[R-sig-ME] singular convergence with lmer()

[Baldwin, Jim -FS]

I wonder if it is a version issue.  Using the data at forums.cirad.fr/logiciel-R/viewtopic.php?t=5071 I get the following (which matches what SAS produces):

> str(dat)
'data.frame':   192 obs. of  3 variables:
 $ Operator: Factor w/ 8 levels "A","B","C","D",..: 1 1 1 1 1 1 1 1 1 1 ...
 $ Part    : Factor w/ 12 levels "1","2","3","4",..: 1 1 2 2 3 3 4 4 5 5 ...
 $ y       : num  0.724 0.699 1.554 1.535 1.786 ...
> fit
Linear mixed model fit by REML
Formula: y ~ (1 | Operator) + (1 | Part) + (1 | Part:Operator)
   Data: dat
    AIC    BIC logLik deviance REMLdev
 -619.7 -603.4  314.9   -630.3  -629.7
Random effects:
 Groups        Name        Variance   Std.Dev.
 Part:Operator (Intercept) 0.00081854 0.028610
 Part          (Intercept) 1.06721993 1.033063
 Operator      (Intercept) 0.00031226 0.017671
 Residual                  0.00063295 0.025159
Number of obs: 192, groups: Part:Operator, 96; Part, 12; Operator, 8

Fixed effects:
            Estimate Std. Error t value
(Intercept)   2.7171     0.2983   9.109


I'm using R 1.15.0 32-bit on Windows XP and Package lme4 version 0.999375-42.

Jim Baldwin
Pacific Southwest Research Station
USDA Forest Service
Albany, California

-----Original Message-----
From: r-sig-mixed-models-bounces at r-project.org [mailto:r-sig-mixed-models-bounces at r-project.org] On Behalf Of Joshua Wiley
Sent: Sunday, July 08, 2012 1:52 PM
To: laurent stephane
Cc: r-sig-mixed-models at r-project.org
Subject: Re: [R-sig-ME] singular convergence with lmer()

Notice that the variance of one of your random effects is estimated at 0.  I suspect that this is the source of the singular convergence.
IIRC proc mixed (which is what I assume you are using in SAS) uses a somewhat different approach to to estimate the random effects than does lme4.

Although it seems to work for Reinhold, again some of the variances are vanishingly small, which seems to me like it may suggest some of the effects are borderline on 0 and perhaps slightly different estimation methods either get "really small" or simply "0" and if 0, you get a warning.  I would also consider simplifying your model (although likelihood ratio tests seem to suggest a significant decrement in the likelihood fixing the variance at 0).

Cheers,

Josh

On Thu, Jul 5, 2012 at 1:01 AM, laurent stephane <laurent_step at yahoo.fr> wrote:
> Dear all,
>
> Using the latest CRAN version of lme4 I get the following warning from lmer() :
>
> Warning message:
> In mer_finalize(ans) : singular convergence (7)
>
> My model is not complicated and it works fine with SAS (if you are
> interested in the details of my model see
> forums.cirad.fr/logiciel-R/viewtopic.php?t=5071 )
>
> What argument could I change in lmer() to overcome this warning ?
>
> Kind regards,
> SL
>
>         [[alternative HTML version deleted]]
>
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models



--
Joshua Wiley
Ph.D. Student, Health Psychology
Programmer Analyst II, Statistical Consulting Group University of California, Los Angeles https://joshuawiley.com/

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





This electronic message contains information generated by the USDA solely for the intended recipients. Any unauthorized interception of this message or the use or disclosure of the information it contains may violate the law and subject the violator to civil or criminal penalties. If you believe you have received this message in error, please notify the sender and delete the email immediately.