1
$\begingroup$

I come across a practice where the logit (Intercept + beta* variables) is once again input as a independant variable in a single variable logistic regression against dependant variable to get intercept ($I$) and a beta ($B_0$). Finally, probability is calculated: $\frac{1}{1+e^{-(I+B_0\text{logit})}}$.

Why is this done -- since probability can be directly calculated from the logit in the first step?

$\endgroup$
  • 1
    $\begingroup$ Are there different dependent variables in the two models? $\endgroup$ – Macro Sep 8 '11 at 4:47
  • 1
    $\begingroup$ Can you post where you're coming across this? $\endgroup$ – Fomite Sep 11 '11 at 5:18
  • $\begingroup$ No different dependant variables.. It is the same default variable. I come across this in a prob of default model for auto loan portfolio $\endgroup$ – Suresh Sep 12 '11 at 4:37
2
$\begingroup$

"Why is this done" -- I don't think it is done ordinarily. I don't see any purpose for it. As you point out, there's another standard, correct way to calculate the probability of each case being "1" on Y. The kind of feedback you describe just seems to confuse things.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.