On 8/27/08, Feiveson, Alan H. (JSC-SK311) <[email protected]> wrote:
> I don't understand your point about causality. It seems to me that
> causality has a lot to do with the intrepretation of model error terms -
> which would affect estimation - but yes, my question was mostly about
> nomenclature - thanks for your suggestion about checking out
> "Measurement Error in Nonlinear Models" .
Well that was a point badly taken, I admit. What I thought was
important is to establish what's observable and what's not. If it is
unobservable, call it a latent variable. If it is observable measured
with error, build a measurement error model around it. But the
instrumental variable is a third variable that's available to you
somehow -- economists sometimes have to come up with instruments out
of thin air, while structural equation modelers in other social
sciences may have rigorous ways of finding those instruments. In your
case, you have your thresholded variable as an instrument, although
arguably that's not such a great instrument since it requires its own
measurement error model to be built around it.
You can think of the causal links going this way or that way between
your "ideal" variable and your observed variable measured with error,
and you get two different models -- classic and Berkson.
--
Stas Kolenikov, also found at http://stas.kolenikov.name
Small print: I use this email account for mailing lists only.
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
