Re: Multiple Imputation: Combining Results for F-Tests - Amelia

15 Nov 2003

On Sat, 15 Nov 2003, Paul Miller wrote:

...
  Dear Dr. King,
  I'm currently using multiple imputation to deal with missing
 observations in an analysis I'm conducting. I want to combine the
 f-tests associated with each of my imputations but it's not clear how I
 should do so. Is there any way that a non-statistician such as myself
 could go about combining these results?

 Paul   
if the f-test is really your quantity of interest, you could use the same
rules as combining any other quantity (see our article for a description;
the point estimate is simply an average of the separate estimates).  
However, most likely the f-test is actually being used as a test of some
kind and you have another quantity of interest, such as a first differnce,
or regression coefficient. In all likelihood, you have one of these, and
I'd suggest that you compute it, following the rules for combining
separate quantities of interest.

If you have a more complicated case with multiple quantities of interest, 
i would consider combining the simulations from each of the analyses of 
the imputed datasets.

Best of luck with your research,

Gary

     : Gary King, King(a)Harvard.Edu    http://GKing.Harvard.Edu :
     : Center for Basic Research      Direct    (617) 495-2027 :
     :   in the Social Sciences       Assistant (617) 495-9271 :
     : 34 Kirkland Street, Rm. 2      HU-MIT DC (617) 495-4734 :
     : Harvard U, Cambridge, MA 02138    eFax   (928) 832-7022 :