07-08-2016 08:54 AM
Need advice: I am trying to find cutoffs in a multivariate logistic regression model composed of a categorical predictor and several continuous variables:
PROC LOGISTIC DATA=work.sdcrossvalid OUTEST=_SNSP_EST;
Bleed/ EXPB ;
To determine the cutoff value, I use the following formula:
X_value = (Logit - intercept)/(Sum of estimates of all predictor variables), as adapted from:
ref: Paper 231-2008
A Macro For Getting More Out Of Your ROC Curve:
X_VALUE = (LOG(_PROB_/(1-_PROB_))-INTERCEPT)/&XVAR
This paper only refer to a single continuous variable. What if you have several?
Is my denominator (Sum of estimates of all predictor variables) correct?
07-09-2016 02:18 AM
If you have multiple levels for Y , that would lead you to two kind of logistic model: 1) cumulative logistic 2)generalize logistic The cutoff value means nothing for these two kind of model. If it is binary logistic model, CTABLE could give you a good vision to cutoff value. proc logistic data=sashelp.class; model sex=age weight height/ctable; run;