[ADMB Users] glmmADMB errors

[Anne Bjorkman]

Dear ADMB users,


I hope this is the right mailing list, I wasn't sure whether to post here
or the mixed models list...


I am attempting to model zero-inflated data with a random effect (repeated
measures on individuals over time).


My data are total and relative abundances of calories of various crop
species (~40 total) consumed within each country (154 countries total) over
the past 40 years, sampled at 5 time periods (1967, 1977, 1987, 1997,
2007). The data appear as follows:


  Year Country          Item_general           Total     Rel.Total

1 1967 Albania                Apples              6.45     0.002903365

2 1967 Albania Bananas_and_plantains     0.00    0.000000000

3 1967 Albania                Barley              17.09    0.007692792

4 1967 Albania                 Beans             41.80    0.018815610

5 1967 Albania   Beverages_Alcoholic      5.60     0.002520751

6 1967 Albania   Beverages_Fermented    0.00     0.000000000


My plan is to run the model for each crop separately, and the output I
would like to get is slope parameters with confidence intervals (this would
tell me the direction and magnitude of change over time for each crop).
[Though if anyone has any better ideas of how I could show change over time
for each crop, I'd certainly be interested in hearing them!]


I have not been able to find a way to account for both zero-inflation and
repeated measures other than using the glmmADMB package (at least, not
within my stats abilities).  Unfortunately I am running into a whole host
of errors in glmmADMB, in particular this one:


Error in glmmadmb(Rel.Total.Rnd ~ I(Year - 1967), random = ~1 | Country,  :

  The function maximizer failed (couldn't find STD file)

In addition: Warning message:

running command './glmmadmb -maxfn 500 -maxph 5 -noinit -shess' had status
1

Need to increase the maximum number of separable calls allowed to at least
20001

Current value is 20000

Use the -ndi N command line option


I have been using the following command:


glmmadmb(Rel.Total~I(Year-1967)+(1|Country),
family="nbinom",zeroInflation=TRUE,data=crops)


where Rel.Total is rounded to nearest (ceiling) integer.  I run into the
same error when I use the absolute total (rounded to nearest integer)
instead of relative total. Interestingly, I do not run into this error when
I run this same model on a subset of only one crop species.


Googling this error brought up another email from someone having a similar
problem, and his problem was solved by downloading an older version of the
software, but I tried that (v. 0.7) and I am still getting the same error
(I am using Mac OX 10.7).  I have also tried adding the "
admb.opts=admbControl(shess=FALSE,noinit=FALSE)" command, which I saw
recommended in some of the list serves, but that does not seem to help.


I have also run into trouble when trying to check assumptions for these
models.  I have tried plotting residuals vs. fitted values by
plot(model.out$fitted,model.out$residuals) and the resulting graph looks no
better than that for a regular mixed model (using nlme) that doesn't
account for zero inflation. The graph shows a horizontal V-shape, so that
the values are close together on the left side of the graph and fan out
widely on the right side.  I was able to make the residuals look a lot
better in the regular mixed model by 4th-root transforming the data (I know
this isn't an appropriate solution, I was just experimenting), so I am
confused as to why the glmmADMB graph of fitted v. residuals wouldn't look
better.


I would also like to try allowing for random slopes in the model (using
(Year|Country)), but when I do this there are only about 5 points on the
graph of fitted v. residuals and they form a straight line from top left to
bottom right. This doesn't happen when I include random slopes in nlme, so
perhaps it has something to do with the way the model accounts for
zero-inflation?


I apologize for the rambling nature of this email.  I am relatively new to
mixed models, and while I have read Zuur from cover to cover, I want to
make sure I really understand what I'm doing rather than just plugging in
numbers.


I would be happy to provide my full dataset off-line if anyone is
interested.


Any help would be greatly appreciated!

Many thanks,

Anne
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.admb-project.org/pipermail/users/attachments/20120301/4f131557/attachment.html>