[R] [R]Intercept in model formulae

[Thomas Lumley]

On Wed, 22 Jan 2003, sadok walid wrote:

>
> Hi,
> I'm a new user of R and I'm trying to make a linear model from this kind of
> dataset
>  x
>  [1] 16.87 19.93 25.85 20.94 17.06 19.49 19.93 25.45 27.74 20.15 25.81
> 21.06 17.17 20.03 25.50 27.79 20.44 16.88 19.93 25.79
>
> z<-x-10
>  y
>  [1] 0.80 1.27 2.22 1.32 0.90 1.18 1.84 2.41 2.97 1.25 2.07 1.41 1.14 1.66
> 2.59 3.51 1.53 0.81 1.26 2.30
>
> plot(x,y)
>
> I want to be able to force the line of the model(mymodel<-lm(y~z)) to pass
> through the origin (zero) and the line of the model(mymodel<-lm(y~x)) to
> pass through 10 as an x-intercept.
> For the first case i have used the (y~x-1) syntax, but the linear model
> returned a regresssion that didn't pass trough the points (with a higher
> Rsq than expected).

It works for me
  plot(z,y,xlim=range(c(0,z)),ylim=range(c(0,y)))
  abline(lm(y~z-1))

gives a line nicely through the points. The fact that the Rsq is higher
than expected may be a defect of the expectations. Rsq is often high
 for models with no intercept because the null model is that the mean of y
is zero.


> For the second case, i did not find a solution yet..

It's the same as the first case, and that's about the only way to do it.

  lm(y~I(x-10)-1)
though I think it looks neater written as
  lm(y~I(x-10)+0)


	-thomas