[R-sig-ME] False Convergence Question

[Albyn Jones]

my guess:  because the estimated variance for PID is on the boundary
(ie 0).  maybe your model is over-parametrized, and the month:PID
interaction is doing all the work? 

Random effects:
  Groups                    Name        Variance Std.Dev.
  factor(Month):factor(PID) (Intercept) 3.857    1.964
  PID                       (Intercept) 0.000    0.000
  Actor2                    (Intercept) 0.757    0.870
  Actor1                    (Intercept) 0.550    0.742
  Month                     (Intercept) 0.333    0.577
 Number of obs: 5225, groups: factor(Month):factor(PID), 5225; PID, 666;
 Actor2, 36; Actor1, 36; Month, 8

On Thu, Aug 18, 2011 at 12:33:56PM -0400, Julie Rushmore wrote:
> Hi all,
> 
> I'm trying to model animal association patterns using a logistic mixed
> effects model with several random effects, and I keep getting the below
> non-convergence warning. Can anyone provide insight on what I am doing wrong
> or how I can get my model to converge? I assume this warning is
> problematic...does anyone have more insight?
> 
> *Warning message:*
> *In mer_finalize(ans) : false convergence (8)*
> 
> In case it is helpful, below is a description of my parameters, followed by
> my model script, and model output. Thanks for any feedback you can provide!
> 
> *Response variables*:
> SucG= number of times two animals (IDs) are observed together in a given
> month
> FailG= number of times one ID is observed without the other in a given month
> 
> *Fixed effects*:
> Age = categorical variable describing the age combination of two IDs
> (AA=adult-adult, AJ=adult-juvenile, JJ=juvenile-juvenile)
> Gender= categorical variable describing the gender combination of two IDs
> (MM=male-male, MF=male-female, FF=female-female)
> Related= categorical variable describing the relationship of two IDs
> (R=related, U=unrelated)
> PrevMonthG= continuous variable for the association index for two IDs in the
> previous month (standardized)
> Rainean2MA= continuous variable for the mean rainfall 2 months prior (a
> proxy for current fruit availability, standardized)
> EstrusAd3= categorical variable for the number of females that are sexually
> receptive in a given month
> FamilyDiff= continuous variable for the difference in family size between
> two IDs
> 
> *Random effects*:
> Month
> PID - Pair ID (i.e., the identifier for every combination of ID pairs)
> Actor1 - ID 1 involved in the association
> Actor2 - ID 2 involved in the association
> PID:Month - an observation-level random effect to account for
> over-dispersion
> 
> *Script:*
> mg=cbind(SucG,FailG)~Age+Gender+Related+PrevMonthGs+RainMean2MAs+factor(EstrusAd3)+FamilyDiff+(1|Month)+(1|PID)+(1|Actor1)+(1|Actor2)+(1|factor(Month):factor(PID))
> glmer(mg,data=standard,family=binomial(link='logit'),nAGQ=1)->SM12
> SM12
> 
> *Output: *
> Number of levels of a grouping factor for the random effects
> is *equal* to n, the number of observations
> Warning message:
> In mer_finalize(ans) : false convergence (8)
> > SM12
> Generalized linear mixed model fit by the Laplace approximation
> Formula: mg
>    Data: standard
>    AIC   BIC logLik deviance
>  25450 25555 -12709    25418
> Random effects:
>  Groups                    Name        Variance Std.Dev.
>  factor(Month):factor(PID) (Intercept) 3.857    1.964
>  PID                       (Intercept) 0.000    0.000
>  Actor2                    (Intercept) 0.757    0.870
>  Actor1                    (Intercept) 0.550    0.742
>  Month                     (Intercept) 0.333    0.577
> Number of obs: 5225, groups: factor(Month):factor(PID), 5225; PID, 666;
> Actor2, 36; Actor1, 36; Month, 8
> 
> Fixed effects:
>                    Estimate Std. Error z value Pr(>|z|)
> (Intercept)         -0.9884     0.5732   -1.72    0.085 .
> AgeAJ                0.2598     0.2099    1.24    0.216
> AgeJJ                0.4971     0.4117    1.21    0.227
> GenderMF             0.2440     0.2066    1.18    0.238
> GenderMM             0.5163     0.3936    1.31    0.190
> Related             -1.5553     0.1935   -8.04  9.2e-16 ***
> PrevMonthGs          1.0444     0.0395   26.42  < 2e-16 ***
> RainMean2MAs         0.6263     0.2531    2.47    0.013 *
> factor(EstrusAd3)1  -0.1683     0.5988   -0.28    0.779
> factor(EstrusAd3)2   0.4705     0.5870    0.80    0.423
> FamilyDiff          -0.0612     0.0317   -1.93    0.053 .
> ---
> Signif. codes:  0 ?***? 0.001 ?**? 0.01 ?*? 0.05 ?.? 0.1 ? ? 1
> 
> Correlation of Fixed Effects:
>             (Intr) AgeAJ  AgeJJ  GndrMF GndrMM Relatd PrvMnG RnM2MA f(EA3)1
> f(EA3)2
> AgeAJ
> -0.258
> AgeJJ       -0.254
> 0.949
> GenderMF    -0.343 -0.033
> -0.032
> GenderMM    -0.337 -0.038 -0.034
> 0.959
> Related     -0.286  0.059  0.036  0.004
> -0.010
> PrevMonthGs -0.047 -0.025 -0.032 -0.008 -0.018
> 0.277
> RainMen2MAs  0.304 -0.001 -0.001  0.000 -0.001  0.012
> 0.045
> fctr(EsA3)1 -0.657  0.002  0.002  0.001  0.001 -0.014 -0.051
> -0.536
> fctr(EsA3)2 -0.536  0.002  0.002  0.001  0.001 -0.014 -0.049 -0.106
> 0.542
> FamilyDiff   0.043 -0.104 -0.021  0.005  0.011 -0.313  0.089  0.004 -0.004
> -0.004
> 
> 
> -- 
> Julie Rushmore
> DVM/PhD Candidate
> College of Veterinary Medicine
> Odum School of Ecology
> University of Georgia
> Athens, GA 30605
> http://ecology.uga.edu/rushmore
> 
> 	[[alternative HTML version deleted]]
> 

> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models


-- 
Albyn Jones
Reed College
jones at reed.edu