On Wed, 13 Sep 2000 19:11:47 GMT, [EMAIL PROTECTED] wrote:
> I have been looking for an errors-in-variables regression to use for
> method comparison experiments. I have found a method by Williamson
> which seems to be the standard method for this problem. My question is
> how I test the hypothesis b=0. I have the slope and se as calculated
> in the Williamson paper, however no formula is given for computation of
> the hypothesis test. In standard linear regression I would normally
- As I understand it, the errors-in-variables regression removes
bias from the coefficients, but does not improve the test against the
null. So you can borrow your test from the OLS regression.
Compare: you use the simple test on the phi (or chi squared) for a
2x2 table, when you report the tetrachoric correlation.
> use a t-distribution but using Monte Carlo simulations it seems that
> using the t-distibution leads to a conservative rejection of the null
> hypothesis?
>
- I don't follow. Is this a reference to Williamson (cite the
literature?) ? Does that paper say that the test that I just
recommended, above, is too conservative?
--
Rich Ulrich, [EMAIL PROTECTED]
http://www.pitt.edu/~wpilib/index.html
=================================================================
Instructions for joining and leaving this list and remarks about
the problem of INAPPROPRIATE MESSAGES are available at
http://jse.stat.ncsu.edu/
=================================================================
