gsem graphical postestimation for time-to-event

[Niels Henrik Bruun]

Dear all,
I've reviewed example 48g in the manual.

The code:

cls
use https://www.stata-press.com/data/r19/gsem_diet, clear
stset dox, failure(fail) origin(time dob) enter(time doe) id(id)
gsem (_t <- energy job height weight, family(loglogistic, failure(_d) ltruncated(_t0))), nolog
estat eform

Gives:

. estat eform
------------------------------------------------------------------------------
             |     exp(b)   Std. err.      z    P>|z|     [95% conf. interval]
-------------+----------------------------------------------------------------
_t           |
      energy |      1.072      0.063     1.19    0.23        0.956       1.202
         job |      1.010      0.030     0.35    0.73        0.953       1.071
      height |      1.011      0.005     2.28    0.02        1.002       1.020
      weight |      1.000      0.003     0.18    0.86        0.995       1.005
       _cons |   3330.551   2608.007    10.36    0.00      717.758   15454.480
------------------------------------------------------------------------------

The interpretation is: Each of the time ratios is just above 1, so an increase in any of the covariates would slightly increase
the expected time to failure (for example, having a job increases the expected time until developing a
disease by 1.01). However, only the time ratio for height is significantly different from 1.

I've found no examples in the documentation on post-estimation showing any graphical reporting possibilities.
I'm having difficulties graphing the results since the outcome is _t, which I usually place on the x-axis.

What are the possibilities using the commands margins, marginsplot, and predict?

[Stephen Jenkins]

I think you might be over-interpreting the example. I think it's there simply to demonstrate -gsem-'s flexibility. If you really wanted to look at these data using survival analysis, you'd use -streg- with its "graphical reporting possibilities.

use https://www.stata-press.com/data/r19/gsem_diet, clear
stset dox, failure(fail) origin(time dob) enter(time doe) id(id)
streg energy job height weight, d(loglogistic)
streg, tratio
help streg postestimation // for graphs,margins,  etc.

[Niels Henrik Bruun]

Stephen Jenkins Thank you for your reply. I agree with you about the model described. I was researching the code

gsem (panss <- i.treat##i.week U[id]@1) (droptime <- i.treat U[id]@gamma, family(weibull, failure(infdrop))

in https://www.stata.com/meeting/uk16/s...henko_uk16.pdf

In this case, I don't think that post-estimation in streg is sufficient.