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: Random thresholds in ordered logit/probit in -gllamm-

From   Maarten buis <[email protected]>
To   [email protected]
Subject   Re: st: Random thresholds in ordered logit/probit in -gllamm-
Date   Sun, 28 Nov 2010 16:26:24 +0000 (GMT)

--- On Sun, 28/11/10, Kristian Karlson wrote:
> Stas, I hope it is okay with one more
> remark on the random thresholds in ordered probit/logit. I
> came across the following paper by sociologists Breen and
> Luijkx,
> They estimate a model with heterogeneous thresholds. They
> present a latent class ordered logit model with
> heterogeneous thresholds and locations

You can see an ordered logit model as a set of logit models that
model the odds getting more than that category versus that 
category and all lower categories, whereby the coefficients are
constrained to be equal across the models (except the constant, 
which is modeled by the thresholds). Maybe you can trick Stata 
into estimating your model by estimating your model by way of 
these separate logits. You would loose the power obtained by the 
assumption that all the effects are constant across models, but 
that assumption is often doubtful anyhow.

Than again, such models are really not my taste. They tend to 
move too far away from the data, and you tend to end up in a
no-win-situattion. If the model confirms what we know from 
simpler models than all is fine (but nothing is won), and if 
they do not confirm what we already knew than we do not know
whether the difference is due to some strong assumptions we had
to impose to be able to estimate the model or whether this is a
real new find, so in that case we have not won anything either.

Hope this helps,

Maarten L. Buis
Institut fuer Soziologie
Universitaet Tuebingen
Wilhelmstrasse 36
72074 Tuebingen


*   For searches and help try:

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