Suppose I run the following command in STATA:
estsimp regress y x1 x2 x3, mi(q) iout
This will produce a set of intermediate results based on each instance of q, a multiple imputation regression estimate, and a series of new variables in the existing data (b1, b2, b3, b4 and b5).
>From the intermediate results, it is possible to calculate a multi-imp estimate of the RMSE by averaging the individual RMSE's.
According to the estsimp documentation, the new variable b5 will contain the simulations of the mean square error. Thus, one should also be able to take the root of the mean of b5 and reach a multi-imp estimate of the RSME for the estimations.
In practice, the result reached by averaging the intermediate RMSE's and taking the root of the mean of b5 are different.
So, what I have (done wrong) (misunderstood) (missed) (etc.)?
Matthew A. Vile
Senior Data Analyst
Goodwill Industries International
15810 Indianola Dr.
Rockville, MD 20855
(240)333-5389
(301)530-1516 (fax)
--
Clarify mailing list served by Harvard-MIT Data Center
[Un]Subscribe/View Archive: http://lists.hmdc.harvard.edu/?info=clarify
