[R-sig-ME] lme/lmer and random effects correlation parameters

Ben Bolker bbolker at gmail.com
Mon Nov 12 17:09:46 CET 2012


Antoine PACCARD <antoine.paccard at ...> writes:

> Dear Mix modelers,
> 
> I am trying to run the following models:
> 
> m1 <- lme(A ~ treatment * lat, random= ~ 1 + treatment | population/family) 
> 
> or with lmer
> 
> m2 <- lmer(A ~ treatment * lat + (treatment |population/family))

 
> Unfortunately m1 won't converge and I always get the following
> message: "iteration limit reached without convergence (10)".  I
> understand this comes from random correlation parameters and this
> issue has been previously discussed in the past. However, I can't
> find a way to make it work and I was wondering if anybody had found
> a way to fix this problem.

> lmer will work and converge properly. 

 I would be very careful about this statement.  Are you *sure* it
has converged properly?  Do the results make sense?  Do you get
non-trivial (not zero, not very very small, correlations not
equal to +/- 1.0) estimates for the variance-covariance matrix?

  Does centering the covariates help?

> Unfortunately the following function to extract p-values won't work
> because of the same reason explained above: pvals.fnc(m2, nsim =
> 10000) "Error in pvals.fnc(m2, nsim = 10000) : MCMC sampling is not
> implemented in recent versions of lme4 for models with random
> correlation parameters"

  http://glmm.wikidot.com/faq offers some alternatives.
I would consider parametric bootstrapping or Kenward-Roger
(perhaps using the functions in the pbkrtest package)
 
> I believe my models are correct since each family and each
> population have experienced treatment (when I get rid of "treatment"
> in the random parameters, the models work fine). Here is a sample of
> the data set:

  What are the missing (blank) values in the A column?
  Is that a glitch in the display, or are there really lots
of missing values?

> population	lat	family	treatment	A
> AA		37.724	35	AB		124
> LLL		37.419	14	AB		249
> AAJ		46.114	8	AB	
> XXX		43.356	1	CD	
> OOO		40.441	27	CD		244.5
> FFF		36.418	6	CD		240
> OOO		40.441	11	AB	
> AA		37.724	28	AB	
> AAF		46.114	17	CD		116.5
> QQQ		39.491	19	CD		93.5
> QQQ		39.491	21	AB		247
> AAF		46.114	10	CD	
> OOO		40.441	20	CD		227
> UUU		42.996	25	AB		95.5
> XXX		43.356	22	CD		209
> AA		37.724	15	AB	
> QQQ		39.491	29	AB		186
> UUU		42.996	19	AB		110
> UUU		42.996	3	AB		247



More information about the R-sig-mixed-models mailing list