xtset - repeated time values within panel error

[Rose Matthews]

Hi why does stata return the error when i run this code

sort ptid year2
xtset ptid year2

//ERROR: repeated time values within panel

using the following dataset:

clear
input float(ptid state year2)
1 1 2003
1 2 .
1 3 .
2 1 2003
2 3 2018
2 2 .
3 1 2003
end label values state q1
label def q1 1 "primary", modify
label def q1 2 "revised", modify
label def q1 3 "dead", modify

As you can see I have successfully run the code here (post 7)using fake data... and I do have repeated years in my variables found in this post, but I didn't obtain the error obtained above. May you please let me know what I'm doing wrong?

https://www.statalist.org/forums/for...n-var-f-status

Update:
it turns out I have around 12 people who died and had the operation on the same day (using the full ddmmyy)

I wouldn't like to lose this data by just dropping them....as this shows a procedure may be high risk

How do I address the problem to allow me to use -xtset-

[George Taylor]

The issue here is that you have repeated years *within* your ID variable - 3 observations have state 1 and year 2003. ID-year pairs need to be unique, as was the case in your other example.

[Carlo Lazzaro]

Rose:
as an aside to George's hepful reply, to avoid -repeated time values within panel- error, just -xtset- your data with -panelid- only, keeping in mind that, while this fix makes time-series operator unvailable, it still allows you to plug -i.year- as a predictor in the right-hand side of your regression equation.

[Rose Matthews]

thanks for this, however,

I can't just

xtset ptid 

///if I do not xtset the year that this means I can't use 

tsfill 

//error: time variable not set

The aim to generate transition probabilities for treatment and control as you can see in the post here:
https://www.statalist.org/forums/for...n-var-f-status

Which I generated successfully using dummy data and with Clyde's insight.

As seen in post 1 of this thread (update at 1100) I have 42 observations in the treatment that were operated on and died on precisely the same day using the format DD MM YY

I can either drop these.... which I wouldn't like to as this may indicate that treatment is high risk...but I'm not sure what else to do....

[Carlo Lazzaro]

Rose:
why not considering survival analysis, then to calulate transition probability (e.g., Weibull regression, if you have a set of other predictors), as reported in: Briggs A, Sculpher M, Claxton K. Decision modelling for health economic evaluation. Oxford: Oxford University Press; 2006:46-56?.

[Rose Matthews]

I have done survival analysis several time before….
but how would this be used to generate transition probabilities if I have 2 failures (death or revised) . And to be honest I haven’t seen this reported on statalist being used to create transition probabilities

would you be able to give me an eg how you would do this?