Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: Decomposition method for discrete-time event-history models?

From   Austin Nichols <[email protected]>
To   [email protected]
Subject   Re: st: Decomposition method for discrete-time event-history models?
Date   Mon, 12 Sep 2011 06:08:08 -0400

Timo Kauppinen <[email protected]>:
I would guess not.  If you compute an average all periods at risk,
people appearing for different numbers of periods, with different
time-varying characteristics in each period, I'm not sure what
population quantity you want to estimate as the difference in averages
(marginal effects) to decompose.  I would think you could use the
first period at risk for each individual and use the generalizations
of Oaxaca (ssc desc oaxaca) on that one period, but that is not
exactly the same exercise as what you asked for.  If you want effects
on duration, I see a simulation in your future.

Maybe if you clarify what you mean by "a difference in the probability
of entry to home-ownership between two groups" you will get better
answers.  This is measuring the differences in X and in the coefs on X
for two groups to look for evidence of discrimination, e.g. in
black-white differences in homeownership?  You would also want to look
for evidence of other kinds of discrimination, e.g. in black-white
differences in predatory lending and subsequent default rates.

On Mon, Sep 12, 2011 at 4:51 AM, Timo Kauppinen <[email protected]> wrote:
> Thanks for the answers. I am sorry for the brevity of my original question, I wrote it with my mobile phone.
> I am aware of decomposition methods for logit models in Stata (such as the fairlie and nldecompose packages), but I was not sure, if there is a problem when these are applied to person-period data (when doing survival/event-history analysis) instead of to data that has only one data line for each person. But I guess that can be done, then.
> Timo Kauppinen Department of Social Research Sociology FI-20014
> University of Turku FINLAND
> ------------------------------
> Date: Thu, 8 Sep 2011 18:21:56 -0400
> From: =?ISO-8859-1?Q?Jorge_Eduardo_P=E9rez_P=E9rez?= <[email protected]>
> Subject: Re: st: Decomposition method for discrete-time event-history models?
> What you call discrete-time event-history models is known in economics
> and other fields as survival analysis, and what you call Fairlie's
> method is actually an extension of what labor economists know as the
> Blinder-Oaxaca decomposition. It is important to provide full
> references when writing to Statalist, so people in other fields can
> understand your problem.
> Now, survival models can be seen as a non-linear model, so the
> extension of Blinder-Oaxaca to non linear models could be used. The
> theory behind this approach is in this presentation and the references
> therein:
> An extension of the Blinder-Oaxaca decomposition for survival models
> is given in this paper.
> Hope this helps,
> _______________________
> Jorge Eduardo P??ez P??ez
> On Thu, Sep 8, 2011 at 5:25 PM, Bryan Sayer <[email protected]> wrote:
>> I am not familiar with Farlie, but you might take a look at predictive
>> margins. ?You can use the difference of the predicted margin between
>> groups to do lots of things.
>> Bryan Sayer
>> Monday to Friday, 8:30 to 5:00
>> Phone: (614) 442-7369
>> FAX: ?(614) 442-7329
>> [email protected]
>> On 9/8/2011 5:12 PM, Timo Kauppinen wrote:
>>> Hello,
>>> I have a simple question for which I haven't been able to find an
>>> answer, though. Is there a decomposition method similar to Fairlie's
>>> method (for example) that could be used with person-period data and a
>>> discrete-time event-history (logit) model? I guess that it is not OK
>>> just to use fairlie, given the data structure? The groups to be
>>> compared have unequal sizes. The gap to be decomposed is a difference
>>> in the probability of entry to home-ownership between two groups.

*   For searches and help try:

© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index