Survival Analysis Help Needed

[pandc]

Hi all,

I am kindly seeking help regarding the Stata implementation of my Political Science research. I will try to be as concise as I can.

I am using a Cox hazard model to examine whether the hazard of political turnover (change in government) is affected by banking crises.

Therefore I believe I require an extended Cox model with the time-dependent covariate BankCrisis. As such, my variables are:

1) Spell: An ID for the government spell
2) Survival: The number of years a spell has survived before turnover occurs
3) Turnover: A dummy for whether turnover occurred (my data is uncensored so this is always a 1)
4) BankCrisis: The number of banking crises the country experienced during the spell
5) Controls

I will work with a fake dataset to illustrate. I begin with a dataset like this:

Spell	Survival	BankCrisis	Turnover
1	4	1	1
2	6	1	1
3	7	2	1
4	2	1	1

I then expand the dataset to allow for BankCrisis to be time-varying:

Spell	Survival	BankCrisis	Turnover
1	1	0	1
1	2	0	1
1	3	0	1
1	4	1	1
2	1	0	1
2	2	0	1
2	3	0	1
2	4	0	1
2	5	0	1
2	6	1	1
3	1	0	1
3	2	0	1
3	3	1	1
3	4	1	1
3	5	1	1
3	6	1	1
3	7	2	1
4	1	0	1
4	2	1	1

In my fake dataset, I have made it that if a government experiences a banking crisis there is a very strong chance its spell will end (see above).

I then stset my data: stset survival , f(turnover) id(spell)

Then I run my Cox regression: stcox bankcrisis, nohr

This does not work for me... I get insignificant results, even though I hardwired the data in this case to get an extreme positive correlation between the experience of a banking crisis and the experience of a turnover following the crisis.

I have spent weeks online and in books trying to understand where I am going wrong. I would greatly appreciate it if someone could take the time to explain whether a) I am approaching the problem correctly, and b) if I am, why this is not working.

Many Thanks,

Ben

[Christophe Kolodziejczyk]

Hi Ben For those observations where a failure is observed turnover should be 0 for the values of survival which are lower than the observed duration. For the first spell for example, turnover should be zero for values of survival is less than 4. Hope this helps. Best Christophe

[pandc]

Thank you very much for this Christophe, that sure does help. If there is any other flaws/comments/improvements people recommend, I would greatly appreciate it. Many thanks