Resources for Multiple Imputation 

 As applied researchers, we all know this situation all too well. Like the alcoholic standing in front of the bar that is just about to open, you just downloaded (or somehow compiled) a new dataset. You open your preferred statistical software and begin to investigate the data. And there again you are struck by lightening: Holly cow - I have missing data!! So what do you do about it? Listwise deletion as usual? In the back of your mind you recall your stats teacher saying that listwise deletion is unlikley to result in valid estimates but hitherto you have simply ignored these caveats. Don't be a fool, you can do better -- use  multiple imputation  (MI). 


 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!