[R-sig-ME] Modelling non-negative non-zero continuous data

[Victoria Pattison-Willits]

Hi Thierry

Thank you so much for the advice and link to your blog post. I have
investigated all the different random effects structures including
modelling year as either a fixed or random variable. depending on the
response variable I  am dealing with, model fit and validation is most
often better modelled when it is a random effect … but will certainly look
into modelling it as a fixed effect again. And to confirm Yes all
exploratory analyses were done and all covariates are fine! And all have a
linear relationship.

Thanks so much again your advice is most welcome and appreciated!

Best wishes

Vicki
On Mon, May 2, 2022 at 3:43 AM, Thierry Onkelinx <thierry.onkelinx using inbo.be>
wrote:

> Dear Vicki,
>
> When you have only one measurement per nest box, then you can't have
> "nest box" as a random effect as it would confound with the residuals.
> I recommend adding "year" as a fixed effect factor. I wrote a blog post on
> the required number of levels for a random effect:
> https://www.muscardinus.be/2018/09/number-random-effect-levels/
> I presume you did an exploratory data analysis and handle covariates with
> strong correlation. Do all covariates have a linear relationship with the
> response?
>
> 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 using inbo.be
> Havenlaan 88
> <https://www.google.com/maps/search/Havenlaan+88?entry=gmail&source=g>
> 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>
>
>
> Op vr 29 apr. 2022 om 20:58 schreef Victoria Pattison-Willits <
> victoriaswillits using gmail.com>:
>
>> Dear all,
>>
>> I am hoping you can help me. I am trying to model chick tarsi (leg length)
>> data. Briefly, I have mean measurements of tarsus length from 457 nests.
>> The data were collected across 31 sites (10 nests in each site) over a
>> six-year period so I have an *a priori *nested random effects structure:
>
>
>> (1|SITE_ID/BOX_NUMBER) + (1|YEAR). (Although  I have had to remove the
>> nested nest_box term due to convergence issues - there is a lot of
>> variance
>> between nests within sites.)
>>
>> The problem that I am running into is that the data is bound between the
>> values 12.73 and 20.12 mm. Both the data itself and the residuals from a
>> lmer model are left-skewed because the data is non-negative and non-zero.
>>
>> The initial suite of models I have tried follows the below: I am running
>> models using both glmmTMB and lme (lmer).
>>
>> (Also I have run the same models using the same length data for a bunch of
>> other response variables with no issues including various breeding
>> outcomes
>> and chick measurements). Fixed covariates are scaled and centred: (sc.)
>> e.g.
>> ```{r}
>> TL_BUILT_FULL_TMB2<-
>>
>> glmmTMB(me_TARSUS~sc.BUILT_PERCENT+sc.GARDEN_PERCENT+sc.RINGING_AGE+sc.AprHatchDate+sc.BROOD_ATRINGING+(1|SITE_ID)+(1|YEAR),
>> data=DF_CHICK_TARSUS, family = gaussian)
>>
>> summary(TL_BUILT_FULL_TMB2)
>> ```
>> I am a little stumped as to what to do - I have run the same model using
>> reflected and log (and/or square root) transformed data - which does seem
>> to resolve the residual issues. However, I know that this is not the best
>> resolution and is rarely done, and transforming data even for the more
>> commonly found right-skewed data is increasingly discouraged. However, I
>> am
>> not finding (and this may be me not using the correct terms in my search!)
>> any other options to overcome the issue of non-negative non-zero data -
>> plenty of advice for ecological data that is right-skewed or left-skewed
>> and zero-inflated!
>>
>> If anyone can help me I would really appreciate it. Thank you all so much
>> as always in advance for your time and knowledge sharing. I am gradually
>> building up my competence in R and mixed modelling and this forum has been
>> really helpful on this steep learning curve! I am hoping I am just missing
>> something obvious! Please let me know if you need any other information
>> from me. Thank you!
>>
>> Very best wishes.
>>
>> Vicki Willits
>>
>>
>> >
>> > _______________________________________________
>> > R-sig-mixed-models using r-project.org mailing list
>> > https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>> >
>>
>>         [[alternative HTML version deleted]]
>
>
>>
>> _______________________________________________
>> R-sig-mixed-models using r-project.org mailing list
>> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>>
>

	[[alternative HTML version deleted]]