# [R] lm without error

Joris Meys jorismeys at gmail.com
Sat Jun 12 02:22:57 CEST 2010

```On Fri, Jun 11, 2010 at 5:28 PM, ivo welch <ivo.welch at gmail.com> wrote:
> thanks, everybody.
>
> joris---let me disagree with you, please.  there are so many
> possibilities of how lm.fit could fail that by the time I am done with
> pre-checking, I may as well write my own lm() routine.
If we all would agree, life would be boring, no? ;-) I see your point,
but the thing is that if the function returns only NA coefficients
(and it can't do anything else than that, as a fit is mathematically
not possible), you have literally no information about what went
wrong. If I fit can't be done, I'd like to know why it happened, and
not just get the answer "Not Available". That's an error message too.

Checking your data before doing an analysis is what we call "Good
Statistical Practice". Using a model on data you didn't check before
is like starting to drive to another country without checking which
direction you have to go. Pretty unlikely you're going to arrive at
the right spot...

This said, you gave us what we need.
>  y= matrix(rnorm(1000), nrow=10, ncol=100)
>  y[,28]= rep(NA, 10)
>  x=rnorm(10)
>  lm( y ~ x )
>       ## now what do you do?  hunt for which column was responsible?
I'd do :
y= matrix(rnorm(1000), nrow=10, ncol=100)
y[,28]= rep(NA, 10)
x=rnorm(10)

getOut <- which(colSums(is.na(y))==dim(y)[1])

lm( y[-getOut] ~ x )

Cheers
Joris
--
Joris Meys
Statistical consultant

Ghent University
Faculty of Bioscience Engineering
Department of Applied mathematics, biometrics and process control

tel : +32 9 264 59 87
Joris.Meys at Ugent.be
-------------------------------
Disclaimer : http://helpdesk.ugent.be/e-maildisclaimer.php

```