Skip to content
Prev 3487 / 20628 Next

cloglog and lmer

What are the results of sessionInfo() ?

  I am using a slightly hacked version of lme4, but I don't *think* that
any of my hacks should affect the functioning of cloglog.

  You ran exactly the toy code example that I provided, right?

  It's possible that this problem is right on the edge of numerical
tractability and thus that there are some tiny platform-dependent
differences that lead to a probability estimate that is very small on
one platform and exactly 0 (due to underflow) on another platform,
triggering the error.

  In fact the error message you got

... mu = 9.35952e-313, i = 1068030290 ...

  suggests this may be the case -- although I don't see how mu =
9.3592e-313 is triggering a (mui<=0 || mui>1) flag ...

  It also suggests there may be another bug in the reporting because I
doubt there are really 1068030290 cases in the data.

  However, there were some old bugs of this type that do not appear in
the current (at least in the development) version of lme4, so I strongly
recommend that you at least check to see if you have the latest released
version, and perhaps switch to the development version
(install.packages("lme4",repos="http://r-forge.r-project.org")).

  It might also be nice if (but this is definitely a can of worms with
reasonable arguments in either direction) lme4 could be tolerant of mu
values on the borders of the allowed region or (???) slightly beyond it,
so that optimizations could proceed to a final answer that might not be
at those boundaries.

  cheers
    Ben Bolker
R version 2.10.1 (2009-12-14)
i486-pc-linux-gnu

locale:
 [1] LC_CTYPE=en_US.UTF-8       LC_NUMERIC=C
 [3] LC_TIME=en_US.UTF-8        LC_COLLATE=en_US.UTF-8
 [5] LC_MONETARY=C              LC_MESSAGES=en_US.UTF-8
 [7] LC_PAPER=en_US.UTF-8       LC_NAME=C
 [9] LC_ADDRESS=C               LC_TELEPHONE=C
[11] LC_MEASUREMENT=en_US.UTF-8 LC_IDENTIFICATION=C

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

other attached packages:
[1] lme4_0.999375-32-2 Matrix_0.999375-38 lattice_0.18-3

loaded via a namespace (and not attached):
[1] grid_2.10.1
Sasha Goodman wrote: