[R-sig-ME] In mer_finalize(ans) : gr cannot be computed at initial par (65)

[Douglas Bates]

On Mon, Aug 2, 2010 at 12:18 AM, Sol Heber
<sol.heber at pg.canterbury.ac.nz> wrote:
> Dear R List –
>
> I am trying to use a GLMM to analyse data on breeding success from crossing experiments between inbred lines of fruit flies but am having problems.
>
> My response variable is a proportion (proportion of eggs that hatched into adults), I have two fixed factors (one of them, “cross”, has 4 levels: inbred, 1st or 2nd generation hybrid, and outbred, while the other fixed factor, “line”, has 3 levels: the initial two lines of fruit flies used for the experiments and the resulting hybrid line), and two random terms – female and male origin (to control for the fact that eggs of the same pair are not independent).
>
> My experimental design therefore looks something like this (resulting in missing cells in some of the combinations, e.g. there would be no data for the interaction inbred*AB, because in the inbred category there is no mix of the two lines):
>
> Cross                           Lines
> Inbred          AA                              BB
> Hybrid F1       A1A2            AB              B1B2
> Hybrid F2       A1A2            AB              B1B2
> Outbred         AA                              BB
>
> I have been using the following commands:
>
>> data<-read.table(file="data.txt",header=TRUE)
>> attach(data)
>> names(data)
>  [1] "cross"     "line"      "forigin"   "morigin"
>  [6] "eggs"       "adults"
>
>> library(lme4)
>
>> y<-cbind(adults,(eggs-adults))
>
>> model1<-lmer(y~cross*line+(1|forigin)+(1|morigin),family=binomial)
>
> But I get the error message: In mer_finalize(ans) : gr cannot be computed at initial par (65)

> I suppose that I receive this error message because the experimental design is unbalanced, because as soon as I take out the interaction between the explanatory variables, R seems to be able to run the model. But it is precisely the interaction between „cross“ and „line“ that I am interested in.

It appears that you have the classic situation of a two-way layout
with missing cells.  This is an example of a situation where the
symbolic analysis performed by model.matrix does not detect rank
deficiency in the result.

The model with the term cross * line is equivalent to fitting a "cell
means" model which includes the interaction term cross:line but not
the main effects terms cross and line.  However, the model matrix for
cross:line will be constructed properly because it will drop the
unused levels (i.e. the missing cells) in the interaction factor.

So, try to fit with cross:line instead of cross*line.

> So I have tried a different approach, by coding the origin of each individual fly: instead of using “cross” and “line” I have tried using maternal grandmother, maternal grandfather, paternal grandmother and paternal grandfather, which combines both the information of “cross” and “line” and should avoid the problem of the unbalanced design.
>
> Again, after attaching the data I have been using the following commands:
>
>> names(data)
>  [1] "mat_granm"  "mat_granf" "pat_granm"  "pat_granf"  "forigin"  "morigin"  "eggs"  "adults"
>
>> y<-cbind(adults,eggs-adults)
>
>> model1<-lmer(y~mat_granm*mat_granf*pat_granm*pat_granf+(1|forigin)+(1|morigin),family=binomial)
>
> This again doesn’t work, but instead of getting an error message, R hangs itself up every time I attempt the analysis (I left it over night to see if it would just take that long, but the program doesn’t react).
>
> Any help would be greatly appreciated, and I would also be happy to make the data available for this purpose.
>
> Sol
>
>
>
> This email may be confidential and subject to legal privilege, it may
> not reflect the views of the University of Canterbury, and it is not
> guaranteed to be virus free. If you are not an intended recipient,
> please notify the sender immediately and erase all copies of the message
> and any attachments.
>
> Please refer to http://www.canterbury.ac.nz/emaildisclaimer for more
> information.
>
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>