[R] qr decomposition issue inside lm (solved)

[Horace Tso]

Folks, apologize for such an obvious oversight on my part. The reason qr
fails is, one of the data points has value of -Inf (response is actually
the log of something, and I have a zero in the original set). That
explains the error message in call to dqrls. I should have taken the
mean of the response before proceeding and that would tell me right away
what's wrong. 

Thanks.

H.

>>> "Horace Tso" <Horace.Tso at pgn.com> 3/6/2007 1:45:28 PM >>>
Dear list,

It's never happened to me before in such a simple exercise but is not
going away and I've checked my data are good. I want a simple lm model
with one response and one predictor, where N is about 4,200 * data set
not exactly small. Both x and y are nice, continuous variables having NA
filtered out with a call to na.omit. So I did

mod = lm( y ~ x, data=x1)

Then the error,

Error in lm.fit(x, y, offset = offset, singular.ok = singular.ok, ...)
: 
        NA/NaN/Inf in foreign function call (arg 4)

I did a trace back and it turned out it's an error thrown by the
Fortran subroutine that seems to be trying a QR decomposition,

traceback()
3: .Fortran("dqrls", qr = x, n = n, p = p, y = y, ny = ny, tol =
as.double(tol), 
       coefficients = mat.or.vec(p, ny), residuals = y, effects = y, 
       rank = integer(1), pivot = 1:p, qraux = double(p), work =
double(2 * 
           p), PACKAGE = "base")
2: lm.fit(x, y, offset = offset, singular.ok = singular.ok, ...)
1: lm(log.p.sales ~ log.mktcap, data = x1)

My question is why would QR fail since the default in lm.fit is
'singular.ok' ? Furthermore, is there a way to get around presumably a
singularity in my design matrix? 

Thanks in advance.

Horace W. Tso

______________________________________________
R-help at stat.math.ethz.ch mailing list
https://stat.ethz.ch/mailman/listinfo/r-help 
PLEASE do read the posting guide
http://www.R-project.org/posting-guide.html 
and provide commented, minimal, self-contained, reproducible code.