[R] lmer - error asMethod(object) : matrix is not symmetric

[Douglas Bates]

On Tue, Feb 16, 2010 at 10:54 AM, Luisa Carvalheiro
<lgcarvalheiro at gmail.com> wrote:
> Dear Douglas,
>
> Thank you for your reply.
> Just some extra info on the dataset: In my case Number of obs is 33,
> and number of groups of factor(Farm_code) is 12.
> This is the information on iterations I get:
>
> summary(lmer(round(SR_SUN)~Dist_NV + (1|factor(Farm_code)) ,
> family=poisson, verbose =TRUE))
>  0:     60.054531:  1.06363  2.14672 -0.000683051
>  1:     60.054531:  1.06363  2.14672 -0.000683051
> Error in asMethod(object) : matrix is not symmetric [1,2]
> In addition: Warning message:
> In mer_finalize(ans) : singular convergence (7)

> When I run a similar model (exp variable Dist_hives) the number of
> iterations is 11:

>  summary(lmer(round(SR_SUN)~Dist_hives + (1|factor(Farm_code)) ,
> family=poisson, verbose =TRUE))
>  0:     61.745238: 0.984732  1.63769 0.000126484
>  1:     61.648229: 0.984731  1.63769 -2.08637e-05
>  2:     61.498777: 0.984598  1.63769 4.11867e-05
>  3:     47.960908: 0.381062  1.63585 6.77029e-05
>  4:     46.223789: 0.250732  1.66727 8.31854e-05
>  5:     46.222223: 0.250732  1.66727 6.97790e-05
>  6:     46.216710: 0.250730  1.66727 7.60560e-05
>  7:     46.168835: 0.230386  1.64883 9.16430e-05
>  8:     46.165955: 0.228062  1.65658 8.70694e-05
>  9:     46.165883: 0.228815  1.65737 8.63400e-05
>  10:     46.165883: 0.228772  1.65734 8.63698e-05
>  11:     46.165883: 0.228772  1.65734 8.63701e-05

> I am very confused with the fact that it runs with Dist_hives and not
> with Dist_NV. Both variables are distance values, the first having no
> obvious relation with the response variable and the second (Dist_NV)
> seems to have a negative effect on SR_SUN.

As you say, Dist_hives has very little relationship to the response
variable.  The two fixed-effects coefficients are the last two
parameters in the iteration output (the first parameter is the
standard deviation of the random effects).  So the slope with respect
to Dist_hives for the linear predictor is 0.0000863.  Either you have
very large magnitudes of Dist_hives or that variable does not have
much predictive power.

For the second (Dist_NV) variable, the optimization algorithm is not
able to make progress from the starting estimates.  This may be an
indication that the problem is badly scaled.  Are the values of
Dist_NV very large?  If so, you may want to change the unit (say from
meters to kilometers) so the values are much smaller.

It may also help to use a starting estimate for the standard deviation
of the random effects derived from the other model.  That is, include
start = 0.22 in the call to lmer.

> Does this information helps identifying the problem with my data/analysis?
>
> Thank you,
>
> Luisa
>
>
>
>
> On Tue, Feb 16, 2010 at 5:35 PM, Douglas Bates <bates at stat.wisc.edu> wrote:
>> This is similar to another question on the list today.
>>
>> On Tue, Feb 16, 2010 at 4:39 AM, Luisa Carvalheiro
>> <lgcarvalheiro at gmail.com> wrote:
>>> Dear R users,
>>>
>>> I  am having problems using package lme4.
>>>
>>> I am trying to analyse the effect of a continuous variable (Dist_NV)
>>> on a count data response variable (SR_SUN) using Poisson error
>>> distribution. However, when I run the model:
>>>
>>> summary(lmer((SR_SUN)~Dist_NV + (1|factor(Farm_code)) ,
>>> family=poisson, REML=FALSE))
>>>
>>> 1 error message and 1 warning message show up:
>>>
>>> in asMethod(object) : matrix is not symmetric [1,2]
>>> In addition: Warning message:
>>> In mer_finalize(ans) : singular convergence (7)
>>
>> So the first thing to do is to include the optional argument verbose =
>> TRUE in the call to lmer.  (Also, REML = FALSE is ignored for
>> Generalized Linear Mixed Models and can be omitted. although there is
>> no harm in including it.)
>>
>> You need to know where the optimizer is taking the parameter values
>> before you can decide why.
>>
>> P.S. Questions like this will probably be more readily answered on the
>> R-SIG-Mixed-Models mailing list.
>>
>>> A model including  Dist_NV together with other variables runs with no problems.
>>> What am I doing wrong?
>>>
>>> Thank you,
>>>
>>> Luisa
>>>
>>>
>>> --
>>> Luisa Carvalheiro, PhD
>>> Southern African Biodiversity Institute, Kirstenbosch Research Center, Claremont
>>> & University of Pretoria
>>> Postal address - SAWC Pbag X3015 Hoedspruit 1380, South Africa
>>> telephone - +27 (0) 790250944
>>> Carvalheiro at sanbi.org
>>> lgcarvalheiro at gmail.com
>>>
>>> ______________________________________________
>>> R-help at r-project.org mailing list
>>> https://stat.ethz.ch/mailman/listinfo/r-help
>>> PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
>>> and provide commented, minimal, self-contained, reproducible code.
>>>
>>
>
>
>
> --
> Luisa Carvalheiro, PhD
> Southern African Biodiversity Institute, Kirstenbosch Research Center, Claremont
> & University of Pretoria
> Postal address - SAWC Pbag X3015 Hoedspruit 1380, South Africa
> telephone - +27 (0) 790250944
> Carvalheiro at sanbi.org
> lgcarvalheiro at gmail.com
>