0
$\begingroup$

I am working on an event history analysis project. I am using a discrete time survival model. Specifically, I am using a logit model. I am interested in examining the substantive effects. Can you plot the predicted probabilities? So far, the only way I have seen scholars interpret the substantive effects is through the odds-ratios. However, I generally find predicted probabilities more intuitive in interpreting GLM models.

Any insight would be greatly appreciated.

$\endgroup$
1
$\begingroup$

It turns out that the inverse cumulative distribution function (so-called quantile function) of the logistic distribution is a generalization of the logit function, see Wikipedia discussion here.

So, if you assume that the probabilities occurring in the logit ratio (log(p/1-p)) are following a uniform distribution, then one can interpret the logits in turns of the standardized (a mean of zero and a scale factor of one) cumulative logistic distribution, hence a probabilistic interpretation of the log-odds ratio.

Here is Wikipedia on this characterization of the logistic distribution:

Its cumulative distribution function is the logistic function, which appears in logistic regression and feedforward neural networks. It resembles the normal distribution in shape but has heavier tails (higher kurtosis). The logistic distribution is a special case of the Tukey lambda distribution.

Note, a Monte Carlo inversion method to simulate a Logistic random deviate is to replace p in the logistic function (and more generally than the standardized distribution, employing the quantile function) with a randomly generated uniform random deviate. I have employed this quite successfully in practice and on this forum used it to explore nonlinear estimators for Logistic distribution parameters.

| cite | improve this answer | |
$\endgroup$

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.