[R-sig-ME] Large variance and SD for random effects

[cumuluss]

Dear Thierry,
thank you for your answer and especially for the quotation.
I will follow your suggestion.
Best regards
Paul



Thierry Onkelinx:
> Dear Paul,
> 
> Yes, I see no point in keeping an unstable model. Remember "All models are
> wrong but some are useful" (Box, 1979). I'd investigate where the complete
> separation occurs and decide which random slope should be removed.
> 
> In case of the strong correlations among random effects see
> https://www.muscardinus.be/2018/02/highly-correlated-random-effects/
> 
> Best regards,
> 
> 
> ir. Thierry Onkelinx
> Statisticus / Statistician
> 
> Vlaamse Overheid / Government of Flanders
> INSTITUUT VOOR NATUUR- EN BOSONDERZOEK / RESEARCH INSTITUTE FOR NATURE AND
> FOREST
> Team Biometrie & Kwaliteitszorg / Team Biometrics & Quality Assurance
> thierry.onkelinx at inbo.be
> Havenlaan 88 bus 73, 1000 Brussel
> www.inbo.be
> 
> ///////////////////////////////////////////////////////////////////////////////////////////
> To call in the statistician after the experiment is done may be no more
> than asking him to perform a post-mortem examination: he may be able to say
> what the experiment died of. ~ Sir Ronald Aylmer Fisher
> The plural of anecdote is not data. ~ Roger Brinner
> The combination of some data and an aching desire for an answer does not
> ensure that a reasonable answer can be extracted from a given body of data.
> ~ John Tukey
> ///////////////////////////////////////////////////////////////////////////////////////////
> 
> <https://www.inbo.be>
> 
> 2018-03-06 11:02 GMT+01:00 cumuluss <cumuluss at gmx.de>:
> 
>> Dear Thierry,
>> sorry one more question. i would like to ask whether you could give me
>> some recommendation for my model. Would you skip random effects (or
>> slopes) even if you think they are necessary, as far as they lead to
>> suffering models? It is maybe a more general question.
>> Thank you!
>>
>>
>>
>>
>> Dear Thierry,
>> thank you for your answer!
>> Ok then I have to rethink the model.
>> Best regards
>> Paul
>>
>>
>>
>> Thierry Onkelinx:
>>> Dear Paul,
>>>
>>> Your random effect structure looks quite complicated. Maybe too complex
>> for
>>> the data. Your model is very likely suffering from (quasi) complete
>>> separation.
>>>
>>> Besides the large variances, you should also be alarmed by the near
>> perfect
>>> correlations among some random effects.
>>>
>>> Best regards,
>>>
>>>
>>>
>>> ir. Thierry Onkelinx
>>> Statisticus / Statistician
>>>
>>> Vlaamse Overheid / Government of Flanders
>>> INSTITUUT VOOR NATUUR- EN BOSONDERZOEK / RESEARCH INSTITUTE FOR NATURE
>> AND
>>> FOREST
>>> Team Biometrie & Kwaliteitszorg / Team Biometrics & Quality Assurance
>>> thierry.onkelinx at inbo.be
>>> Havenlaan 88 bus 73, 1000 Brussel
>>> www.inbo.be
>>>
>>> ////////////////////////////////////////////////////////////
>> ///////////////////////////////
>>> To call in the statistician after the experiment is done may be no more
>>> than asking him to perform a post-mortem examination: he may be able to
>> say
>>> what the experiment died of. ~ Sir Ronald Aylmer Fisher
>>> The plural of anecdote is not data. ~ Roger Brinner
>>> The combination of some data and an aching desire for an answer does not
>>> ensure that a reasonable answer can be extracted from a given body of
>> data.
>>> ~ John Tukey
>>> ////////////////////////////////////////////////////////////
>> ///////////////////////////////
>>>
>>> <https://www.inbo.be>
>>>
>>> 2018-03-05 20:24 GMT+01:00 cumuluss <cumuluss at gmx.de>:
>>>
>>>> Hello,
>>>> the result of my GLMM with binomial error structure revealed for one of
>>>> the random intercepts and slopes variances and Sd's larger then 1
>>>>
>>>>> Generalized linear mixed model fit by maximum likelihood (Laplace
>>>> Approximation) ['glmerMod']
>>>>>  Family: binomial  ( logit )
>>>>> Formula: obs.yn ~ z.lengthxyz + z.obsX + Spe_tr_subspecie + (1 |
>> commu) +
>>>>>     (1 + z.lengthxyz + z.obsX | Siteun) + (1 + z.lengthxyz +
>>>>>     z.obsX + Spe_tr_subspecie_a.c + Spe_tr_subspecie_b.c +
>>>>>     Spe_tr_subspecie_c.c | behavior)
>>>>
>>>>> Random effects:
>>>>>  Groups   Name                  Variance Std.Dev. Corr
>>>>>  commu    (Intercept)           0.614004 0.78358
>>>>>  Siteun   (Intercept)           0.521198 0.72194
>>>>>           z.lengthxyz           0.001016 0.03188   1.00
>>>>>           z.obsX                0.306931 0.55401  -0.17 -0.19
>>>>>  behavior (Intercept)           2.966139 1.72225
>>>>>           z.lengthxyz           0.030171 0.17370   0.77
>>>>>           z.obsX                0.412169 0.64200   0.20 -0.36
>>>>>           Spe_tr_subspecie_a.c  1.853903 1.36158   0.58  0.62  0.11
>>>>>           Spe_tr_subspecie_b.c  3.973439 1.99335   0.51  0.23  0.25
>> 0.65
>>>>>           Spe_tr_subspecie_c.c  7.401343 2.72054   0.39  0.30  0.47
>>>> 0.79  0.60
>>>>> Number of obs: 4413, groups:  commu, 144; Siteun, 108; behavior, 31
>>>>
>>>> Now I wonder, whether that is a reason to worry, that the result could
>>>> be not valid?
>>>> Thanks in advance for any comments!
>>>> Paul
>>>>
>>>> _______________________________________________
>>>> R-sig-mixed-models at r-project.org mailing list
>>>> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>>>>
>>>
>>>       [[alternative HTML version deleted]]
>>>
>>> _______________________________________________
>>> R-sig-mixed-models at r-project.org mailing list
>>> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>>>
>>
> 
> 	[[alternative HTML version deleted]]
> 
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>