topic Re: Proc Logistic: predicted probabilities question in SAS Procedures

Proc Logistic: predicted probabilities question

jfine73 — Fri, 27 Jun 2014 18:54:50 GMT

I have a a data set where the dependent variable has 4 levels (low, low-medium, medium, and high) with three independent continuous variables.

So my code currently looks like this:

proc logistic data=updated;

model groupedSet = bioMarker age bmi;

run;

the output gives me 3 intercepts, which is great, but only gives me 1 estimate and odds ratio estimate for the bioMarker, age, and bmi. Is there a way to see what the predicted probability is for the bioMarker at each level of the groupedSet? I understand the main problem is because it is a continuous variable. So is there any way around that? I've been trying to find some papers on it but am have bad luck so far.

Thanks.

Re: Proc Logistic: predicted probabilities question

Reeza — Fri, 27 Jun 2014 18:58:45 GMT

You can use the estimate statement or the odds ratio statement.

Here's an example of the estimate statement:

SAS Data Analysis Examples: Ordinal Logistic Regression

Re: Proc Logistic: predicted probabilities question

jfine73 — Fri, 27 Jun 2014 19:26:42 GMT

Thank you for the link Reeza,

In the example they have the 'paired' variable either be 0 or 1 (a categorical variable). Mine, however, is continuous. Should I just pick the cutoff points as the same as the cutoff points for the dependent variable or just leave it as it is? Also if I wanted to standardize for age and bmi in this estimate would I set this to be 0 or 1?

something like this?

estimate "pr prb groupedSet = 1" intercept 1 bioMarker 1 age 1 bmi 1 \ ilink category = '1';

estimate "pr prb groupedSet = 2" intercept 1 bioMarker 1 age 1 bmi 1 \ ilink category = '2';

estimate "pr prb groupedSet = 3" intercept 1 bioMarker 1 age 1 bmi 1 \ ilink category = '3';

Re: Proc Logistic: predicted probabilities question

Reeza — Fri, 27 Jun 2014 19:29:28 GMT

GPA was continuous.

You usually look for the estimate at the mean of the value but that depends on what you're looking for.