[R] Problem with aovlmer.fnc in languageR

[Mats Exter]

Dear R list,

I have a recurring problem with the languageR package, specifically the 
aovlmer.fnc function. When I try to run the following code (from R. H. 
Baayen's textbook):


     # Example 1:
     library(languageR)
     latinsquare.lmer <- lmer(RT ~ SOA + (1 | Word) + (1 | Subject),
                              data = latinsquare)
     x <- pvals.fnc(latinsquare.lmer,
                    withMCMC = TRUE)
     aovlmer.fnc(latinsquare.lmer,
                 mcmc = x$mcmc,
                 which = c("SOAmedium", "SOAshort"))


I get the following error message (German locale):


     Fehler in anova(object) : Calculated PWRSS for a LMM is negative


Invoking traceback yields the following result:


     > traceback()
     4: .Call(mer_update_projection, object)
     3: anova(object)
     2: anova(object)
     1: aovlmer.fnc(latinsquare.lmer, mcmc = x$mcmc, which = c("SOAmedium",
            "SOAshort"))


By contrast, the following code (without the aovlmer.fnc command) runs 
without error:


     # Example 2:
     library(languageR)
     latinsquare.lmer <- lmer(RT ~ SOA + (1 | Word) + (1 | Subject),
                              data = latinsquare)
     pvals.fnc(latinsquare.lmer,
               withMCMC = TRUE)


Similarly, the following code (without the pvals.fnc command, and 
consequently running aovlmer.fnc *without* MCMC sampling) also runs 
without error:


     # Example 3:
     library(languageR)
     latinsquare.lmer <- lmer(RT ~ SOA + (1 | Word) + (1 | Subject),
                              data = latinsquare)
     aovlmer.fnc(latinsquare.lmer,
                 noMCMC = TRUE)


However, the following code results in exactly the same error message as 
given above, which puzzles me because up to and including the pvals.fnc 
command, the code is basically the same as in Example 2 (which runs 
fine); and the code in the final aovlmer.fnc command is the same as in 
Example 3 (which also runs fine) and, crucially, does not refer to the 
object created by pvals.fnc (named "x") at all:


     # Example 4:
     library(languageR)
     latinsquare.lmer <- lmer(RT ~ SOA + (1 | Word) + (1 | Subject),
                              data = latinsquare)
     x <- pvals.fnc(latinsquare.lmer,
                    withMCMC = TRUE)
     aovlmer.fnc(latinsquare.lmer,
                 noMCMC = TRUE)


Trying to run the examples given in the languageR documentation, no 
error occurs for me with the example in ?pvals.fnc; however, the same 
error as above occurs when running the example given in ?aovlmer.fnc.

I am using Windows XP SP3, R version 2.8.0, and updated packages. The 
following is the result of sessionInfo:


     > sessionInfo()
     R version 2.8.0 (2008-10-20)
     i386-pc-mingw32

     locale:
 
LC_COLLATE=German_Germany.1252;LC_CTYPE=German_Germany.1252;LC_MONETARY=German_Germany.1252;LC_NUMERIC=C;LC_TIME=German_Germany.1252

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

     other attached packages:
     [1] languageR_0.953    lme4_0.999375-27   Matrix_0.999375-16 
zipfR_0.6-5        lattice_0.17-17

     loaded via a namespace (and not attached):
     [1] grid_2.8.0  tools_2.8.0


To sum up, the error occurs when invoking aovlmer.fnc, but somehow seems 
to be caused by previous invocation of pvals.fnc. I have reproduced the 
error on several other computers (running on Windows XP SP2, Windows XP 
SP3, and Windows Vista) with fresh installations of R, so I do hope I 
have excluded any trivial blunders on my part. Any help is greatly 
appreciated.

Thank you very much in advance!

Mats Exter

-- 
Mats Exter
Institut für Linguistik
Allgemeine Sprachwissenschaft
Universität zu Köln
50923 Köln
Germany