[R] problems using lqs()

[ripley@stats.ox.ac.uk]

On Mon, 10 Feb 2003, Luca Scrucca wrote:

> I found a strange behaviour in the lqs function.

Why is this strange?  It *is* covered in the Note on the help page: please
read it and the reference there.

> Suppose I have the following data:

[...]

> Now, instead of using the formula, I want to provide the design matrix and
> the response, then:
> 
> X <- cbind(1, x1, x2)
> mod2 <- lqs.default(X, y, intercept=F, method="lms", nsamp="exact")
> mod2$coefficients
>                    x1         x2
> 35.4217275  0.4276641 -1.2834731
> mod2$bestone
> [1]  6 14 15
> 
> The results are not the same (?!). Furthermore, if I create the design
> matrix without the column of 1's for the intercept:

Right, it's not the same algorithm.

> X <- cbind(x1, x2)
> mod3 <- lqs.default(X, y, intercept=T, method="lms", nsamp="exact")

> I get the same result I had using the formula (see mod1).
> This is confusing me!

That *is* the same algorithm.

> Another small problem appears if I re-fit the first model using the
> formula and asking to return the x-matrix and the response:
> 
> lqs(y ~ x1 + x2, method="lms", nsamp="exact", x.ret=T, y.ret=T)
> 
> Error in model.frame(formula, rownames, variables, varnames, extras,
> extranames,  :
> 	variable lengths differ
> 
> Does anyone know what's going on?

No, but R has debugging tools for you to find out ....


-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595