Skip to content

possible bug in model.matrix

2 messages · Whit Armstrong, Peter Dalgaard

Original
Whit Armstrong
# 
Is this a bug, or have I misunderstood the proper use of lm?

Thanks,
Whit


code:
x <- rnorm(50)
y <- matrix(as.logical(round(runif(100),0)),ncol=2)
NROW(x)==NROW(y)
lm(x~y)
[1] TRUE
Error in "[[<-.data.frame"(`*tmp*`, nn, value = c(2, 1, 2, 1, 1, 1, 2,
:
        replacement has 100 rows, data has 50
However, the call to lm works if the matrix is a numeric instead of
logical:
x <- rnorm(50)
y <- matrix(runif(100),ncol=2)
NROW(x)==NROW(y)
lm(x~y)


Seems to be a problem in model.matrix.default:

debug: for (nn in namD[isF]) if (is.null(attr(data[[nn]], "contrasts")))
contrasts(data[[nn]]) <- contr.funs[1 +
    isOF[nn]]
Browse[1]>
Error in "[[<-.data.frame"(`*tmp*`, nn, value = c(1, 2, 2, 2, 2, 2, 2,
:
        replacement has 100 rows, data has 50
$platform
[1] "i686-pc-linux-gnu"

$arch
[1] "i686"

$os
[1] "linux-gnu"

$system
[1] "i686, linux-gnu"

$status
[1] "alpha"

$major
[1] "2"

$minor
[1] "2.0"

$year
[1] "2005"

$month
[1] "09"

$day
[1] "12"

$"svn rev"
[1] "35558"

$language
[1] "R"
Peter Dalgaard
# 
"Whit Armstrong" <whit at twinfieldscapital.com> writes:
Dunno. It appears that logicals like factors are not supposed to have
matrix structure. What actually happens is that setting contrasts
strips dimension attributes
 
Browse[1]>
debug: for (nn in namD[isF]) if (is.null(attr(data[[nn]], "contrasts"))) contrasts(data[[nn]]) <- contr.funs[1 +
    isOF[nn]]
Browse[1]> zz <- data[["y"]]
Browse[1]> contrasts(zz) <-  contrasts(zz)
Browse[1]> zz
  [1] TRUE  TRUE  TRUE  FALSE TRUE  TRUE  TRUE  TRUE  TRUE  FALSE TRUE FALSE
 [13] FALSE TRUE  TRUE  TRUE  FALSE TRUE  FALSE TRUE  TRUE  FALSE FALSE TRUE
 [25] TRUE  TRUE  FALSE FALSE FALSE FALSE TRUE  FALSE FALSE TRUE  TRUE FALSE
 [37] FALSE FALSE FALSE TRUE  TRUE  TRUE  TRUE  FALSE FALSE FALSE TRUE
..
 [85] TRUE  TRUE  TRUE  TRUE  TRUE  TRUE  TRUE  FALSE TRUE  TRUE  TRUE  TRUE
 [97] TRUE  TRUE  TRUE  FALSE
Levels: FALSE TRUE
 
which in turn comes from

if (is.logical(x)) x <- factor(x, levels = c(FALSE, TRUE))

and the fact that factor() throws away dimensions.

*If* it's a bug, I don't think it is easily fixable....