[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
Re: st: RE: stset with multiple spells
Mark and Juan Manuel,
I also rarely work with survival models, but I'm guessing that the
"Multiple-record-per-subject survival data" discussed in the manuals
is a rather large superset of the "Multiple-failure-per-subject
survival data" that Juan Manual wants to analyze. I liked the
suggestion of treating each "new" failure as a new "patient" and
allowing the "frailty" of these patients to be correlated, since they
are in fact multiple observations of time periods within the same
country (though note that a country is even less the same entity over
time than is a person). What was _wrong_ with that approach?
Another method would be to look at the number of episodes in a given
time frame with a count data model (-poisson- or -xtpoisson-). Still
another method might redefine the problem in a larger way... it
strikes me that you have likely defined "periods of financial
distress" using economic indicators, and then chosen cutoffs for
levels of these indicators using some theory. Am I right? Then it
seems to me that these economic indicators should be the Y variables
whose movements you wish to model--which might argue for an -xtreg-
panel model, possibly using -suest- if you have multiple Ys.
Now speculating wildly,
On 4/24/06, Schaffer, Mark E <M.E.Schaffer@hw.ac.uk> wrote:
> Juan Manuel,
> > -----Original Message-----
> > From: email@example.com
> > [mailto:firstname.lastname@example.org] On Behalf Of
> > Juan Manuel Jauregui
> > Sent: 24 April 2006 03:32
> > To: email@example.com
> > Subject: st: stset with multiple spells
> > I have an unbalanced panel of almost 200 individuals
> > (countries in this case) with 100 records each (25 years of
> > quarterly data) and I want to study the duration of some
> > periods of financial distress. I have an event that marks the
> > begining of each crisis, and another event that marks its
> > end. This crises periods appear more than once for each
> > country and they are separated from each other by periods of
> > tranquility, that is, analysis time should begin from 0 when
> > a new crisis begins.
> I rarely work with duration data, so this is possibly a silly question,
> but -help stset- has a heading near the top that reads
> "Multiple-record-per-subject survival data"
> and later in the help file, explains how to handle it.
> How is this different from what you want to do?
> > The closest thing I think I can do is to take each episode as
> > an individual and take countries as groups and use a shared
> > frialty model, but that wouldn't be exactly what I want to
> > do. I'd prefer to take it as repeated episodes of the same individual.
* For searches and help try: