Is there something like r-squared for exponential models?

[Christoph Feucht]

I'm working on a survival model. And I would like to compare different models in how good they are at predicting a duration.
For example I would like to compare the following models:
streg i.gender i.industry, distribution(e) nohr
streg yrbirth i.industry, distribution(e) nohr
streg i.gender yrbirth i.industry, distribution(e) nohr

I read up that r-squared can be used only in linear regressions https://statisticsbyjim.com/regressi...ar-regression/

What is an alternative measure for non-linear regressions? And what command can I use to display it?

[Carlo Lazzaro]

Christoph:
see -help estat ic-.

[Nick Cox]

https://www.stata.com/support/faqs/s...ics/r-squared/ should annoy everybody somehow.

[Joro Kolev]

Nick shows one approach.

Another approach is in Magee, Lonnie. "R 2 measures based on Wald and likelihood ratio joint significance tests." The American Statistician 44, no. 3 (1990): 250-253.

You can always get an R-square like measure as the McFadden's R squared = 1 - (log likelihood of model)/(log likelihood of model with a constant only).