p.dalgaard@biostat.ku.dk writes:
> Try this
>
> data(Assay)
> as1 <- lme(logDens~sample*dilut, data=Assay,
> random=pdBlocked(list(
> pdIdent(~1),
> pdIdent(~sample-1),
> pdIdent(~dilut-1))))
>
> update(as1,random=pdCompSymm(~sample-1))
> update(as1,random=pdCompSymm(~sample-1))
> update(as1,random=pdCompSymm(~sample-1))
> update(as1,random=pdCompSymm(~sample-1))
>
> I'm getting different results on different invocations!
I'm not sure exactly what the problem was but it could be triggered by
calling lme directly rather than through update. That is, successive
calls to
lme(logDens~sample*dilut, data=Assay, random=pdCompSymm(~sample-1))
would alternate between a successful fit and an error.
The problem does *not* occur with the current r-patched, which will be
released as R-1.7.1 in early June. At least it does not occur under
Linux. I would appreciate it if someone could check with r-patched
under Windows.
I have added this example as one of the tests in nlme_3.1-40, which
will be released before R-1.7.1