R devel - Mar 2011 - discrepancy between lm and MASS:rlm

Dear R-devel,

There seems to be a discrepancy in the order in which lm and rlm evaluate their
arguments. This causes rlm to sometimes produce an error where lm is just fine.

Here is a little script that illustrate the issue:
> library(MASS)
> ## create data
> n <- 100
> dat <- data.frame(x=rep(c(-1,0,1), n), y=rnorm(3*n))
>
> ## call lm, works fine
> summary(lm(y ~ as.factor(x), data=dat, subset=x!=0))
Call:
lm(formula = y ~ as.factor(x), data = dat, subset = x != 0)

Residuals:
     Min       1Q   Median       3Q      Max
-2.60619 -0.82160  0.06307  0.65501  2.56677

Coefficients:
              Estimate Std. Error t value Pr(>|t|)
(Intercept)   0.061010   0.100027   0.610    0.543
as.factor(x)1 0.001332   0.141459   0.009    0.992

Residual standard error: 1 on 198 degrees of freedom
Multiple R-squared: 4.479e-07,  Adjusted R-squared: -0.00505
F-statistic: 8.868e-05 on 1 and 198 DF,  p-value: 0.9925
> ## call rlm, error
> summary(rlm(y ~ as.factor(x), data=dat, subset=x!=0))
Error in rlm.default(x, y, weights, method = method, wt.method = wt.method,  :
  'x' is singular: singular fits are not implemented in
rlm
>

My guess is that rlm first converts x to a factor, which becomes a three-level
factor, then subsets on x!=0, which effectively eliminates a level, and then
creates a "regression" matrix, which becomes singular due to the
absence of data for a level.

Is there a simple way to work around it. The simplest I could think of is
with(subset(dat, x!=0), rlm(y ~ as.factor(x))
which is ok, but most of my scripts make use of data arg to regressions and
I'd like to stay consistent as much as practical.

Thanks,
Vadim


Note: This email is for the confidential use of the named addressee(s) only and
may contain proprietary, confidential or privileged information. If you are not
the intended recipient, you are hereby notified that any review, dissemination
or copying of this email is strictly prohibited, and to please notify the sender
immediately and destroy this email and any attachments.  Email transmission
cannot be guaranteed to be secure or error-free.  Jump Trading, therefore, does
not make any guarantees as to the completeness or accuracy of this email or any
attachments.  This email is for informational purposes only and does not
constitute a recommendation, offer, request or solicitation of any kind to buy,
sell, subscribe, redeem or perform any type of transaction of a financial
product.

> -----Original Message-----
> From: r-devel-bounces at r-project.org 
> [mailto:r-devel-bounces at r-project.org] On Behalf Of Vadim Ogranovich
> Sent: Monday, March 14, 2011 10:37 AM
> To: 'r-devel at r-project.org'
> Subject: [Rd] discrepancy between lm and MASS:rlm
> 
> Dear R-devel,
> 
> There seems to be a discrepancy in the order in which lm and 
> rlm evaluate their arguments. This causes rlm to sometimes 
> produce an error where lm is just fine.
It may not be a problem with the order of evaluation.  rlm()
might not be calling model.frame() with drop.unused.levels=TRUE.
I've made that mistake before with similar symptoms.

Bill Dunlap
Spotfire, TIBCO Software
wdunlap tibco.com 
> 
> Here is a little script that illustrate the issue:
> 
> > library(MASS)
> > ## create data
> > n <- 100
> > dat <- data.frame(x=rep(c(-1,0,1), n), y=rnorm(3*n))
> >
> > ## call lm, works fine
> > summary(lm(y ~ as.factor(x), data=dat, subset=x!=0))
> 
> Call:
> lm(formula = y ~ as.factor(x), data = dat, subset = x != 0)
> 
> Residuals:
>      Min       1Q   Median       3Q      Max
> -2.60619 -0.82160  0.06307  0.65501  2.56677
> 
> Coefficients:
>               Estimate Std. Error t value Pr(>|t|)
> (Intercept)   0.061010   0.100027   0.610    0.543
> as.factor(x)1 0.001332   0.141459   0.009    0.992
> 
> Residual standard error: 1 on 198 degrees of freedom
> Multiple R-squared: 4.479e-07,  Adjusted R-squared: -0.00505
> F-statistic: 8.868e-05 on 1 and 198 DF,  p-value: 0.9925
> 
> > ## call rlm, error
> > summary(rlm(y ~ as.factor(x), data=dat, subset=x!=0))
> Error in rlm.default(x, y, weights, method = method, 
> wt.method = wt.method,  :
>   'x' is singular: singular fits are not implemented in rlm
> >
> 
> 
> My guess is that rlm first converts x to a factor, which 
> becomes a three-level factor, then subsets on x!=0, which 
> effectively eliminates a level, and then creates a 
> "regression" matrix, which becomes singular due to the 
> absence of data for a level.
> 
> Is there a simple way to work around it. The simplest I could 
> think of is
> with(subset(dat, x!=0), rlm(y ~ as.factor(x))
> which is ok, but most of my scripts make use of data arg to 
> regressions and I'd like to stay consistent as much as practical.
> 
> Thanks,
> Vadim
> 
> 
> Note: This email is for the confidential use of the named 
> addressee(s) only and may contain proprietary, confidential 
> or privileged information. If you are not the intended 
> recipient, you are hereby notified that any review, 
> dissemination or copying of this email is strictly 
> prohibited, and to please notify the sender immediately and 
> destroy this email and any attachments.  Email transmission 
> cannot be guaranteed to be secure or error-free.  Jump 
> Trading, therefore, does not make any guarantees as to the 
> completeness or accuracy of this email or any attachments.  
> This email is for informational purposes only and does not 
> constitute a recommendation, offer, request or solicitation 
> of any kind to buy, sell, subscribe, redeem or perform any 
> type of transaction of a financial product.
> 
> ______________________________________________
> R-devel at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-devel
>