Hi all,

I have run a logistic regression using the user-written command boxtid, and would now like to know how calculate confidence intervals for the power estimates defined as p1 in the results table below:

Box-Tidwell regression model

Logistic regression Number of obs = 185,042
Wald chi2(12) = 503.07
Prob > chi2 = 0.0000
Log pseudolikelihood = -1763.3257 Pseudo R2 = 0.1007

(Std. Err. adjusted for 30,502 clusters in ddyadid)


Robust
midint | | Coef. | Robust Std. Err. | z | P>|z| | [95% Conf. Interval]
------------------------------------------------------------------------------------


ktbitr 2.130029 .4880317 4.36 Nonlin. dev. 20.272 (P = 0.000)
p1 4.832425 1.261686
kigojm 1.16586 .3525975 3.31 Nonlin. dev. 9.763 (P = 0.002)
p1 5.959293 2.862814
gdppc2000_1 -.0000383 .0000168 -2.28 Nonlin. dev. 8.778 (P = 0.003)
p1 .1208477 .2674343
mcap_1 9.392144 1.854957 5.06 Nonlin. dev. 44.082 (P = 0.000)
p1 .1215187 .0529244
tpop_1 -9.65e-07 5.05e-07 -1.91 Nonlin. dev. 10.339 (P = 0.001)
p1 .0991047 .1147337

I've seen on various threads that confidence intervals can be manually calculated using the standard error, degrees of freedom and the inverse estimator function - see, for instance, #3 in https://www.statalist.org/forums/for...ls-after-nlcom.

The problem is that I am not sure what the current estimator function is in this particular case - I assume logit, but I'm also inclined to think that the calculation of power estimates is in some way distinct from the standard coefficients? I'd be incredibly appreciative if someone could tell me the correct function, and also explain the underlying logic.

All the best,
Matthew