censoring-failure-survival

[turan keles]

Dear friends,

I am examining a survival analysis regarding tax recovery speed of post-conflict states. In my paper, I basically created a dummy for recovery and coded 0 if a state reaches its pre-conflict tax levels and coded 1 if the state could not reach the pre-war tax levels. Additionally, I created a variable to code within how many years did the state reached its pre-conflict levels. For example, state A has a civil conflict in 1980 and if state A reaches its 1979 tax levels within the available data time interval, I coded 0 and if not I coded 1. If the state could reach its 1979 tax levels in 1990 I coded durationtime as 10. As to command in stata I could not be sure which type stse to use. For the relevant issue should I use "stset durationtime, failure(censored)" or "stset durationtime, failure(censored==0)" ?
Thanks for your help!
Regards

[Clyde Schechter]

If you use -failure(recovery)-, Stata will interpret that to mean that the observation reflects a return to pre-conflict tax-levels if recovery!= 0, and a censored observation (never achieved pre-conflict tax-levels during the observation period) when recovery == 0. But you coded your variable recovery in exactly the opposite way.

Given the way you constructed the variable you call recovery, you need to use -failure(recovery == 0)-.

By the way you don't say how you set the variable durationtime for those entities where pre-conflict tax levels were never regained during the period of observation. If you just left it missing, then you will get wrong answers from your analyses, because they will be omitted from analysis rather than being treated as censored observations. You need, instead, to provide a value for durationtime which equals the last duration from the conflict at which they were observed and still did not achieve pre-conflict tax levels.