Different regression result on same dataset from both Enterprise Miner and SAS base

[JinboZhao]

Hi SAS,

I used SAS Enterprise Miner to generate a scorecard, then I got the corresponding logistic regression. My regression result is as follows:

Event Classification Table

Data Role=TRAIN Target=TARGET Target Label=' '

False True False True
Negative Negative Positive Positive

16 31 384 3928

Then I output the regression generating code and opened it in SAS base. I run it on the same dataset. This time I got the frequency table of my old and new predicted target: 

Table of TARGET by U_TARGET
	U_TARGET(Unnormalized Into: TARGET)	Total
1
TARGET		415	415
0	Frequency
1	Frequency	3944	3944
		4359	4359
Total	Frequency

It gave a totally different result with the one from E-Miner. 

I do not know why, and I doubt of the accuracy of the result from E-miner. Could anyone give me some clue? Thank you. 

[Reeza · Accepted solution]

WIth what you've posted, not really.

Most likely the regression node in Base SAS has different default settings than in EM. 

Start by verifying the models have the same variables and selection method. 

Then move to the regression options, probably easiest to use the EM settings on the left hand pane and then check the Base SAS settings.

[ballardw]

Data and code for everthing to see possible differences.

Note that the Regression procedures by default will remove any record that has missing values for any of the variables on the model statement. I don't know wheter Enterprise Miner may just treat missing values as a different level of a a variable, which is an option in most of the regression procs depending on the role the variable plays and which procedure.