[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

From |
"Simon Oertel" <Sim.Oertel@t-online.de> |

To |
<statalist@hsphsun2.harvard.edu> |

Subject |
st: Event history/Survival analysis_how to introduce time varying variables |

Date |
Mon, 19 Nov 2007 16:18:13 +0100 |

Dear Stata Users, I need some advice concerning survival analysis. I would like to analyze the effect of CEO exits on the mortality rate of organizations. My problem is that I am not sure about the way how to introduce the exit of a CEO since this variable is not time constant. Following the book of Blossfeld/Golsch/Rohwer (Event History Analysis with Stata), there are three basic approaches to analysis such data: 1.) by using piecewise constant exponential model, 2.) by applying the method of episode splitting in parametric and semiparametric transition rate models, 3.) by specifying the distributional form of the time-dependence and directly estimating its parameters using the ML method. I first used the second approach. I split (if a CEO exit has occurred in the organization) the "living time" of each organization in two episodes. The first episode starts from the time of foundation and ends at the exit-time of the CEO. The second spell starts from the time of the CEO exit and ends at the closing day of the organization. Then I used parametric/semiparametric models to analyse the influence of the CEO exit on the risk of an organizational closure. However, I would like to compare these results with the results of a piecewise constant exponential model. Following examples in the basic literature, I used "stsplit time, at (0, 365, .)" and "tab time, ge(time)" for splitting the data into yearly records and defining time dummies. Then I generated period-specific dummies. For example: "for any branch branch . \ any time1 time2 .: gen branchtimeY = X*Y". However, I am not sure if I can use the same command for modelling period-specific dummies concerning CEO exits. In the dataset I have two variables for the CEO exit. One dummy, which tells my, if a CEO exit has occurred in the organization and a second one which tells me the exactly exit-date. Is it right, to use the date of the CEO exit (lets say the CEO of organization1 would exits 725 days after the foundation of the organization) for defining the period-specific dummies? This would mean that the command would be for example: "for any CEO_exit_time CEO_exit_time . \ any time1 time2 .: gen CEO_exit_timetimeY = X*Y", right? Since my results of this analysis-approach are totally different from the one of the parametric/semiparametric transition rate models, I suggest that I make some failure (The results of the piecewise constant exp. model show a very weak, positive influence of CEO exits on the mortality risk of orgnizations, while the parametric models show a strong negativ effect of CEO exits on the mortality risk of organizations.) Or is it possible that the results are different, because the case number is too small for a piecewise constant exponential model (around 1200 exits in 15000 organizations in 15 years)? Also if I suggest that this question is very simple, I will really appreciate any answer. Thanks for your help Simon * * For searches and help try: * http://www.stata.com/support/faqs/res/findit.html * http://www.stata.com/support/statalist/faq * http://www.ats.ucla.edu/stat/stata/

- Prev by Date:
**Re: st: RE: simulated data for logistic regression... remedial algebra help?** - Next by Date:
**Re: st: problem with mvprobit** - Previous by thread:
**st: problem with mvprobit** - Next by thread:
**st: ologit data handling** - Index(es):

© Copyright 1996–2016 StataCorp LP | Terms of use | Privacy | Contact us | What's new | Site index |