topic Re: Input Variables in Logistic Regression in Statistical Procedures

Input Variables in Logistic Regression

juanvg1972 — Thu, 12 Apr 2018 17:25:36 GMT

Hi,

I am working in a logistic regression using 'proc logistic'. One of the input variables is 'age' (from 18 to 78). I have created a new var 'age_levels' that is the result of making ranges from 'age' using a clustering, it has 4 levels.

Using proc freq I can see that there is a dependency between 'age_levels' and the target var of the model.

My doubt is wheter var I have to use in the model of proc logistic , 'age' or 'age_levels' as input var.

Can anybody help me??

Thanks

Re: Input Variables in Logistic Regression

PaigeMiller — Thu, 12 Apr 2018 17:57:40 GMT

You can use any variable(s) that you think are appropriate in such a model.

I usually avoid clustering of input variables such as age into four levels. This seems to me to throw away potentially useful information that is contained in age.

Re: Input Variables in Logistic Regression

Reeza — Thu, 12 Apr 2018 19:50:24 GMT

When you use age as a categorical variable you're saying that the boundaries reflect some grouping. So if it makes sense that a 41 year old has different outcomes than a 39 year old, assuming one age cut off is 40, then use the categories. Otherwise the general consensus is don't group the variables unless there's some reason to do so.