PROC MI : Differences between FCS REG and MONOTONE REG

[grep2638]

Hi everyone,

I am doing simulations to test multiple imputation procedures implemented in PROC MI.

For one of my datasets, I had the following error message :

ERROR: The set of classification levels for centre, generated for the imputed variable y_4, does
not include the level c3 for observations that have missing y_4 values.

It occured when I was imputing with the MONOTONE REGRESSION technique (on a monotone dataset, of course).

The code below (on SAS V9.4) reproduces such error:

DATA myData;
	INPUT centre $ y_1 y_2 y_3 y_4;
	DATALINES;
	c3 11 27 22 .
	c3 10 14 21 .
	c3 2 21 15 .
	c3 1 10 12 .
	c2 24 57 21 12
	c2 3 21 14 7
	c2 2 45 65 12
	c2 1 2 3 4
	c1 4 5 12 1
	c1 12 32 54 12
	c1 15 12 32 14
	c1 14 21 10 12
	;
RUN; 

PROC MI DATA=myData OUT=imputed;
	CLASS centre;
	VAR centre y_1 y_2 y_3 y_4;
	MONOTONE REG;
RUN;

 But with FCS REGRESSION, no problem !

PROC MI DATA=myData OUT=imputed;
	CLASS centre;
	VAR centre y_1 y_2 y_3 y_4;
	FCS REG;
RUN;

Yet, the SAS user guide for PROC MI  (pdf version available here) seems to assert that for monotone datasets, FCS REGRESSION fill-in phase acts like a MONOTONE REGRESSION  (page 5894 to 5895 and 5902 to 5903).

How to explain that the error occurs only for MONOTONE REGRESSION ? 

Thanks in advance

[SAS_Rob · Accepted solution]

You are exactly right.  With FCS, the problem should also surface.  

I suspect that this is an oversight in the development that does not catch the missing level and you should report it to Technical Support.

http://support.sas.com/ctx/supportform/createForm 

[Jonas_p]

Dear grep2638,

I'm facing exactly the same problem as you... have you been able to solve it eventually ?

Best wishes

Jonas