stcox: convergence issues using profile likelihood (-pllf- from SSC)

[Andrew Lover]

* Re-post with more informative subject line.

Hi Statalisters,

I am trying to extract useful information from 'infinite' coefficients/CIs as suggested here using profile likelihoods and -pllf- from SSC:

http://www.ctspedia.org/do/view/CTSp...initeEstimates

However, I am having (not totally surprising) convergence issues; any ideas or suggestions?

thanks-

Andrew Lover

****** start example

// ssc install pllf
use "http://www.stata-press.com/data/r11/repair", clear

logistic foreign ib3.repair   // complete separation

* generate some survival times
set seed 12345
gen time = 20*runiform(),, if foreign == 0
replace time = 5*runiform(),,  if foreign != 0
tabstat time, by(foreign)

* need explicit dummies for -pllf-
tab repair, gen(rep_dum)

* general Cox model
stset time, fail(foreign)
stcox rep_dum*

* Profile likelihood

pllf stcox rep_dum*, profile(rep_dum1) diff trace // fails to converge

pllf stcox rep_dum*, profile(rep_dum1) /// // not convincing
    diff range (-1000 50) trace
    
****** end example