[R-sig-ME] clustered data with glmer() and glmmPQL()

[Thierry Onkelinx]

You can use tank as a fixed effect instead of a random effect. In that case
your model reduces to a general linear model. Personally I prefer a
likelihood based model (glmer) over a penalised quasi-likelihood model
(glmmPQL) unless I need things that are not available with glmer.

You need to use the log(weight) of the fish as an offset factor instead of
calculating the ratio.

ir. Thierry Onkelinx
Instituut voor natuur- en bosonderzoek / Research Institute for Nature and
Forest
team Biometrie & Kwaliteitszorg / team Biometrics & Quality Assurance
Kliniekstraat 25
1070 Anderlecht
Belgium

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

2015-06-16 1:54 GMT+02:00 Marsela Alvanopoulou <marselalv op gmail.com>:

> Hi again,
>
> Thank you very much for your response. I found most of the answers in glmm
> wiki faq. I used MASS::glmmPQL for the model, car::Anova and multcomp::glht
> for the hypothesis testing and I still need some work to check the effect
> of the tank, if any.
>
> I also want to check if there is a significant difference on the number of
> parasites per gram (continuous response variable). I multiplied all values
> by 100 to get a discrete variable like before. Does that affect the final
> conclusions?
>
> Thanks again!
>
> On Mon, Jun 15, 2015 at 4:34 PM, Thierry Onkelinx <
> thierry.onkelinx op inbo.be> wrote:
>
>> Dear Maria,
>>
>> The assumption of normality is only required for the residuals of linear
>> (mixed) models, not for the residuals of generalised linear (mixed) models.
>>
>> You can't use aov() for two reasons: it assumes a Gaussian distribution
>> and it assumes independent observations.
>>
>> mod1 and mod2 are in principle the same model (but fitted differently).
>> Both assume the same correlation structure.
>>
>> 3 levels is not enough to get a sensible variance estimate for a random
>> effect. See glmm wiki faq for more details.
>>
>> Best regards,
>>
>> ir. Thierry Onkelinx
>> Instituut voor natuur- en bosonderzoek / Research Institute for Nature
>> and Forest
>> team Biometrie & Kwaliteitszorg / team Biometrics & Quality Assurance
>> Kliniekstraat 25
>> 1070 Anderlecht
>> Belgium
>>
>> 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
>>
>> 2015-06-15 15:06 GMT+02:00 Marsela Alvanopoulou <marselalv op gmail.com>:
>>
>>> Hello,
>>>
>>> I'm
>>> ​ ​
>>> a master student from
>>> ​ ​
>>> Greece. I´m trying to model count data with
>>> ​ ​
>>> GLMM (lme4
>>> ​ ​
>>> package), using as discrete response variable the number of parasites per
>>> fish and as categorical predictor variable three
>>> ​ ​
>>> different species.
>>> ​ ​
>>> I'm using as random effect the three different tanks I used and as fixed
>>> the infection level
>>> ​​
>>> .
>>> ​
>>> ​
>>> This is the model I'm running:
>>>
>>> mod
>>> ​ ​
>>> <-
>>> ​ ​
>>> glmer
>>> ​ ​
>>> (parasite~species+(1|tank),
>>> ​ ​
>>> family=poisson
>>> ​​
>>> , data=mydata)
>>>
>>> I noticed that the estimate of the intercept does not give the mean of
>>> the
>>> first species, so I ran a simple glm model to get the estimate. With
>>> summary() I got the p values that allow me to reject my hypothesis and
>>> continue
>>> ​ ​
>>> to the Tukey test. Is it legal to use
>>> ​
>>> TukeyHSD(aov(parasite~species, data=mydata))
>>> ​ ?
>>> ​
>>> ​
>>> Finally I tested the assumptions
>>> ​ ​
>>> and
>>> ​ ​
>>> I found violation of normality and independence.
>>>
>>> I also tried MASS package where the assumption of independent residuals
>>> was
>>> not violated anymore but the histogram gave me a much more skewed
>>> distribution, but also anova() is not available for QTLs.
>>>
>>> mod2 <- glmmPQL (parasite~species, random=~1|tank, family=poisson,
>>> data=mydata)
>>>
>>> Thank you in advance for your help.
>>>
>>> Maria
>>>
>>>         [[alternative HTML version deleted]]
>>>
>>> _______________________________________________
>>> R-sig-mixed-models op r-project.org mailing list
>>> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>>>
>>
>>
>
>
> --
> Marsela Alvanopoulou
> MSc Student, Biology Dept.,
> University of Bergen, Norway
>
> *e-mail: Marsela.Alvanopoulou op student.uib.no
> <Marsela.Alvanopoulou op student.uib.no>*
> *e-mail: marsela.alvanopoulou op imr.no <marsela.alvanopoulou op imr.no>*
> *linkedin: gr.linkedin.com/pub/marsela-alvanopoulou/69/3b3/410/
> <http://gr.linkedin.com/pub/marsela-alvanopoulou/69/3b3/410/>*
>

	[[alternative HTML version deleted]]