Exponentiating bootstrapped coefficients (redux)

[Andrew Lover]

Hi 'listers,

I am bootstrapping errors from a firthlogit model (SSC), but am getting an error with addition of 'or' or 'eform' options to either -firthlogit- itself or bootstrap.

webuse lbw //, clear

xi: firthlogit low age lwt i.race smoke ptl ht ui, or

* tiny # reps for ease of illustration

* fails immediately
xi: bootstrap, reps(10): firthlogit low age lwt i.race smoke ptl ht ui, or

* fails postestimation
xi: bootstrap, eform reps(10): firthlogit low age lwt i.race smoke ptl ht ui

There was an earlier discussion of this issue, but the same solutions don't appear to work with -firthlogit- surprisingly....

http://www.stata.com/statalist/archi.../msg00140.html

cheers!
Andrew

[Richard Williams]

Don't ask me why, but it seems to work with replay, using either bs or firthlogit:

webuse lbw, clear
xi: firthlogit low age lwt i.race smoke ptl ht ui, or
* tiny # reps for ease of illustration
xi: bootstrap, reps(10): firthlogit low age lwt i.race smoke ptl ht ui
bs, or
firthlogit, or

[Richard Williams]

Going back to that example you linked to from 2008 -- the replay option takes care of that problem too:

sysuse auto, clear
* does not work
xi: bootstrap _b, eform reps(10) dots: logistic foreign i.rep78 price
* Does work
xi: bootstrap _b, reps(10) dots: logistic foreign i.rep78 price
bs, or

[Andrew Lover]

Hi Richard,

That's bizarre, and in hindsight maybe should have been an obvious thing to try.... but that's incredibly helpful- thanks!

cheers
Andrew

[Richard Williams]

What is even more bizarre is that, after the seemingly fatal error, replay works:

webuse lbw, clear
* fails postestimation
xi: bootstrap, eform reps(10): firthlogit low age lwt i.race smoke ptl ht ui
bs
bs, or

If this isn't buggy behavior it is certainly sub-optimal. I'd be annoyed if I had some monster job get a seemingly fatal error after two days of running, especially since it wasn't really fatal.

[Andrew Lover]

A rare and elusive Quasibug?!