Is there a preferred and simple method in Stata10 for external
validation of a model based on logistic regression?
How do you generate an ROC on the validation dataset using the
coefficients of the primary dataset. Are there other tests that would
then be run?
Would it be valid to run the two regressions separately and somehow
compare them-- lrtest, for example. Or somehow compare the coefficients
for being statistically different?
If a second, separate dataset is not available, Hosmer and Lemeshow
(Applied Logistic Regression) 2nd ed, and others describe an internal
validation: running a logistic regression on a subset (70%?) of a data set,
then using the coefficients from the primary regression on the remaining
subset (30%?) of the dataset. 70/30 is my guess-- is this standard?
How would you generate an ROC on the smaller subset using the
coefficients of the primary regression to generate the predicted values?
Is a 100x bootstrap resampling technique a more complete version of
this 70/30 internal validation?
Thanks in advance.
Mitchell F. Berman
Columbia University
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
