Dear Statalist users,
I am applying a parametric survival (i.e. duration) model to a sample of individuals adopting a farming technology. However, I have problems choosing the distribution option because my hazard function exhibits a bump (i.e. increase in risk) by year 5 (i.e. 2005) that prevents the curve from looking like the smooth distribution curves presented in influential duration modelling papers (i.e. Kiefer, 1985). I do have documented information of a technology promotion program carried out in the year 2005, so perhaps I could model that year as a dummy variable...
My hazard function graph looks as follows:
Given that my survival function graph exhibits an asymptote, I've decided to go with the CUREREGR command in Stata 15.1 so as to apply a split-population (i.e. cure) regression model. I've run CUREREGR with different combinations of class, link and distribution options. I consistently get smaller AIC and BIC values for a non-mixture, lml, exponential distribution model.
Am I missing something here? The exponential distribution is supposed to be a straight line. But I also saw an example of how Kiefer (1985) treats drastic variations in risk as smooth lines and therefore exponential distribution makes sense.
Any thoughts would be much appreciated. Jesus Pulido.
I am applying a parametric survival (i.e. duration) model to a sample of individuals adopting a farming technology. However, I have problems choosing the distribution option because my hazard function exhibits a bump (i.e. increase in risk) by year 5 (i.e. 2005) that prevents the curve from looking like the smooth distribution curves presented in influential duration modelling papers (i.e. Kiefer, 1985). I do have documented information of a technology promotion program carried out in the year 2005, so perhaps I could model that year as a dummy variable...
My hazard function graph looks as follows:
Given that my survival function graph exhibits an asymptote, I've decided to go with the CUREREGR command in Stata 15.1 so as to apply a split-population (i.e. cure) regression model. I've run CUREREGR with different combinations of class, link and distribution options. I consistently get smaller AIC and BIC values for a non-mixture, lml, exponential distribution model.
Am I missing something here? The exponential distribution is supposed to be a straight line. But I also saw an example of how Kiefer (1985) treats drastic variations in risk as smooth lines and therefore exponential distribution makes sense.
Any thoughts would be much appreciated. Jesus Pulido.
