RE: Re: Using NonlinearFit/Regress?
- To: mathgroup at smc.vnet.net
- Subject: [mg33906] RE: [mg33876] Re: [mg33856] Using NonlinearFit/Regress?
- From: "DrBob" <majort at cox-internet.com>
- Date: Mon, 22 Apr 2002 00:57:44 -0400 (EDT)
- Reply-to: <drbob at bigfoot.com>
- Sender: owner-wri-mathgroup at wolfram.com
There's no way to tell if a model fits data without trying. It's not a matter of whether it fits, anyway --- the question is how poorly does it fit? (Models always fit badly.) Life would be so easy, if some book somewhere had all the answers, but it just isn't true. Bobby Treat -----Original Message----- From: Johannes Ludsteck To: mathgroup at smc.vnet.net [mailto:johannes.ludsteck at wiwi.uni-regensburg.de] Subject: [mg33906] [mg33876] Re: [mg33856] Using NonlinearFit/Regress? Dear njg, before trying anything else, you should check whether the model fits your data. If parameters show the false sign, this indicates that the model may not be identified or simply false. So please consult a introductory statistics or econometrics book before posting such questions. However, if you anyway have to use the resrictions anyway, you can do this, for example by applying a barrier constraint, for example NonlinearFit[x/(1+a (x-1) + a b x (x-1)) - If[a<=0||b<=0,Infinity,0.0],...] However, this will complicate the minimization and you have to provide two starting points for every parameter, for example {a, {0.3,0.6}},{b,{0.3,0.6}}. There are more sophistic methods to apply restrictions. You will find them in intermediate or advanced statistics or econometrics books. Best regards, Johannes On 20 Apr 2002, at 2:49, redrooz at yahoo.com wrote: > I am trying to estimate the parameters {a,b} in the > function: x / (1 + a (x-1) + a b x (x-1)) using > NonlinearFit[...]. > > The model is known to be a CONCAVE function with > 0 < a,b < 1 and b < a. NonlinearFit/Regress > always produces negative parameters (a discontinuous > function); even if I try to condition them to be > positive e.g., {a, 0, 1} and {b, 0, 1}. > > Any suggestions about using NonlinearFit[...] with the > above function? > > --njg > <><><><><><><><><><><><> Johannes Ludsteck Economics Department University of Regensburg Universitaetsstrasse 31 93053 Regensburg Phone +49/0941/943-2741