another predict(*, newd) error

[mmaechler]

From an e-mail to lme4-authors from Patrick Breheny (U.Iowa) - thank you -
example slightly simplified:

require(lme4)
data(Orthodont,package="nlme")
Orthodont <- within(Orthodont, nsex <- as.numeric(Sex == "Male"))

fm1 <- lmer(distance ~ age + (age|Subject) + (0 + Sex|Subject), data=Orthodont)
## (warning about identifiability, not a problem)
p1 <- predict(fm1, Orthodont) # works fine
## but  result  is *WRONG*  [this is "the same bug" as the problem below !]
all.equal(p1, predict(fm1)) # 0.03218


fm2 <- lmer(distance ~ age + (age|Subject) + (0 + nsex|Subject), data=Orthodont)
predict(fm2, Orthodont) #
## Error in t(.Call(Csparse_dense_crossprod, y, x)) :
##    error in evaluating the argument 'x' in selecting a method for
## function 't': Error: Cholmod error 'X and/or Y have wrong dimensions' at
## file ../MatrixOps/cholmod_sdmult.c, line 90
traceback()

and the bug is -- once more -- in mkNewReTrms() which returns a wrong (length) b vector
which then gives an error in b %*% Zt

[mmaechler]

It's worse: I've edited the example code above: When debugging, I found that also the first case
works wrongly... but --unfortunately-- does not give an error, but a wrong result...
--> we have a bad problem, so I call this release critical.

[bbolker]

wow, that was a quick fix ...