Mathematical calculation of confidence intervals & comment on statacorp post

[Tara Boyle]

Hello all
I came across this post on stata corp post https://www.stata.com/support/faqs/s...nce-intervals/

https://www.stata.com/support/faqs/s...nfidence-inter

vals/

I'm trying to understand this code written by an author on a publication

Author displays his code on how he calculated how Surgical cases in urology surgeons affects recurrence in prostate cancer.

So as you can see the author predicts xb (linear predictors) – as per post in stata corp
The author then goes on to generate the Upper Bound and lower bound CI (seen above in my green notes)

My questions

1. Why does the author use -exp- and a -xb ? (stata corp is positive)
2. Why does the author also include a^(1/gamma) ^-1 ? Does this have any relation with Weibull at all ?
3. If I may push my luck for another question - I don't understand what stata corp is trying to generate in the sections marked in blue

*Note the gamma is a scalar saved from the parametric model log-logistic model performed prior to this .

I don’t understand why the red steps are performed as this is not given in stata corp post...can anyone explain this ?

Also, with the -predict- in stata corp it says can be used with logistic regression. Can this be used with COX/ Parametric tests? From what I read one needs to take care as it is essentially summarising means for all values.
Is this correct?

[Leonardo Guizzetti]

The two methods are not the same because they are for different models. There is no reason they should be calculated in the same way or yield the same values.

The Stata FAQ discusses confidence limits for a logistic regression model. The confidence limits calculations in your screenshot are for some kind of survival model (apologies, but the image is blurry and my eyes are not what they once were). The author of the article should make clear what type of regression model they are using.