R help - May 2007 - Fitting model with response and day bias

Hi,

I'm trying to fit a model which has a response bias, but also a day to day 
bias.
If I try to simulate the data, I don't get the right values with optim, and 
also I can't
use the function to give a prediction interval.

My simulated data are:

DF <- as.data.frame(cbind(x=rep(1:10,2),dag=rep(1:2,each=10)))
bias <- c(-0.2,0.5)
DF$y <- ((DF$x-0.1) * 5)+2 + bias[DF$dag]+rnorm(20,0,sd=0.5)


Which I try to fit with:
fn <- function(x){
    a <- x[1]
    b <- x[2]
    c <- x[2]
    sum((DF$y - (((DF$x-c)*a)+b + x[DF$dag+2]))^2)
    }
optim(c(1,1,1,1,1),fn)

But with poor succes.

Also, in the real model, I have a response which is y/time (like in 
lm(y/time~x1 + x2,...) ) , but if I put the time variable at the right side 
(lm(y~I(x1 + x2)*time) , it gets an coefficient.
Is there a way to avoid this?

Thanks


Bart

Bernardo,

thanks for your solution, I was short in time last weeks, but I tried it and 
it works fine.
I'll post it to the list, as it may help other people too.

Regards


Bart

----- Original Message ----- 
From: "Bernardo Rangel Tura" <tura at centroin.com.br>
To: "Bart Joosen" <Bartjoosen at hotmail.com>
Sent: Wednesday, May 30, 2007 8:36 PM
Subject: Re: [R] Fitting model with response and day bias

> On Wed, 2007-05-30 at 17:20 +0200, Bart Joosen wrote:
>> Hello,
>>
>> you are right about lm(I(y/time)~x1+x2), but what I would like to do
is:
>> lm(y~(x1+x2)*time), where x1 and x2 should have coefficients, but time 
>> not.
>> Time is just used as a multiplier and should not have a coefficient.
>> But I don't know if its possible.
>>
>> The reason why time is not in my code is that it's simplified to
make it
>> a
>> bit understandable.
>>
>>
>> Thanks anyway
>>
>> Bart
>
> Bart,
>
> If I understand you problem, you have a dataset with 4 variables (y, x1,
> x2 and time) and you need fit them model:
>
> y = A+B*(x1*time)+C*(x2*time)+error
>
> and use command: lm(y~(x1+x2)*time).
>
> I think the comand is: lm(y~I(x1*time)+I(x2*time))
>
> I sugest try this ...
>
> I hope this help you ...
>
> If is solve your problem put solution in list to help other people
> -- 
> Bernardo Rangel Tura, M.D, PhD
> National Institute of Cardiolgy
> Rio de Janeiro - Brasil
>
>