appropriate to use event history analysis?

[Thomas Moliterno]

Hello ... I am working on a research project and the reviewers have suggested that event history analysis may be an appropriate modeling technique. I'm not so sure, but it's also been over 20 years since I've run event history models in Stata, so I thought I'd seek some input before I invest the time in structuring the dataset (up to now we have been running xtgee models). So I'll ask my general question first, and then if event history *is* a good modelling technique I think I'll have some follow-up data structure questions.

We have population level career history data on individuals in a particular profession:

• when a person they started and all the organizations they worked for
• the positions they held in the organizations
• organization-level for the organizations they worked (i.e., performance data like profits, sales, etc.)
• each observation is one-person year
• we only have complete career histories (no left or right censored career histories): so a person's first record in the data is their first job, and their last is their last job

Our DV is mobility: whether from one organization to another. We are interested theoretically in whether aspects of an individuals prior career history associate with mobility. In particular:

• the point their career at which they are *first* promoted to a managerial role
• the number of years they served in a managerial role at the organization they worked for immediately prior to the mobility event

Example:

+--------------------------------------------------------------------------+
| id year teamid mobility hlevel CnY_1M~d TnY_Mgr ln_pop~w |
|--------------------------------------------------------------------------|
5868. | 1323 1993 2 0 3 0 0 15.75831 |
5869. | 1323 1994 2 0 3 0 0 15.78362 |
5870. | 1323 1995 2 0 3 0 0 15.80727 |
5871. | 1323 1996 2 0 3 0 0 15.83056 |
5872. | 1323 1997 2 0 3 0 0 15.85479 |
|--------------------------------------------------------------------------|
5873. | 1323 1998 2 0 3 0 0 15.87775 |
5874. | 1323 1999 2 0 4 7 1 15.90068 |
5875. | 1323 2000 2 0 4 7 2 15.92297 |
5876. | 1323 2001 2 0 4 7 3 15.941 |
5877. | 1323 2002 2 1 4 7 4 15.95655 |
|--------------------------------------------------------------------------|
5878. | 1323 2003 14 0 6 7 1 17.37807 |
5879. | 1323 2004 14 0 6 7 2 17.38714 |
5880. | 1323 2005 14 1 6 7 3 17.39424 |
5881. | 1323 2006 2 0 6 7 5 16.0299 |

• This individual (id==1323) worked for organization 2 (teamid==2) starting in 1993 and was promoted to a managerial role (i.e., any hlevel>=4) in the 7th year of their career. So the time to their first managerial role is 7 (CnY_1M~d==7)and keeps that value for all subsequent observations (that becomes an attribute of their prior career history that they carry with them for the remainder of their career).
• They then work in that managerial role for 4 years, with TnY_Mgr counting that experience.
• In 2002 they experience a mobility event (mobility==1): at this point they have had 4 years tenure in that managerial role (TnY_Mgr==4).
• They move to organization 14, also in a managerial role (hlevel==7)and so the count variable TnY_Mgr resets to 1 in 2005
• In 2005 there is another mobility event (back to organization 2). Since they are returning to an organization where they worked before, TnY_Mgr resumes the count at 5.
• ln_pop is one of our time variant organization-level attributes (here, population of the city where the organization is based)

Another example:

+--------------------------------------------------------------------------+
| id year teamid mobility hlevel CnY_1M~d TnY_Mgr ln_pop |
|--------------------------------------------------------------------------|
21090. | 4669 1990 25 0 2 0 0 15.40537 |
21091. | 4669 1991 25 0 3 0 0 15.43006 |
21092. | 4669 1992 25 0 3 0 0 15.45659 |
21093. | 4669 1993 25 0 3 0 0 15.47922 |
21094. | 4669 1994 25 0 3 0 0 15.4973 |
|--------------------------------------------------------------------------|
21095. | 4669 1995 25 0 3 0 0 15.5168 |
21096. | 4669 1996 25 1 3 0 0 15.53286 |
21097. | 4669 1997 19 0 3 0 0 16.74171 |
21098. | 4669 1998 19 0 3 0 0 16.74702 |
21099. | 4669 1999 19 0 3 0 0 16.75376 |
|--------------------------------------------------------------------------|
21100. | 4669 2000 19 1 3 0 0 16.76004 |
21101. | 4669 2001 6 0 3 0 0 16.34031 |
21102. | 4669 2002 6 0 3 0 0 16.34328 |
21103. | 4669 2003 6 0 3 0 0 16.34571 |
21104. | 4669 2004 6 0 3 0 0 16.3484

• This person was never promoted to a managerial role (all observations of hlevel<4). So CnY_1M~d and TnY_Mgr take values of 0 for their whole career (in an alternative specification we set CnY_1M~d==. since they were not *instantly* promoted to manager, but that's a separate matter).
• They have mobility events in 1996 and 2000

OK ... so up to this point we have been modelling (this is an abbreviated model without all controls, etc., just to illustrate)

xtgee mobility ln_pop CnY_1Mgr TnY_Mgr i.teamid i.year, family (binomial 1) link(logit) vce(robust)

That is, does the point in time at which they first became a manager and the amount of time they held a managerial role at the immediately preceding organization associate with the mobility event.

The reviewers have suggest that we model this with EHA: the time to the mobility event, with the "waiting time" being 1) the number of years it took be promoted to a managerial role and 2) the number of years they served in a managerial role. I'm not so sure because we not saying that everyone who experiences the event "waits" in the same way: in the examples above id==4669 experiences the event in year 7 of their career and id==1323 in year 10 ... OK, but we aren't interested in how long it takes them to experience that event in career years. Instead, we are interested in how id==1323 experienced the event when their resume showed that they had been promoted to a managerial role in year 7, and then had 4 years of tenure in that role (which just happened to be in year 10 of their career)... and in contrast id==4669 experienced that event when their resume showed that they had NEVER been promoted to a managerial role (which just happened to be in year 7 of their career).

Many thanks for any help (and if you've gotten to this point, for your time reading through).

Tom