to sci.stat.edu and SR Millis -

On 29 Mar 2002 06:23:07 -0800, [EMAIL PROTECTED] (SR Millis)
wrote:

> Thanks to Roy St Laurent & Rich Ulrich for their advice. In follow-up:
> a reference for the Bradley-Blackwood procedure:
> 
> Bradley, E & Blackwood, L (1989). Comparing paired data: a simultaneous
> test of means and variances. Am Stat, 43, 234-235.

By the way, I want to thank you again for that citation of Lin's
concordance -
   "Lin has discussed the shortcomings of the t-test for assessing
   concordance between raters (Biometrics, 1989, 45, 255-268). Among
   other things, the paired t-test fails to detect poor agreement in 
  [ ... ] "

The article, and the measures, are better than I expected:  
this C_b  measure of accuracy  *does*  have a potential 
advantage over the t-test, since it weighs in those variances.
The article draws on some appropriate examples which are
different from the ones that I have ever been concerned with.

That is, there are examples with machine-measures, where 
the r's are (say) over 0.90.   I look at t-test and correlations
as important pieces of information in training or assessing 
human raters using subjective scales, with a focus on 
improving the positive information.  The article looks as
concordance as the single, summary measure for selection 
of an alternative, or approval of a device, with a focus on 
the combined errors, which are small.   So it is not just
editors (who I mentioned before) who need one measure,
instead of using two, separately.

So, I will try to keep Lin's coefficient in mind in the future.
And he provides a formula for multiple raters.
His measure of accuracy may be important by itself, too.

Further notes to myself:  check on the articles in google
about further developments;  pay attention to asymptotic 
behavior,  for smallest N and for dichotomous measures.

-- 
Rich Ulrich, [EMAIL PROTECTED]
http://www.pitt.edu/~wpilib/index.html
.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
.                  http://jse.stat.ncsu.edu/                    .
=================================================================

Reply via email to