[R-sig-ME] convergence warning lmer depends on order expl. vars

[Ben Pelzer]

Dear list,

When using lmer I ran I'm into a strange situation.
Using:

M1 <- lmer (AIMSontw.lt ~ (1+age | P.nr)+age+baseline+treatm,
             verbose=TRUE, family=gaussian, REML=FALSE)

generates the warning : In mer_finalize(ans) : singular convergence (7). 
However, using:

M2 <- lmer(AIMSontw.lt ~ (1+age | P.nr)+age+treatm+baseline,
             verbose=TRUE, family=gaussian, REML=FALSE)

with only the order of two independents changed, does NOT generate the 
above warning.

The fixed and random effect estimates are the same for both models.
AIMSontw.lt is continuous and ranges from 0 to 18.
age (after having been centered around some meaningful value) is 
continuous and ranges from -8 to 46.
obaseline is continuous and ranges from 26 to 83.
treatm is a dichotomous 0/1 variable.
P.nr is a factor with 22 levels, i.e., respondents.
The total N=149.

There may be a problem with this model/data: the correlation between the 
random intercept and the random slope of age equals 1.000. But leaving 
out one of these two random effects greatly inflates the deviance.  In 
short: I'm in doubt about how to proceed.

Thanks for any comment/advice!!

Ben.