Thanks to Brian Ripley for his response. He suggests using offset instead of
fiddling with the coefficients of the model; I'll look into his suggestion. In
the mean time I found out what was going on. By using the assignment:
ttt$coef[1]_ aut4.freq.glm.final$coef[1] + log(0.96)
Splus was creating a new component of the object ttt called "coef". ttt$coef
was accessing this new component while coef(ttt) was accessing the old
component, called "coefficients"; the new one reflected the change while the old
one didn't of course.
> ttt_aut4.freq.glm.final
> names(ttt)
[1] "coefficients" "residuals" "fitted.values"
[4] "effects" "R" "rank"
[7] "assign" "df.residual" "weights"
[10] "prior.weights" "family" "linear.predictors"
[13] "deviance" "null.deviance" "call"
[16] "iter" "y" "contrasts"
[19] "terms" "formula" "control"
> ttt$coef[1]_ aut4.freq.glm.final$coef[1] + log(0.96)
> names(ttt)
[1] "coefficients" "residuals" "fitted.values"
[4] "effects" "R" "rank"
[7] "assign" "df.residual" "weights"
[10] "prior.weights" "family" "linear.predictors"
[13] "deviance" "null.deviance" "call"
[16] "iter" "y" "contrasts"
[19] "terms" "formula" "control"
[22] "coef"
Thanks to prof. Ripley for his suggestion, original post follows.
Gérald Jean
Analyste-conseil (statistiques), Actuariat
télephone : (418) 835-8839
télecopieur : (418) 835-5865
courrier électronique: gerald.jean@spgdag.ca
"In God we trust all others must bring data"
>
> Dear S-users,
>
> for non-statistical reasons I would like to adjust, before using predict, some
> coefficients of a glm. For the sake of argument suppose I only want to adjust
> the constant as follows.
> > ttt_aut4.freq.glm.final
> > ttt$coef[1]_ aut4.freq.glm.final$coef[1] + log(0.96)
> > ttt$coef[1]
> (Intercept)
> -5.385395
> > coef(ttt)[1]
> (Intercept)
> -5.344573
>
> What is going on? The correct answer is -5.385395, see the original
> constant below.
> > aut4.freq.glm.final$coef[1]
> (Intercept)
> -5.344573
> > coef(aut4.freq.glm.final)[1]
> (Intercept)
> -5.344573
>
> When I call predict passing ttt as the model and using the newdata argument
> which version of the constant will be used?
>
> Thanks for your help,
-----------------------------------------------------------------------
This message was distributed by s-news@wubios.wustl.edu. To unsubscribe
send e-mail to s-news-request@wubios.wustl.edu with the BODY of the
message: unsubscribe s-news
|