applied rearchers. Last but not least, I recommend searching for "multiple imputation" on Andrew Gelman's blog; you will find many of interesting entries on the topic. Good luck!