how to judge the best of 2 fits...

Rob Learmonth rlearmon at metz.une.edu.au
Sun Jun 19 23:51:58 EST 1994


Some comments on curve fitting from a non-mathematical perspective.
One simple way to check the appropriateness of fit is to visually compare
fitted curves plotted against the original data.  Often you can get a good
impression of whether e.g. a 1 exponential or 2 exponential fits the curve 
better.

Another simple way is to produce plote of the (weighted) residuals (see
Genest & Wahl BBA 521, 502-9, 1978) and look at the scatter of points
and any local variations.

I have found that using the Chi-Squared statistic for evaluating goodness
of fit *not* to be helpful, especially for curves that approach Zero.

Some references I have found helpful include:
Motulsky & Ransnas (1987) FASEB J. 1, 365-374.
Noggle, J.H. "Practical Curve Fitting and Data Analysis" 1993,
Ellos Horwood - PTR Prentice Hall,  ISBN - 0-13-677394-X (bundle of
book plus software for curve fitting.  For review, see Biochemical
Education, 21(4), 210-211).

I hope these comments are helpful.

Rob Learmonth.
______________________________________________________________________________

Dr. Robert Learmonth					rlearmon at metz.une.edu.au
Department of Biochemistry, Microbiology & Nutrition
University of New England,				Phone: 61 67 73 2705
Armidale NSW 2351, Australia.				Fax: 61 67 72 8235
______________________________________________________________________________



More information about the Bioforum mailing list