Re: [R] Conditional piece-wise dependent regression

From: Dimitris Rizopoulos <dimitris.rizopoulos_at_med.kuleuven.be>
Date: Mon 01 Aug 2005 - 18:39:48 EST

Since you want least squares, I think you could use lm() here, i.e.,

U <- 50
V <- 100
Time <- 1:150
dat <- data.frame(y = rnorm(150), Time, f1 = as.numeric(Time > U), f2 = as.numeric(Time > V))
###############
m <- lm(y ~ Time + I(Time - U):f1 + I(Time - V):f2, data = dat)

# check also the design matrix
model.matrix(m)

I hope it helps.

Best,
Dimitris

Dimitris Rizopoulos
Ph.D. Student
Biostatistical Centre
School of Public Health
Catholic University of Leuven

```Tel: +32/16/336899
Fax: +32/16/337015
Web: http://www.med.kuleuven.be/biostat/
http://www.student.kuleuven.be/~m0390867/dimitris.htm

```
• Original Message ----- From: "Arie" <poohdov@yahoo.com> To: <r-help@stat.math.ethz.ch> Sent: Monday, August 01, 2005 10:17 AM Subject: [R] Conditional piece-wise dependent regression

Hi, after reading some R docs, I couldn't figure out how can I find the solution for the following

We're making a least square approximation for an experiment described by the following model:

T is the time,
Y is some measured value.

>From time=0 till time=U:
Y = b + p*T
>From time=U and on (some effect added):
Y = b + p*T + q*(T-U)
Y = b + p*T + q*(T-U) + r*(T-V)

Measured: Yi, Ti pairs.
Wanted: b, p, q, r.
b and p are the same for all time ranges; q is the same for time=U and on.

Thanks,
Arie.

R-help@stat.math.ethz.ch mailing list