logistic regression intercept is negative and not significant
- Thread starter Tucker04
- Start date
It's not my field of expertise (though I'd like it to be--I really wish I knew more about categorical models) but my understanding is that a negative intercept in a logit model meand that if all the explanatory variables were 0, you'd have less than even odds of observing a positive result for your dependent variable. If the intercept is not itself significant, I suppose that would mean the odds in such a case are not significantly different from even.
If I'm wrong on this someone please speak up; I'd like to know also.
If I'm wrong on this someone please speak up; I'd like to know also.
When using logistic regression, you are fitting the following linear model:
log(p/1-p) = b0 + b1*x1 + b2*x2 + ... + bk*xk + E
As you can see, you are modeling the changes in the log odds. That's usually not very helpful. When analyzing this type of models, it is often more informative to obtain odds ratios, which are the exponentiated coefficients. And there's no odd ratio for the intercept, so it won't matter whether is negative or not. You can safely remove it from the model if not significant.
log(p/1-p) = b0 + b1*x1 + b2*x2 + ... + bk*xk + E
As you can see, you are modeling the changes in the log odds. That's usually not very helpful. When analyzing this type of models, it is often more informative to obtain odds ratios, which are the exponentiated coefficients. And there's no odd ratio for the intercept, so it won't matter whether is negative or not. You can safely remove it from the model if not significant.