[R] drop1() seems to give unexpected results compare to anova()

[Prof Brian Ripley]

On Sun, 3 Aug 2008, Thomas P C Chu wrote:

> Thanks to Mr Dalgaard for his advice and everyone else who has contributed. 
> Inclusion of an error term at the end of sim.set$y = ... line did cure my 
> problems with drop1() and step().
>
> I suppose it is my own inexperience in carrying out simulations caused such 
> gaffe.

R 2.8.0 will warn if asked to compute F tests that are essentially 0/0 
(where 'essentially' is relative to the total sum of squares).  It does so 
in your examples, so in future it will be easier to notice this.

-- 
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