Thanks to all who responded: Madeline Bauer, Henrik Aalborg Nielsen, Hong Ooi,
Renaud Lancelot, F. Tussell, and Robert Key.
All suggestions don't provide some sort of regression coefficients, or rules,
with which to get fitted values at other "x" values. I probably wasn't clear
enough on this one; the Splus function predict IS NOT AN OPTION, whatever I will
come up with will be used in a production environment on a main frame computer
in I don't even know which language, maybe SAS but that's just a guess.
Let me explain the problem. I have a glm model fitting some tabled data.
Suppose amongst the predictors, all factors by the way, I have an "age" factor,
the levels being c("0-20", "20-30", "30-40", ... , "over 100"), say. I got
regression estimates for all levels of that factor, the reference level having
coefficient 0. Given we have a person 32 years old, we want to "refine" a
little bit the coefficient we got for the "30-40" group. How can that be done.
Using the coefficients as dependent and the middle of the age brackets as
predictors I can get very good predictions at intermediate age levels using some
of the Splus smoothers, or something like «lm(y ~ ns(x), ...) but I don't get
"rules" that could be used in a different language to get estimates at
intermediate values? I could use linear interpolation between the values at
which I get estimates, but the original curve sure doesn't seem linear. Any
suggestions welcomed.
Thanks to all who responded,
Gérald Jean
Analyste-conseil (statistiques), Actuariat
télephone : (418) 835-4900 poste (7639)
télecopieur : (418) 835-6657
courrier électronique: gerald.jean@spgdag.ca
"In God we trust all others must bring data"
-----------------------------------------------------------------------
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
|