Duration model, cox or whatever?

[Javier Moreno]

We have to do a duration model with ens_dur.dta to look for spell to quit smoking, and what covariates affects its duration and how to reduce it. We have made this code, but we dont know what hazard function to specify (weibull, log logistic...) , or to use a cox hazard. We are just beginners and we have written that, can you help us? Have we to test for Proporitonal Hazard?

Dataset: https://gofile.io/?c=SL3Jow

use ens_dur.dta

/*Descriptive Statistics */

des

sum

tab smoker
tab physical_act
tab bmi
tab health_q
tab d_compl_init
tab d_compl_quit

 /* Drop observations for Ceuta and Melilla */

drop if ccaa>=18

/* Create binary dummies */

gen native=0
replace native=1 if Spanish==1

gen female=0
replace female=1 if sex==2

gen chrill=0
replace chrill=1 if chronic==1

/* Setting the duration and the censoring variables */

stset dur_quit, failure(d_compl_quit)

/* Empirical hazard */

sts graph, hazard

/* Estimation of continuous time duration models */

/* Basic models */

streg, dist(exponential) nohr
streg, dist(weibull) nohr
streg, dist(loglogistic)
streg, dist(lognormal)
streg, dist(ggamma)
streg, dist(gompertz)
stcox, estimate

/* Complete model */

streg c.age##c.age female native i.ccaa i.size_mun i.marital_st /*
      */  i.health_q chrill i.bmi i.physical_act, dist(exponential) nohr

streg c.age##c.age female native i.ccaa i.size_mun i.marital_st /*
      */  i.health_q chrill i.bmi i.physical_act, dist(weibull) nohr
      
streg c.age##c.age female native i.ccaa i.size_mun i.marital_st /*
      */  i.health_q chrill i.bmi i.physical_act, dist(loglogistic)
      
streg c.age##c.age female native i.ccaa i.size_mun i.marital_st /*
      */  i.health_q chrill i.bmi i.physical_act, dist(lognormal)
      
streg c.age##c.age female native i.ccaa i.size_mun i.marital_st /*
      */  i.health_q chrill i.bmi i.physical_act, dist(ggamma)
      
streg c.age##c.age female native i.ccaa i.size_mun i.marital_st /*
      */  i.health_q chrill i.bmi i.physical_act, dist(gompertz) nohr
      
stcox c.age##c.age female native i.ccaa i.size_mun i.marital_st /*
      */  i.health_q chrill i.bmi i.physical_act, nohr      
      
estat phtest

[Javier Moreno]

Is the phtest specified correctly? And if it were (thus rejected) , should we use weibull or log logistic?