R devel - Nov 2009 - model.frame error with formula=~1 and na.action=na.fail (PR#14066)

Full_Name: Sebastian Meyer
Version: 2.10.0
OS: x86_64-pc-linux-gnu
Submission from: (NULL) (77.135.124.96)


When using model.frame (model.frame.default) on an intercept-only model and with
na.action=na.fail SPECIFIED, i.e.

   data(iris)
   model.frame( ~ 1, data = iris, na.action = na.fail)

one gets the following error:

   Error in complete.cases(object) : negative length vectors are not allowed

There is no error if the na.action argument is left UNSPECIFIED (though the
default value also is na.fail!):

   model.frame( ~ 1, data = iris)

In either case, there should be no error. Specifing other na.action functions
(na.omit, na.pass, na.exclude) works.
If there is at least one (non-intercept) variable in the formula, e.g.

   model.frame( ~ 1 + Species, data = iris, na.action = na.fail)

there is obviously no issue. 

Best regards,
    Sebastian Meyer

Thank you for the report -- but a bit more investigation of the root 
cause would have been even more helpful: this is not 'model.frame 
error', nor even reported by R as such.

The issue was that complete.cases() did not know how to find the 
number of rows of a 0-column data frame (it knew nothing about data 
frames). That has been changed in R-patched and R-devel, together with 
a more informative error message in other cases where the number of 
rows is unspecified.

On Sun, 15 Nov 2009, Sebastian.Wastl.Meyer at campus.lmu.de wrote:
> Full_Name: Sebastian Meyer
> Version: 2.10.0
> OS: x86_64-pc-linux-gnu
> Submission from: (NULL) (77.135.124.96)
>
>
> When using model.frame (model.frame.default) on an intercept-only model and
with
> na.action=na.fail SPECIFIED, i.e.
>
>   data(iris)
That line is unnecessary: 'iris' is a visible object in package 
'datasets'.
>   model.frame( ~ 1, data = iris, na.action = na.fail)
>
> one gets the following error:
>
>   Error in complete.cases(object) : negative length vectors are not allowed
>
> There is no error if the na.action argument is left UNSPECIFIED (though the
> default value also is na.fail!):
>
>   model.frame( ~ 1, data = iris)
>
> In either case, there should be no error. Specifing other na.action
functions
> (na.omit, na.pass, na.exclude) works.
> If there is at least one (non-intercept) variable in the formula, e.g.
>
>   model.frame( ~ 1 + Species, data = iris, na.action = na.fail)
>
> there is obviously no issue.
>
> Best regards,
>    Sebastian Meyer
>
> ______________________________________________
> R-devel at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-devel
>
-- 
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