Have you looked at the correlation of the parameter estimates?
In S-Plus, this is automatically produced by summary(lm(...)).
There is probably a simple, elegant way to get this in R, but I don't
know it. Therefore, I produced it as follows:
> DF <- data.frame(x=1:12, y=rep(1:6, each=2))
> fit <- lm(y~x, DF)
> sumfit <- summary(fit)
> cov.fit <- sumfit$cov.unscaled
> seb <- sqrt(diag(cov.fit))
> cov.fit / outer(seb, seb)
(Intercept) x
(Intercept) 1.000000 -0.883176
x -0.883176 1.000000
I developed this by looking at attributes(fit), then attributes(sumfit),
etc.
Is this helpful?
spencer graves
Kyriakos Kachrimanis wrote:
>Dear list members,
>
> this is not an R question and forgive me for using the list for irrelevant
>questions, but this is the only place I know where I can find some good
>statisticians and I need an expert opinion.
>There is this power law kinetic model of the form:
>M=kt^n
>where t is the time, M is the fraction of drug released, k is the rate
>constant and n is an exponent related to the mechanism of release.
>I fit this equation to 100 different datasets (by linear regression, after
>logarithmic transformation) and then I plot n vs k. What I get is a decay
>pattern of n as k increases.
>QUESTION 1: Is some theoretical reason for this kind of relation. Is it
>expected? Is it "normal" that model parameters (intercept and
slope in the
>linearised form) should be correlated?
>QUESTION 2: Is it theoretically sound to compare the magnitude of constants,
>k, that correspond to different exponents, n? (k has dimensions of [time]^-n
>so we are comparing quantities with different physical interpretation).
>Practically speaking, does k=0.2 with an exponent n=0.5 mean that we have
>the same release rate as in the case of k=0.2 but with n=1.0?
>
>Thank you very much in advance.
>
> Kyriakos.
>
>
>
>---
>
>______________________________________________
>R-help at stat.math.ethz.ch mailing list
>https://www.stat.math.ethz.ch/mailman/listinfo/r-help
>
>