Log vs. Logit Transformation in SAS Enterprise Miner

[Tmac]

Hello,

I have question about the transformation in SAS Enterprise Miner. It seemed kind of easy but being a newbie Im need for some help:

I have numerous continous variables (scale 0 to 1) and would like to linearise them to be able to do regression models.

As far as I learned one would use a logit transformation to put them onto a - infinity to + infitiy intervall to be able to use regression models. As there are are numerous variables this would take way to long to do it with formulas in the transformation tab. So is there are way to do it for all wanted variables in fairly quick way?

I tried to do just log transformations (under Default methods --> Intervall inputs) which gave me a quite similar distribution of the variable as when I am doing a logit transformation (I compared around 10 logit vs log transformation). So would that be a legit workaround?

Best regards 🙂 

[tomrvincent]

You might want to tag this with Miner instead of Guide.

[PaigeMiller]

---

@Tmac wrote:

I have numerous continous variables (scale 0 to 1) and would like to linearise them to be able to do regression models.

---

Are these continuous variables (scale 0 to 1) the independent variables or the response variables?

As far as I learned one would use a logit transformation to put them onto a - infinity to + infitiy intervall to be able to use regression models. As there are are numerous variables this would take way to long to do it with formulas in the transformation tab. So is there are way to do it for all wanted variables in fairly quick way?

It sounds like these are the independent variables you are talking about, and there's generally no need to transform them. Certainly, if you mean the independent variables, you would not want to use a logit or a log on them, without a very strong reason (which I don't see).

I tried to do just log transformations (under Default methods --> Intervall inputs) which gave me a quite similar distribution of the variable as when I am doing a logit transformation (I compared around 10 logit vs log transformation). So would that be a legit workaround?

The distribution of the independent variables is not particularly relevant in regression.