Logistic regression

[ks94]

hello,

please how write Logistic regression without specifying the variable selection method ?

thanks.

[ballardw]

List the variables you want on the model statement.

Dummy example code:

model result = variable1 variable2 variable3 ;

[ks94]

Thanks for your answer. Between my variables ;there are numeric and character variables ,so when i try this :

PROC logistic data=africa descending ;

model FLAG_CRISE_SYSTEMIQUE(event="1")=FLAG_CRISE_SYSTEMIQUE
TAUX_CHANGE_USD TOTAL_DETTE_PIB TAUX_INFLATION
FLAG_CRISE_INFLATION FLAG_DETTE_INTERIEURE
PAYS FLAG_DETTE_EXTERIEURE
FLAG_INDEPENDANCE FLAG_CRISE_MONETAIRE FLAG_CRISE_BANCAIRE;
run;

I've an error message , which required numeric variable or i must specified in the CLASS statement.

But my purpose is that I want to use this method to recognize my significant variables between the differents variables... my target variable is  FLAG_CRISE_SYSTEMIQUE.

I hope you understand my problem ?

[ks94]

thank 

i found the solution to my problem with the code bellow:

proc logistic data=africa;
class FLAG_DETTE_INTERIEURE(param=ref ref='1') FLAG_DETTE_EXTERIEURE(param=ref ref='1') FLAG_INDEPENDANCE(param=ref ref='1') FLAG_CRISE_MONETAIRE(param=ref ref='1') FLAG_CRISE_INFLATION(param=ref ref='1');
model FLAG_CRISE_SYSTEMIQUE(event="1") = ANNEE TAUX_CHANGE_USD TOTAL_DETTE_PIB TAUX_INFLATION FLAG_DETTE_INTERIEURE FLAG_DETTE_EXTERIEURE FLAG_INDEPENDANCE FLAG_CRISE_MONETAIRE FLAG_CRISE_INFLATION;
run;

the variables which are Pr>khi-2 <0.05 are my significant variables. 

[SteveDenham · Accepted solution]

Which of the variables are not continuous?  Put those in a class statement and add /PARAM=GLM after the variable list.  Add /SOLUTION to the MODEL statement.

You should then get a Type 3 table of tests for the variables. In particular, there will be a test for the main effect of FLAG_CRISE_SYSTEMIQUE.  I hope this makes sense.  If not, please post your code and any part of the log that outlines ERRORs or WARNINGs.

SteveDenham

[SteveDenham]

I guess I was a bit late in my suggestion. You seem to have solved the problem.

SteveDenham