Logistic Regression - Part 3 - Result Interpretation

How to interpret the SAS output of Proc Logistic ?

Once we run Proc Logistic in SAS with various options illustrated in previous article "Logistic Regression - Part 2 - How to do it ?" , it gives various tables in the output window.
which of all those tables are important and what do those tables convey? How to check if the model is good or not so good ? Let's learn it in this article .




In logistic Regression, maximum likelihood criteria is used. The Estimates in the table "Analysis of Maximum Likelihood Estimates" are nothing else but beta coefficients the regression equation.



Unlike R-Square in Linear Regression, we have multiple things in Logistic Regression that indicate the health of our model

1.  Concordance and it's derived statistics such as Somer's D, C etc.
2.  HL Test
3.  ROC Curve



Percentage concordance should be around 75% or more for a model to be  in good health.





Classification table is the table that is easier to understand and also speaks language of business. We can take a cut-off of Probability value looking at this itself ( however there are other methods too for that).

In the  table below, we can say that correctness being highest at 0.5 prob level, it can be considered as cut-off. In "English", at this probability cut off, model is most efficiently separating 1 and 0 ( Event and Non-event).



Also called as confusion matrix, I hope after our explanation, you won't use this name again.


As both the Sensitivity and Specificity indicate the correctness of the model, both should be on higher side.

So if we take 1- Specificity in lieu of Specificity, it should be on the lower side.

Hence when Sensitivity and 1 - Specificity are plotted on X and Y axis and it is checked how close is the curve to Y axis. Curve closer to Y axis means that with at High Sensitivity level, Specificity is also high.


Also, it is checked how similar are the ROC curves for Test and Validation samples. It should be similar.

In the next episode of Logistic Regression, you would learn other methods of taking Probability Cut-off.


Enjoy reading our other articles and stay tuned with ...

Kindly do provide your feedback in the 'Comments' Section and share as much as possible.