Generalized linear mixed models: problem of interpretation

#1
Hi,

I am dealing with a problem of interpretation concerning an error message due to a generalized linear mixed model.
Datafile is attached to this message.

I try to model the percentage of flowering of different varietes of mustards in function of sum of degrees days (DEGJ_B3), variety (VAR_LIB), photoperiod (PTP).
In our field trials, we score the flowering of mustards at different dates.

Code:
library(lme4)

MYDATA[,7]=as.factor(MYDATA[,7])
class(MYDATA[,7])
attach(MYDATA)
Y=cbind(FLEURI,NON_FLEURI)

glmm1=glmer(Y~DEGJ_B3*VAR_LIB+PTP*VAR_LIB+(1|IND_ID),family="binomial",control=glmerControl(optCtrl=list(maxfun=2e6)))
summary(glmm1)
anova(glmm1,type='Chisq'))
And the following message appears:

Code:
Warning messages:
1: In checkConv(attr(opt, "derivs"), opt$par, ctrl = control$checkConv,  :
  Model failed to converge with max|grad| = 0.0593365 (tol = 0.001, component 1)
2: In checkConv(attr(opt, "derivs"), opt$par, ctrl = control$checkConv,  :
  Model is nearly unidentifiable: very large eigenvalue
 - Rescale variables?;Model is nearly unidentifiable: large eigenvalue ratio
 - Rescale variables?
I did few research on the internet:
To solve the first message, I tried to increase the number of iterations and to use a "bobyqa" optimization but without any result.
To solve the second issue, I have no idea how to proceed.
I do not understand very well (or at all...) the theory behind this kind of models so I am not very comfortable with making huge modifications in this model.

Do you have an idea of the problem behind these message errors?
Does it come from my dataset which is not good enough?

Do you confirm that I am in a situation in which I have to use a mixed model? A generalized linear model without considering a mixed effect shows good results... :)

Thanks a lot.

Paul.
 

Attachments

#2
If I include IND_ID as a fixed effect too (proposed by j58):
Code:
glmm1=glmer(Y~DEGJ_B3*VAR_LIB+PTP*VAR_LIB+IND_ID+(1|IND_ID),family="binomial",control=glmerControl(optCtrl=list(maxfun=2e6)))
I get this:
Code:
fixed-effect model matrix is rank deficient so dropping 22 columns / coefficients
Warning messages:
1: In checkConv(attr(opt, "derivs"), opt$par, ctrl = control$checkConv,  :
  unable to evaluate scaled gradient
2: In checkConv(attr(opt, "derivs"), opt$par, ctrl = control$checkConv,  :
  Model failed to converge: degenerate  Hessian with 3 negative eigenvalues
>
 

j58

Active Member
#3
Actually, I realized that was a dumb idea as soon as I posted it, and deleted my post—apparently not quickly enough. Sorry about that.