input coding for EM regression

[deleted_user]

Hi,

Will appreciate some advice from EM users.

I was trying to build a logistic regression model using EM and my input data contains class variables (e.g. binary indicators). The input coding options provided in EM are "GLM" or "Deviation". Is there any way in EM that we can code by "Dummy" with "1" as my interpretation reference?

This is because I noticed that if we use "Deviation", the results are interpreted with reference to "0". This is not so ideal as more often than not, we will like to know whether having that effect (i.e. "1") will have an impact on my target or not rather than the other way.


Thank you for sharing.

[Karsten_SAS]

Hi, the variable metadata item 'order' also determines the encoding reference value. As documented in the help, the default value for 'order' is 'descending'. This entails that 1 is the encoding reference level as it is the first value of 1 and 0 in descending order. Hence, the dummy variables are VARNAME_0 dummies. Simply change (e.g., using a Metadata node) the Enterprise Miner variable metadata 'order' from default to ascending. Thereby, you can enforce 0 to be the encoding references to create VARNAME_1 dummies. Hence, the regression coefficients are interpreted with respect to 1 or not 1 instead of 0 nor not 0. HTH and cheers, Karsten

[deleted_user]

Thanks for sharing! It's great to hear your reply and it's very useful.

Cheers,

Nic

[deleted_user]

Hi Karsten,

I've just tried with the method suggested and yes it's by tweaking the order to set the reference point. Probably a minor correction is to change to descending order instead of ascending. In any case thanks for directing me to the solution! 🙂

Cheers,

Nic

[Karsten_SAS]

Hi Nic, nice to hear it works. Cheers, Karsten