Survival analysis using panel data

[Joel de Bruijn]

Dear sir, madam,

I have a question regarding certain data management challenges that would allow me to perform a survival analysis using panel data from the Panel Study of Income Dynamics. I am analyzing the period leading up to the decision for an individual to become an entrepreneur.

To be able to perform a survival analysis (to stset the data) I need to create a spell length variable which I want to do on the basis of the data available to me. The commands that I have been using are as follows:

generate BusStart = 0
bysort ID (year): replace BusStart=1 if (BusOwn[_n]==1 & BusOwn[_n-1]==0)

BusOwn is a dummy variable that equals 1 if an individual owns a business at year n. My aim is to only analyse the first business start of any individual. The commands that I have been using however create a dummy that is 1 whenever a person owns a business in period n after not having owned a business in period n-1. Therefor it is possible that a single person has more than one period where BusStart == 1, and indeed it does occur.

ID	BusStart	BusOwn	year
5170	1	1	1996
5170	0	1	1997
5170	0	0	1999
5170	1	1	2001
5170	0	0	2003

Is there an easy way to create a dummy variable (BusStart) that only equals 1 at time n if all previous periods of BusOwn equal zero?

ID	BusStart	BusOwn	year
5170	1	1	1996
5170	0	1	1997
5170	0	0	1999
5170	0	1	2001
5170	0	0	2003

Thanks in advance.

Kind regards,
Joel

[Stephen Jenkins]

search tsspell

[Joel de Bruijn]

Hi Stephen,

Thank you for your reply. I have posted the example below again in the proprietary format, apologies for not doing that earlier. I will give tsspell (and tsset) a try.

* Example generated by -dataex-. To install: ssc install dataex
clear
input float(ID BusStart BusOwn) int year
5170 0 0 1985
5170 0 0 1986
5170 0 0 1987
5170 0 0 1988
5170 0 0 1989
5170 0 0 1990
5170 0 0 1991
5170 0 0 1992
5170 0 0 1993
5170 0 0 1994
5170 0 0 1995
5170 1 1 1996
5170 0 1 1997
5170 0 0 1999
5170 1 1 2001
5170 0 0 2003
5170 0 0 2005
5170 0 0 2007
5170 0 0 2009
5170 0 0 2011
5170 0 0 2013
5170 0 0 2015
end

Kind regards,
Joël

[Joel de Bruijn]

Hi Stephen, Statalist members,

Thank you for your suggestion:

search tsspell

I have been working with the tsspell command but I am running into a few issues. My aim is to define episode's (or spells) as starting at age 25 and ending as soon as a business is owned for the first time. To achieve this I attempted to use the command below.

tsspell, cond(Age > 25 & BusOwn == 0)

My concern is that I haven't defined my command ensuring that only one spell is created per ID. Individual 5170 now has three spells, I only want to keep the first spell.

* Example generated by -dataex-. To install: ssc install dataex
clear
input float ID byte(_seq _spell _end) float BusOwn int year float Age
5170 0 0 0 0 1985 20
5170 0 0 0 0 1986 21
5170 0 0 0 0 1987 22
5170 0 0 0 0 1988 23
5170 0 0 0 0 1989 24
5170 0 0 0 0 1990 25
5170 1 1 0 0 1991 26
5170 2 1 0 0 1992 27
5170 3 1 0 0 1993 28
5170 4 1 0 0 1994 29
5170 5 1 1 0 1995 30
5170 0 0 0 1 1996 31
5170 0 0 0 1 1997 32
5170 1 2 1 0 1999 34
5170 0 0 0 1 2001 36
5170 1 3 0 0 2003 38
5170 2 3 0 0 2005 40
5170 3 3 0 0 2007 42
5170 4 3 0 0 2009 44
5170 5 3 0 0 2011 46
5170 6 3 0 0 2013 48
5170 7 3 1 0 2015 50
end

Is there a (better) way to define the command to measure spells that I am after, starting at age 25 and ending when an individual owns a business?

Thanks.

Kind regards,

Joël de Bruijn

PS. An earlier version of this message concerned a problem with spells not ending at BusOwn != 0, however, this issue has been solved by using & instead of if.