From: John Philo <jphilo@mailway.com>
  To  : Chandra Boon <boon@uhnres.utoronto.ca>
  Date: Thu, 26 Apr 2001 18:10:18 -0700

RE: Weighted vs nonweighted fits

Chandra,

A couple of additional points to add to Olin's response:

1) I think one thing that is confusing you is that when you do a weighted
fit in the Beckman software, what is returned under the label 'variance' is
not actually the variance. I believe the number displayed there is the
reduced chi squared, as would be appropriate to characterize the quality of
a weighted fit. The actual variance of the weighted fit will always be >=
that of the unweighted fit.

2) For interference data when you ask for a weighted fit all the data are
weighted equally (it's really a non-weighted fit) and what is returned under
the variance heading is really the variance. The same applies if you fit
absorbance data that were taken using only a single replicate.

3) Yes, there will be a difference in the best fit parameters for weighted
versus non-weighted fits. In general the values from the unweighted fit
'should' fall within the confidence interval for the weighted fit (i.e. the
difference between the two fits would not be significant), but this is
certainly not always true.

John Philo
Alliance Protein Laboratories
www.ap-lab.com

-----Original Message-----
From: Chandra Boon [boon@uhnres.utoronto.ca]">mailto:boon@uhnres.utoronto.ca]
Sent: Thursday, April 26, 2001 11:50 AM
To: rasmb@alpha.bbri.org
Subject: Weighted vs nonweighted fits


Hello.  I am using the Beckman software (Origin) to analyse some
sedimentation equilibrium data.  I obtained fairly different results (for
Ka and the variance) when using the weighted vs the nonweighted fit for
the same model.  What is the difference between the two fits and which
one is more reliable?  Thanks.

Chandra Boon
Dept. of Medical Biophysics
University of Toronto