Message-ID: <009a01c9174c$aef3dae0$50111481@PR017080>
Date: 2008-09-15T16:04:07Z
From: Benoit Boulinguiez
Subject: Question abour nonlinear regressions
In-Reply-To: <Pine.LNX.4.64.0809141427520.10089@parser.ilovebacon.org>
Hi all,
I try to use R to compare models of adsorption onto activated carbon
materials.
Three models are compared:
(1)Langmuir
qe~a*b*Ce/(1+b*Ce)
(2)Freundlich
qe=b*Ce^a
(3)Langmuir-Freundlich
qe~a*b*Ce^z/(1+b*Ce^z)
Experimentally I measure Ce and qe, thus I obtain a dozen of paired values
(Ce,qe) which represents an isotherm of adsorption.
I use the nls() function to assess the parameters of the models and
confint() to get their confidence intervals.
To estimate the fit, I plot the residuals for each models and compare the
SSR values. However, as the model (3) has 3 parameters where models (1) and
(2) have 2, I use the standard deviation to estimate models' fitting each
other.
May I use a sort of R2 determined with the cor() function between the data
and the fitted values, to compare the models as well?
Regards/Cordialement
Benoit Boulinguiez