Date: Mon, 23 Feb 1998 09:54:59 -0500
Reply-To: Mike Davenport <Mike.Davenport@RICHMOND.PPDI.COM>
Sender: "SAS(r) Discussion" <SAS-L@UGA.CC.UGA.EDU>
From: Mike Davenport <Mike.Davenport@RICHMOND.PPDI.COM>
Content-Type: text/plain; charset=us-ascii
Well, the "best" way is to compute the least square means, if you can
define a glm model that describe the data. Then it is posible(i think)
to output the matrix with the data in the desired form.
Alternatively, you can compute the mean of each parameter using proc
means and output the results to a data set and merge the mean data back
into the original data set and replace missing values with the mean
The assumptions for this concept in any case are that the mean value is
a reasonable estimate (preferably a LSM which is BLUE) of a missing
value. It would be nice if the data is normally distributed, but if you
are using LSM results the ANOVA can deal with departures from normality
as long as the residuals are iid. Look at the residual plots and make
sure you dont have any discernable paterns, like an arc or trumpet.
If you have a "large" amount of missing data this method is going to be
difficult to defend. You would be better off trying a boot-strap
technique to determine the means of each parameter before assigning a
mean value to the missing data.
Sharon Alexander wrote:
> Missing Data
> I am a relatively new SAS user so my question will seem quite
> simple I am sure.
> I have been asked to replace missing data in a dataset containing
> approximately 200 numeric variables with the mean of the variable.
> I am not sure how to set up an array to do this. Could someone
> give me some guidance?