[Date Index] [Thread Index] [Author Index]

Differing curvature measures in NonlinearRegress

• To: mathgroup at smc.vnet.net
• Subject: [mg90253] Differing curvature measures in NonlinearRegress
• From: "Gy. Csanády" <csanady at helmholtz-muenchen.de>
• Date: Thu, 3 Jul 2008 06:13:21 -0400 (EDT)

Dear Madam Dear Sir;
I have a question related to the NonlinearRegress-procedure where 
Mathematica calculates the FitCurvatureTable. In this table the 
values of =93Max Intrinsic=94,  =93Max Parameter-Effects=94 and of =9395% 
confidence region=94 are given.
For a statistical comparison the intrinsic and parametric curvature 
should be compared to the curvature over the 95% confidence region, 
which is 1/(2*Sqrt(F(p,n-p,a)), where F is the corresponding F-
distribution at a significance level a, with  p the number of 
parameters and n the number of data points (see e.g. Ratkowsky (1983) 
Nonlinear regression modeling).
Mathematica seems to give only the radius of curvature (1/(Sqrt(F(p,n-
p,a)) which cannot be compared directly with the other curvature 
measures listed in the FitCurvatureTable.
A clarification would be appreciated.
With best regards
Gy=F6rgy Csan=E1dy