anova of lme objects (model1, model2) gives different results depending on order of models

Hello-

I understand that it's convention, when comparing two models using
the anova function anova(model1, model2), to put the more
"complicated" (for want of a better word) model as the second model.
However, I'm using lme in the nlme package and I've found that the
order of the models actually gives opposite results. I'm not sure if
this is supposed to be the case or if I have missed something
important, and I can't find anything in the Pinheiro and Bates book
or in ?anova, or in Google for that matter which unfortunately only
returns results about ANOVA which isn't much help. I'm using the
latest version of R and nlme, just checked both.

Here is the code and output:

PHQmodel1=lme(PHQ~Age+Gender+Date*Treatment, data=compfinal,

random=~1|Case, na.action=na.omit)

PHQmodel2=lme(PHQ~Age+Gender+Date*Treatment, data=compfinal,

random=~1|Case, na.action=na.omit,
+              correlation=corAR1(form=~Date|Case))

anova(PHQmodel1, PHQmodel2) # accept model 2

                    Model df      AIC      BIC    logLik   Test
L.Ratio p-value
PHQmodel1     1  8 48784.57 48840.43 -24384.28
PHQmodel2     2  9 48284.68 48347.51 -24133.34 1 vs 2 501.8926 <.0001

PHQmodel1=lme(PHQ~Age+Gender+Date*Treatment, data=compfinal,

random=~1|Case, na.action=na.omit,
+              correlation=corAR1(form=~Date|Case))

PHQmodel2=lme(PHQ~Age+Gender+Date*Treatment, data=compfinal,

random=~1|Case, na.action=na.omit)

anova(PHQmodel1, PHQmodel2) # accept model 2

                     Model df      AIC      BIC    logLik   Test
L.Ratio p-value
PHQmodel1     1  9 48284.68 48347.51 -24133.34
PHQmodel2     2  8 48784.57 48840.43 -24384.28 1 vs 2 501.8926 <.0001

In both cases I am led to accept model 2 even though they are
opposite models. Is it really just that you have to put them in the
right order? It just seems like if there were say four models you
wouldn't necessarily be able to determine the correct order.

Many thanks,
Chris Beeley, Institute of Mental Health, UK

...session info follows

sessionInfo()

R version 2.15.0 (2012-03-30)
Platform: i386-pc-mingw32/i386 (32-bit)

locale:
[1] LC_COLLATE=English_United Kingdom.1252  LC_CTYPE=English_United
Kingdom.1252
[3] LC_MONETARY=English_United Kingdom.1252 LC_NUMERIC=C
[5] LC_TIME=English_United Kingdom.1252

attached base packages:
[1] grid      stats     graphics  grDevices utils     datasets
methods   base

other attached packages:
 [1] gridExtra_0.9      RColorBrewer_1.0-5 car_2.0-12
nnet_7.3-1         MASS_7.3-17
 [6] xtable_1.7-0       psych_1.2.4        languageR_1.4
nlme_3.1-104       ggplot2_0.9.1

loaded via a namespace (and not attached):
 [1] colorspace_1.1-1 dichromat_1.2-4  digest_0.5.2     labeling_0.1
lattice_0.20-6   memoise_0.1
 [7] munsell_0.3      plyr_1.7.1       proto_0.3-9.2
reshape2_1.2.1   scales_0.2.1     stringr_0.6
[13] tools_2.15.0

packageDescription("nlme")

Package: nlme
Version: 3.1-104
Date: 2012-05-21
Priority: recommended
Title: Linear and Nonlinear Mixed Effects Models
Authors at R: c(person("Jose", "Pinheiro", comment = "S version"),
person("Douglas", "Bates", comment =
           "up to 2007"), person("Saikat", "DebRoy", comment = "up
to 2002"), person("Deepayan",
           "Sarkar", comment = "up to 2005"), person("R-core", email
= "R-core at R-project.org", role =
           c("aut", "cre")))
Author: Jose Pinheiro (S version), Douglas Bates (up to 2007),
Saikat DebRoy (up to 2002), Deepayan
           Sarkar (up to 2005), the R Core team.
Maintainer: R-core <R-core at R-project.org>
Description: Fit and compare Gaussian linear and nonlinear
mixed-effects models.
Depends: graphics, stats, R (>= 2.13)
Imports: lattice
Suggests: Hmisc, MASS
LazyLoad: yes
LazyData: yes
License: GPL (>= 2)
BugReports: http://bugs.r-project.org
Packaged: 2012-05-23 07:28:59 UTC; ripley
Repository: CRAN
Date/Publication: 2012-05-23 07:37:45
Built: R 2.15.0; x86_64-pc-mingw32; 2012-05-29 12:36:01 UTC; windows