Stata The Stata listserver
[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: Logit and percentage of correct classification

From   Fred Wolfe <>
Subject   Re: st: Logit and percentage of correct classification
Date   Fri, 03 Sep 2004 05:07:05 -0500

In addition to Ronan's important point, another point that might be made is that model performance as assessed by simple logistic regression is overly optimistic. Running the model with 10-fold cross validation (as may be obtained in CART or by programming in Stata) is often illuminating as to the truer predictive ability of models.

Fred Wolfe

At 04:48 AM 9/3/2004, you wrote:

Herve STOLOWY wrote:

Dear All:
Is there an easy way to determine the percentage of correct
classification (predicted probabilities) after a logit (or logistic)
regression (equivalent to Predicted/Percentage correct in SPSS)?
You can, using -predict-, generate the predicted probability. You can then decide on an appropriate threshold for a predicted event. The usual one is 0.5, but there are many situations where the consequences of incorrect prediction are quite different for false positive and false negative predictions.

For instance, the current European guidelines for prevention of cardiovascular disease recommend treatment when a person has a 10% chance of a fatal heart attack or stroke. Except, of course, for the British, who use a threshold of 30% risk of fatal or non-fatal heart attack. The cutpoints at which you evaluate your model fit need to be established pragmatically.

You can then assess your model performance using -diagt-, which gives detailed statistics on the pattern of errors in your prediction system.

Ronan M Conroy (
Senior Lecturer in Biostatistics
Royal College of Surgeons
Dublin 2, Ireland
+353 1 402 2431 (fax 2764)
Just say no to drug reps
* For searches and help try:

Fred Wolfe
National Data Bank for Rheumatic Diseases
Wichita, Kansas
Tel (316) 263-2125     Fax (316) 263-0761

*   For searches and help try:

© Copyright 1996–2022 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index