Hi,
I want to analyse the time a president or prime minister stay’s in office with a survival model and encountered some problems and hope that you could help me.
The basic idea is to use duration (time in office) as dependent variable and regress it on some covariates. I know the exact day a president or prime minister enters/leaves office, so I could use a continuous time model, but my covariates (economic growth, GDP, …) are all measured in years, hence I have to use a discrete time model. Considering the underlying continuous distribution I settled so far for the discrete time proportional hazard model using cloglog in stata, following Jenkins (1995).
Now to my problems: My approach is most likely inappropriate considering the nature of different political systems. The first problem I encountered are term limits. The fact that the time a president is in office is predetermined by the political system makes it necessary for me to correct for selection, e.g. with the Heckman selection model. Does anybody know a good paper where the Heckman selection model was used in a duration model or any other material (preferable with references to stata)? So I could figure out how to use it.
The second problem is that presidents or prime ministers can lose office because of different reasons (losing election, vote of no confidence, coup, retirement …), hence I should use a competing risk model for my analysis. I found in the material from Prof. Jenkins (https://www.iser.essex.ac.uk/resourc...sis-with-stata) that under the assumption that exit can only occur at the end of an interval that I could use a multinomial logit model. Unfortunately, the mentioned assumption does not hold in my case, a president could leave office before election and not only at the end of the term. Does anybody know how to estimate a competing risk model in this case with discrete time data?
Thanks
Michael
I want to analyse the time a president or prime minister stay’s in office with a survival model and encountered some problems and hope that you could help me.
The basic idea is to use duration (time in office) as dependent variable and regress it on some covariates. I know the exact day a president or prime minister enters/leaves office, so I could use a continuous time model, but my covariates (economic growth, GDP, …) are all measured in years, hence I have to use a discrete time model. Considering the underlying continuous distribution I settled so far for the discrete time proportional hazard model using cloglog in stata, following Jenkins (1995).
Now to my problems: My approach is most likely inappropriate considering the nature of different political systems. The first problem I encountered are term limits. The fact that the time a president is in office is predetermined by the political system makes it necessary for me to correct for selection, e.g. with the Heckman selection model. Does anybody know a good paper where the Heckman selection model was used in a duration model or any other material (preferable with references to stata)? So I could figure out how to use it.
The second problem is that presidents or prime ministers can lose office because of different reasons (losing election, vote of no confidence, coup, retirement …), hence I should use a competing risk model for my analysis. I found in the material from Prof. Jenkins (https://www.iser.essex.ac.uk/resourc...sis-with-stata) that under the assumption that exit can only occur at the end of an interval that I could use a multinomial logit model. Unfortunately, the mentioned assumption does not hold in my case, a president could leave office before election and not only at the end of the term. Does anybody know how to estimate a competing risk model in this case with discrete time data?
Thanks
Michael