[R] split-plot analysis with lme()

[i.m.s.white]

Dear R-help,

Why can't lme cope with an incomplete whole plot when analysing a split-plot
experiment? For example:

R : Copyright 2006, The R Foundation for Statistical Computing
Version 2.3.1 (2006-06-01)

> library(nlme)
> attach(Oats)
> nitro <- ordered(nitro)
> fit <- lme(yield ~ Variety*nitro, random=~1|Block/Variety)
> anova(fit)
              numDF denDF   F-value p-value
(Intercept)       1    45 245.14333  <.0001
Variety           2    10   1.48534  0.2724
nitro             3    45  37.68560  <.0001
Variety:nitro     6    45   0.30282  0.9322

# Excellent! However ---

> fit2 <- lme(yield ~ Variety*nitro, random=~1|Block/Variety, subset=
+ !(Variety == "Golden Rain" & nitro == "0"))
Error in MEEM(object, conLin, control$niterEM) : 
	Singularity in backsolve at level 0, block 1
-- 
************************************************
*    I.White                                   *
*    University of Edinburgh                   *
*    Ashworth Laboratories, West Mains Road    *
*    Edinburgh EH9 3JT                         *
*    Fax: 0131 650 6564   Tel: 0131 650 5490   *
*    E-mail: i.m.s.white at ed.ac.uk              *
************************************************

----- End forwarded message -----

-- 
************************************************
*    I.White                                   *
*    University of Edinburgh                   *
*    Ashworth Laboratories, West Mains Road    *
*    Edinburgh EH9 3JT                         *
*    Fax: 0131 650 6564   Tel: 0131 650 5490   *
*    E-mail: i.m.s.white at ed.ac.uk              *