Fixed effects (in Accelerated Failure Time Survival Model), Cross-sectional

[MIchael Jefferson]

Hi,

For my research I am examining private equity strategies and their subsequent exit-types. My data is Stset in the following way:

stset E_Date, failure(Successful==1) id(Strategy_Number) enter(time P_Date) origin(time P_Date)

                id:  Strategy_Number
     failure event:  Successful == 1
obs. time interval:  (E_Date[_n-1], E_Date]
 enter on or after:  time P_Date
 exit on or before:  failure
    t for analysis:  (time-origin)
            origin:  time P_Date

------------------------------------------------------------------------------
      1,197  total observations
          0  exclusions
------------------------------------------------------------------------------
      1,197  observations remaining, representing
      1,197  subjects
        251  failures in single-failure-per-subject data
  3,031,231  total analysis time at risk and under observation
                                                at risk from t =         0
                                     earliest observed entry t =         0
                                          last observed exit t =     8,216

For every ID I have one observation moment, event moment, the moment of exit (E_Date). The variables are observed only at one moment and are therefore static.
However, I want to control for fixed effects due to the type of industry, the year of exit, and the country of the firm. How should I proceed with my analysis? I have considered the xtstreg, but this is not applicable to my dataset, as it does not involve panel data, am I correct? When using the

xi

command for the three fixed effect variables,

i.CountryFE i.YearFE i.IndustryFE

the accelerated failure time (AFT) models are indicating 'cannot compute an improvement, discontinuous region encountered' and 'hessian is not negative semidefinite'. I hope someone is able to offer me some guidance. Including the three variables as regular variables in the parametric analysis would not be appropriate for my research?

Thanks in advance,

Michael

[Clyde Schechter]

I'm not certain this is why your model is not converging, but it seems to me that trying to include a fixed effect for the year of exit is almost the same as including the time-to-exit outcome variable as a covariate in the model. That doesn't make sense, and I would imagine it makes the likelihood function pretty bizarre, and quite possibly non-convergent.

As an aside: unless you are using an ancient version of Stata, there is no reason to use xi: here. Use factor-variable notation instead. It's easier and opens up the wonderful world of -margins- to you. It also gives you more readable regression outputs. See -help fvvarlist- for details.

[MIchael Jefferson]

Thank you very much Clyde!
Another problem has presented itself while running the log-logistic AFT model. I want to use the

ibn.

command for some categorical variables in my analysis, while also suppressing the constant term (I read this is necessary). Without doing so, just by using

i.

, and one category is obviously omitted, there are no running problems. Whereas while using the

ibn.

command and suppressing the constant term, the iteration process is never-ending. Do you have any suggestion on how to overcome this problem?
Should I open a new thread for this question?

[Clyde Schechter]

Yes, this should be a new thread. And when you do re-post, you should show the code and outputs for the convergent model with i., as well as the code you are using with ibn.

[MIchael Jefferson]

Hi Clyde,

I'm sorry I haven't replied earlier. I have started a new thread under:
https://www.statalist.org/forums/for...t-i-versus-ibn

Thanks for all your help.