On 8/25/06, Guido Heineck <[email protected]> wrote:
Jean,
thank you again for your comments.
You're welcome.
I was imprecise: I do not use predicted values from a regression but rather
calculate/generate them on basis of coefficients I have from another paper (and I am well
aware of the limitations of doing so).
In particular: I have self-reported height and weight and follow a procedure as proposed by
Cawley and Burkhauser (2006). They have validation data to correct for the self-reporting
bias and provide prediction equations which can be 'transported' to other data sets that
include only self-reported indicators, as in my case.
So, I just cannot use bootstrapping in the 'first step' since this is merely
addition/multiplication.
IMHO you don't need to bootstrap the multinomial logit either. I
cannot really see the use of it. What I would (try to) do in your case
is a kind of "sensitivity analysis":
(1) Compare the results from using the self-reported and the corrected
measures of height and weight
(2) Try to change to coefficients from this other paper, and look what
happens to your results.
(3) Try to find similar coefficients from another paper and look again
what happens.
All this is quite ad hoc, but can be interesting and instructive. But
maybe someone else will come up with something more elegant and/or
accurate.
jean
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
