[R-sig-ME] lme4 glmer convergence question

Ben Bolker bbolker at gmail.com
Fri Nov 6 22:04:03 CET 2015


On Fri, Nov 6, 2015 at 1:58 PM, Sally A. Roman <saroman at vims.edu> wrote:
> Hello -
> I am trying to use the lme4 package for a glmm and am getting a convergence code of 0 and a statement: Model failed to converge with max|grad| = 0.00791467 (tol = 0.001, component 1). I am interested in using the lme4 package because I would like to have AIC values to determine the appropriate model as I add in additional covariates.

 In contrast to glmmPQL, I'm guessing?

> Two weeks ago when I tried the same approach I got a warning message that the model failed to converge because of the max|grad| issue, but am not getting the warning message this time, just the statement at the end of the summary output.
>
> Summary output below:
> Generalized linear mixed model fit by maximum likelihood (Laplace
>   Approximation) [glmerMod]
> Family: Gamma  ( log )
> Formula: Meat_Weight ~ logsh + SAMS_region_2015 + (1 | StationID)
>    Data: datad
> Control: glmerControl(optCtrl = list(maxfun = 100000))
>
>     AIC      BIC   logLik deviance df.resid
>  29841.2  29893.0 -14912.6  29825.2     4748
>
> Scaled residuals:
>     Min      1Q  Median      3Q     Max
> -5.6389 -0.5089  0.0376  0.5660  7.3199
>
> Random effects:
> Groups    Name        Variance Std.Dev.
> StationID (Intercept) 0.007073 0.0841
>  Residual              0.026626 0.1632
> Number of obs: 4756, groups:  StationID, 426
>
> Fixed effects:
>                      Estimate Std. Error t value             Pr(>|z|)
> (Intercept)          -9.21833    0.10890  -84.65 < 0.0000000000000002
> logsh                 2.62984    0.02223  118.33 < 0.0000000000000002
> SAMS_region_2015ET    0.09299    0.03174    2.93             0.003393
> SAMS_region_2015HC    0.12031    0.03347    3.59             0.000325
> SAMS_region_2015HCsr  0.08405    0.03892    2.16             0.030810
> SAMS_region_2015LI    0.07721    0.03209    2.41             0.016107
>
> (Intercept)          ***
> logsh                ***
> SAMS_region_2015ET   **
> SAMS_region_2015HC   ***
> SAMS_region_2015HCsr *
> SAMS_region_2015LI   *
> ---
> Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
>
> Correlation of Fixed Effects:
>                  (Intr) logsh  SAMS__2015E SAMS_regn_2015HC
> logsh            -0.968
> SAMS__2015E      -0.179 -0.036
> SAMS_regn_2015HC -0.166 -0.038  0.699
> SAMS_rgn_2015HCs -0.139 -0.037  0.600       0.569
> SAMS__2015L      -0.186 -0.027  0.728       0.691
>                  SAMS_rgn_2015HCs
> logsh
> SAMS__2015E
> SAMS_regn_2015HC
> SAMS_rgn_2015HCs
> SAMS__2015L       0.593
> convergence code: 0
> Model failed to converge with max|grad| = 0.00764043 (tol = 0.001, component 1)
>
> Does this mean that the model is not converging? I also used the glmmPQL method. The coefficient parameter estimates are similar between the two model types.
>
> Here is glmer (lme4) model code. I increased the maxfun to deal with other issues I had when I ran the model last time.
>
> l1<-glmer(Meat_Weight~logsh+SAMS_region_2015+(1|StationID),
>         family="Gamma"(link="log"),data=datad,control=glmerControl(optCtrl=list(maxfun=100000)))
> Here is the glmmPQL code.
>
> m1<-glmmPQL(fixed=Meat_Weight~logsh+SAMS_region_2015,random=~1|StationID,
>         family=Gamma(link="log"),data=datad)
>
> I am sure this is not information to diagnosis the problem, but if anyone has suggestions I can provide more data.
>
> Thanks
>
> Sally Roman
> Fisheries Specialist
> Virginia Institute of Marine Science
> Marine Advisory Services
>
> Phone: 804-684-7165
> Fax: 804-684-7161


  Have you looked at ?convergence ?  The bottom line (as commented on
here recently by Doug Bates in this forum) is that the convergence
tests give a lot of false positives; I have thought a lot about
removing them, or at least about increasing the tolerances
considerably, but have been afraid to make changes that would lead to
a lot more false *negatives* (i.e. unreported problems with models)
without a lot more time & effort evaluating these rules and makng the
decision carefully (which I don't have right now ...)

  Especially if you are getting similar-enough results between glmmPQL
and glmer, I would feel free to ignore the warnings.



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