topic Re: proc logistic - generalized logit model OR in Statistical Procedures

proc logistic - generalized logit model OR

SasNewb1 — Fri, 23 Dec 2016 00:17:46 GMT

I'm using SAS 9.4 and I'm running a generalized logit model in proc logistic (stepwise selection) and my outcome variable has 4 levels: A, B, C, and D.

How can I set up my code such that the odds ratio compares A to B, B to C, C to D? Right now SAS default compares A to B, A to C, A to D.

proc logistic data=data1 descending;

class sex(ref='M') /param=ref;

model levels=weight age sex height /slstay=0.05 slentry=0.05 scale=none selection=stepwise aggregate link=glogit;

run;

Re: proc logistic - generalized logit model OR

DHINESHSHANKAR0 — Fri, 23 Dec 2016 00:40:43 GMT

One way for comparison among the predictor can be by standardized estimate option in the model statement ,.ie stb .

Then the absolute value of the standardized estimate can be used to provide the approzimate ranking based on relative importance of the x variables in logistic model.

Re: proc logistic - generalized logit model OR

clim072 — Fri, 23 Dec 2016 01:00:19 GMT

Not sure if this will work but you can try adding in REF in your MODEL statement after your dependent variable.

For example, level B is not your reference group, this will compare B vs A, B vs C, and B vs D.

proc logistic data=data1 descending;

class sex(ref='M') /param=ref;

model levels (REF = "B") =weight age sex height /slstay=0.05 slentry=0.05 scale=none selection=stepwise aggregate link=glogit;

run;

Hope this helps.

Re: proc logistic - generalized logit model OR

Reeza — Fri, 23 Dec 2016 01:22:17 GMT

Have you tried the ODDSRATIO statement?

Re: proc logistic - generalized logit model OR

Ksharp — Sat, 24 Dec 2016 04:30:44 GMT

It is called  adjacent-category logit model.
try link=alogit.


data data1;
 set sashelp.class;
 if _n_ lt 6 then levels='A';
  else if _n_ lt 12 then levels='B';
   else levels='C';
run;
proc logistic data=data1 descending;
class  sex(ref='M') /param=ref;
model levels=weight age sex height /slstay=0.05 slentry=0.05  
scale=none selection=stepwise aggregate link=alogit;
run;


OR if you don't have continuous variable, you could also try PROC CATMOD.

proc catmod.........
response alogit;
model......