Re: st: Averages for missing values

["Richard. Williams" <Richard.A.Williams.5@nd.edu>]

At 06:01 PM 9/8/2005, SamL wrote:
> These methods are frowned upon?  Which method in particular is the object
> of the frown?  I guess I am puzzled, because mean substitution comes in a
> few flavors.  There is substituting the mean and leaving it at that.  I
> assume this is widely understood as problematic.  And then there is
> substituting the mean and adding a dummy variable to the model that
> indicates whether the mean has been substituted.  Is there any problem
> with this second approach?
I taught and recommended the latter method for several years - and then 
retracted my support a few years ago!   See Allison's book on Missing 
Data.  For the Cliff's Notes version, see pp. 5-6 of

http://www.nd.edu/~rwilliam/stats2/l12.pdf

Allison gives some great examples showing that the procedure gives biased 
estimates and is worse than listwise deletion.  In fact, in general he 
shows that most of the traditional methods for handling missing data are 
worse than listwise deletion.  You should either use listwise deletion, or 
go with a more advanced method like multiple imputation.

I really need to update that handout some time, because it is full of 
advice on how to implement methods that you probably shouldn't 
implement!  Part of the problem though is that the better methods can be 
problematic or hard to implement too.


-------------------------------------------
Richard Williams, Associate Professor
OFFICE: (574)631-6668, (574)631-6463
FAX:    (574)288-4373
HOME:   (574)289-5227
EMAIL:  Richard.A.Williams.5@ND.Edu
WWW (personal):    http://www.nd.edu/~rwilliam
WWW (department):    http://www.nd.edu/~soc 
*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/