Merlin package

[giu ferrante]

I am assessing the merling package in Stata 14.2 for a "three-level survival model" and a "joint frailty model for recurrent and terminal events"

I get an error "initial values not feasible" for both datasets, therefore the package is not working.

Other commands such as stcox, shared (id) work fine

I would be very grateful to you if you could give me any hints on what I'm doing wrong or what I should do

Thanks

[Paul Dickman]

I'm far from a -merlin- expert, but if you haven't done so then read the help. There are a number of options for setting starting values (separate options for fixed effects, random effects, and ancillary parameters) and optimising the estimation (e.g., integration method and options specific to the integration method). For example, I believe -merlin- uses default starting values of 1 for random effects and variances of random effects, which may not be optimal. I would suggest giving more details of your problem and the exact command you tried so that those who have more knowledge than me might be tempted to make suggestions. I suggest mentioning which version you are using as there's both an SSC version and a development version. If you can get a simpler model to converge then it might be useful to show the estimates as this can give insight into which options to tweak.

[giu ferrante]

Thank you for your help
The commands are:
"joint frailty model for recurrent and terminal events"
merlin (TimeEvent1 x1 M1[id]@1, family(rp, df(3) failure(Event1))) (TimeEvent2 x1 M1[id], family(rp, df(1) failure(Event2)))


"three-level survival model"
merlin(TimeEvent1 x1#M1[Cluster]@1 M2[Cluster]@1 M3[Cluster>id]@1, family(rp, df(3) failure(Event1)))

I suspect there might be an issue with the database format , I am checking this again, including matrix

although I've already reshaped the initial dataset (one row one id) by using the command:
msset, id(id) states(Event1 Event2) times(TimeEvent1 TimeEvent2)