topic Re: proc logistic outmodel= scoring in Statistical Procedures

proc logistic outmodel= scoring

trekvana — Tue, 02 Aug 2011 15:55:10 GMT

so i created a simple data set and ordinal logistic model to illustrate the issue i am having. i am modeling my data (xxx) and then create a new data set (predict) for which i want to score and get estimated probabilities. everything works out fine ( i am able to get the estimated probabilities in the scores data set) but i keep getting these warning messages that I dont know how to fix. see below for warning messages

data xxx;

input y x1 x2;

datalines;

1 1 1

1 2 1

1 2 2

2 1 1

3 2 2

2 1 2

3 1 1

;

run;

proc format;

value y 1='satisfied' 2='neither' 3='dissatisfied';

value xone 1='no' 2='yes';

value xtwo 1='cat' 2='dog';

run;

proc sort data=xxx; by y; run;

proc logistic data=xxx order=data outmodel=estimates;

format y y. x1 xone. x2 xtwo.;

class x1(ref='no') x2(ref='cat') / param=ref;

model y = x1 x2;

run;

data predict;

do x1=1 to 2 by 1;

do x2=1 to 2 by 1; output;

end;

run;

proc logistic inmodel=estimates;

score data=predict out=scores;

format x1 xone. x2 xtwo.;

run;

sas keeps telling me that x1 and x2 are not in the estimates dataset but if you look into the dataset you can clearly see the two variables are in there.

7052 proc logistic inmodel=estimates;

7053 score data=predict out=scores;

7054 format x1 xone. x2 xtwo.;

WARNING: Variable X1 not found in data set WORK.ESTIMATES.

WARNING: Variable X2 not found in data set WORK.ESTIMATES.

7055 run;

NOTE: The data set WORK.SCORES has 4 observations and 6 variables.

never the less I still get the predicted probabilites in the score dataset. however when I reverse the score and format statements i get this error:

7056 proc logistic inmodel=estimates;

7057 format x1 xone. x2 xtwo.;score data=predict out=scores;

WARNING: Variable X1 not found in data set WORK.ESTIMATES.

WARNING: Variable X2 not found in data set WORK.ESTIMATES.

7058

7059 run;

NOTE: The scored data contains missing values due to a class level which is not in the training data set.

NOTE: The data set WORK.SCORES has 4 observations and 6 variables.

the predicted probabilities do not get calculated.

if i leave out the format statement the predicted probabilities still do not get calculated:

7060 proc logistic inmodel=estimates;

7061 score data=predict out=scores;

7062 run;

NOTE: The scored data contains missing values due to a class level which is not in the training data set.

NOTE: The data set WORK.SCORES has 4 observations and 6 variables.

i am thinking it has something to do with the format statment. its driving me nuts. any help?

Re: proc logistic outmodel= scoring

Reeza — Tue, 02 Aug 2011 16:19:04 GMT

Where does that estimates dataset come from? I see the data set XXX and predict have x1 and x2, but the estimates dataset kinda pops up outta nowhere, or is that just a typo for the demo here?

Your also missing a semi-colon after the run statement (with the proc sort).

Re: proc logistic outmodel= scoring

trekvana — Tue, 02 Aug 2011 16:23:16 GMT

reeza- oops my mistake. in the original proc logistic it should be outmodel=estimates instead of outmodel=scores. i have corrected this

edit: ok corrected the semicolon. i am stilling getting those warnings. any suggestions?

Re: proc logistic outmodel= scoring

Reeza — Tue, 02 Aug 2011 16:29:45 GMT

From the SAS DOCs :

FORMAT statements are not allowed when the INMODEL= data set is specified; variables in the DATA= and PRIOR= data sets in the SCORE statement should be formatted within the data sets.

Re: proc logistic outmodel= scoring

trekvana — Tue, 02 Aug 2011 16:32:07 GMT

would you be able to tell me how i could format the variables within the data sets?

Re: proc logistic outmodel= scoring

trekvana — Tue, 02 Aug 2011 16:37:12 GMT

haha. the answer was so simple:

data predict;

do x1=1 to 2 by 1;

do x2=1 to 2 by 1; output;

end;

format x1 xone. x2 xtwo.;

run;

and now everything works ok. thanks reeza for the eye opener