[R-sig-ME] Hierarchical structure for preservation of observations

[Jacob Bukoski]

Dear Thierry,

Thank you kindly for your advice. You were right in that the model is too
complex for the dataset. I began examining the intervals of the
coefficients and was getting "Non-positive definitive approximate
variance-covariance" errors. Hopefully I'll have more data that becomes
available in the future, but in the meantime will try to simplify the model.

Best wishes,
Jacob



On Mon, Oct 19, 2015 at 2:49 AM, Thierry Onkelinx <thierry.onkelinx at inbo.be>
wrote:

> Dear Jacob,
>
> The random effect specification is correct for the model that you have in
> mind.
>
> I'd rather think of the effects of site and genus as crossed rather
> nested. The formula would become (1|Site) + (1|Genus). Assuming that you
> have 5 or more genera. If not, then it better to add Genus to the fixed
> effects or keep your current model.
>
> I'm a bit worried about the complexity of your model. 65 observations is
> not a lot for a model with 6 parameters.
>
> 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
>
> 2015-10-18 16:02 GMT+02:00 Jacob Bukoski <jbukoski1 at gmail.com>:
>
>> Dear all,
>>
>> I am currently trying to develop a predictive model of ecosystem carbon
>> stocks for forests of Southeast Asia. The difficult part is that the
>> dataset I've compiled (from published data in the peer-reviewed
>> literature)
>> only consists of 65 or so observations.
>>
>> I've resorted to a mixed effects model under the understanding that I can
>> specify a grouping structure and maintain spatially correlated
>> observations
>> within each study. However, I'm not sure that my specification of the
>> hierarchical structure is acceptable, as this is my first attempt at using
>> mixed models.
>>
>> I am using R version 3.2.2., and the lmer() function of the "lme4"
>> package.
>>
>> I have specified basal area (a metric of tree stem cross sectional area at
>> 1.3 meters height), latitude, and categorical variables for small (mean
>> stem diameter < 5 cm) and large (mean stem diameter > 15 cm) forests as
>> fixed effects.
>>
>> I have specified the dominant genus of tree in plots and the site as
>> random
>> effects. My thinking here is that plots with the same dominant genera of
>> tree would not be independent of one another, nor would plots within the
>> same site. Thus, by specifying random effects for Genus in Site... I can
>> maintain the individual observations.
>>
>> The model I have specified as follows:
>>
>> *lmer1 <- lmer(Biomass ~ Basal.area + Latitude + Small + Large +
>> (1|Site/Genus), REML=FALSE)*
>>
>> *My two-part question is*: (i) Does the logic behind my random effect
>> specification hold, and (ii) have I specified it under the lmer() function
>> correctly?
>>
>> Additionally, I am unsure of whether the number of groups has an impact on
>> the degrees of freedom. The summary reports 34 groups for Genus:Site, and
>> 21 for Site. My specified model has 40 residual degrees of freedom. Have I
>> violated my degrees of freedom?
>>
>> Any advice and external resources will be hugely appreciated!
>>
>> Many kind thanks,
>> Jacob
>>
>> --
>> Jacob J. Bukoski
>> Master of Environmental Science Candidate, 2016
>> School of Forestry and Environmental Studies, Yale University
>> jbukoski1 at gmail.com | jacob.bukoski at yale.edu | LinkedIn
>> <
>> https://www.linkedin.com/profile/view?id=AAIAAAdWVW8BMzqU_2EGNbEkyuy8O7K1Jyhd8ps&trk=nav_responsive_tab_profile_pic
>> >
>>
>>         [[alternative HTML version deleted]]
>>
>> _______________________________________________
>> R-sig-mixed-models at r-project.org mailing list
>> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>>
>
>


-- 
Jacob J. Bukoski
Master of Environmental Science Candidate, 2016
School of Forestry and Environmental Studies, Yale University
jbukoski1 at gmail.com | jacob.bukoski at yale.edu | LinkedIn
<https://www.linkedin.com/profile/view?id=AAIAAAdWVW8BMzqU_2EGNbEkyuy8O7K1Jyhd8ps&trk=nav_responsive_tab_profile_pic>

	[[alternative HTML version deleted]]