Retrieving probabilities from ordered multinomial regression

[Bright]

Hi,

I have a dataset like the one below, where Lag_level and Level can take any value from {0,1,2,3,4} and LOS_category takes values from set  {1,2,3,4}.

Lag_level  Level  LOS_category

0                1        1

4                2         2

I am trying to estimate the probability of Level as the dependent variable and independent variables are  Lag_level  and LOS_category. I used ordered multinomial logit regression with the following code.

proc logistic data =have;
class lag_level (param = ref ref="4") LOS_category(param = ref ref="4");
model Level (descending)=lag_level  LOS_category /unequalslopes=(lag_level);
output out=outreg  predicted=p1;
store  out=Level_transition;
run;

Now I want to predict the (individual) probabilities (as opposed with cumulative probabilities) for a test dataset like this:

data test2;
   input lag_level  LOS_category;
   datalines;
0 1
1 1
2 1
3 1
4 1
0 3
1 3
2 4
3 3
4 2
;

I tried to use proc plm and restore the model as follows:

proc plm restore=level_transition;
   score data=test2 out=testout2 predicted / ilink;
run;
proc print data=testout2;
run;

But then I received an error "ERROR: The file WORK.LEVEL_TRANSITION does not exist or it is not a valid item store." Because these procedure works for standard logistic regression, I was wondering if anybody can help with this cumulative multinomial one.

Thanks!

[Rick_SAS]

I believe the problem is that you are using the UNEQUALSLOPES option. The doc for the LOGISTIC PROCEDURE says,

"The STORE statement is not available for models created with the LINK=ALOGIT option, the EQUALSLOPES or UNEQUALSLOPES options, a STRATA statement, or an EXACT statement."

You can work around the problem by removing the option.  You can create a separate PROC LOGISTIC call that performs the unequal slopes analysis but does not contain the STORE statement.

[Bright]

Thanks for the reply. So do you know any other way that I can estimate probabilities?

[Rick_SAS · Accepted solution]

If you want to do it in a single call, you can use the SCORE statement

proc logistic data =Have;
class lag_level (param = ref ref="4") LOS_category(param = ref ref="4");
model Level (descending)=lag_level  LOS_category /unequalslopes=(lag_level);
output out=outreg  predicted=p1;
SCORE data=test2 out=pred;
run;

proc print data=pred;
run;

[Bright]

Thanks for your help!