Ordinal Predictor in SAS VDMML Regression task

[andreas_zaras]

Hello!

Let us assume that we have an ordinal predictor and we want to run a linear or a logistc regression in VDMML. How does SAS VDMML handles the ordinal predictor e.g. as interval or using a dummy coding e.g. stairwise coding?

Thanks in advance,

Andreas

[sbxkoenk]

Hello @andreas_zaras ,

Any idea which procedures you are talking about?

• REGSELECT for Linear Regression?
• LOGSELECT for Logistic Regression?

An ordinal predictor should be put in the CLASS statement, such that it is considered as a discrete | categorical input. It should NOT be entered into the model as an interval-scaled input!

For ordinal inputs, SAS supports bathtub coding and THERMOMETER (cumulative parameterization) coding. I think bathtub coding is only supported when constructing the input layer in a neural network.

I do not know about "stairwise coding" but you can make a little macro for it.

Cheers,

Koen

[andreas_zaras]

Any details about bathtub and thermometer encoding? I mean how this is done using summy variables (i guess?)

[sbxkoenk · Accepted solution]

From the SAS 9.4 M7 help :

For bathtub coding, see here :

SAS® Enterprise Miner™ 15.2: Reference Help
Neural Network Node: Reference
https://go.documentation.sas.com/doc/en/emref/15.2/p0zbgj1tu3h1uhn1x6regixbdg7v.htm

CTRL-F : bathtub coding

Cheers,

Koen