Survival analysis

[Robin Hofset]

Hello!

My partner and I are currently working on our master thesis. We are investigating survival of CEOs in family firms in Norway. We are, at the moment, struggling to figure out how to perform a good survival analysis.

the data set looks something like this:


ID of company------------------ID of CEO------------------Year ------------------------- -------> several other independent covariates relevant to test

---------1--------------------------------1-------------------------2000
---------1--------------------------------1-------------------------2001
---------1--------------------------------2-------------------------2002
---------2--------------------------------3-------------------------2000
---------2--------------------------------3-------------------------2001
---------2--------------------------------3-------------------------2002


Other information:

1) We have constructed a instrumental variable for when a change of CEO occurs (which is the failure variable)
2) We have both left and right censoring (as we do not know the tenure of each CEO before nor after the specified time (2000-2015)
3) We set the data as panel data using "xtset"
4) We set the data ready for survival using "stset"

Example of how the do file could look like:
------------------------------------------------------------------------------------------------------------
xtset pcid yr
stset yr, id(pcid) failure(SuccessionD==1)
stcox LargestFamCEO_15304
stdescribe
sts graph, hazard
estat phtest, detail
-------------------------------------------------------------------------------------------------------------

We have tried Cox model, where our covariates are significant. But fail the Cox proportional assumption (we use "estat phtest, detail").

Long story short for the data: CEO is the patient, and being related to the family owning the company is the "medicine" we want to test.

We would be so thankful if anyone could point us in the right direction. We have looked through Chuck Huber's stata videos of the subject, and the survival manual in Stata. We are still, however, unsure of whether we do things correctly or not.

[Marcos Almeida]

Welcome to the Stata Forum / Statalist.

Please read the FAQ, particularly on how to present data and command.

This is surely the best way to entice insightful replies.

That said, I wish to make two comments:

The Stata manual on SV models is a must read. There you will find good information.

In (quite) generally terms, and considering the model is "appropriate" in terms of measures and rational, the "issue" related to the proportional hazards may be deal with by using taking a further step and select an "extend" model of Cox regression.

To end, in spite of the paucity of information, I got the impression the time variable is discrete. Being this so, perhaps you could select an appropriate model for this as well.

Hopefully that helps.

[Carlo Lazzaro]

Robin:
as an aside to Marcos's helpful advice, most part of -st- machinery (included interval censoring) is covered in http://www.stata.com/bookstore/survi...-introduction/
-strata- option in -stcox- might be another fix to the seemingly lack of proportional hazard along the way.
If no fix applies to PH assumption fulfillment, you may want to take a look at https://www.ncbi.nlm.nih.gov/pubmed/15757960 (the second author, Svend Juul, was a very active Statalist member until some years ago and authored and coauthored two Stata press textbooks).

[Robin Hofset]

Thank you both for a quick and thorough response. My apologies for not presenting the data the correct way. This forum is surely amazing. We will consider extended Cox model and look into a model that is appropriate for discrete time.

[Carlo Lazzaro]

Robin:
you may be also interested in the following textbook: https://global.oup.com/academic/prod...cc=it&lang=en&