[R-sig-ME] Data structure issue for GLMM models
Thierry Onkelinx
thierry.onkelinx at inbo.be
Fri Nov 20 11:37:24 CET 2015
Dear Anne,
I've posted an Rpubs on adding a variable to both the fixed and the random
effects. Have a look at http://rpubs.com/INBOstats/both_fixed_random I
think that it answers your question.
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
2015-11-20 9:24 GMT+01:00 Anne Blach Overgaard <anne.overgaard op bios.au.dk>:
> Dear Thierry,
>
>
>
> Thank you very much for your reply.
>
> Below is the syntax of the base model with altitude included as fixed
> effect:
>
>
>
> GLMM.b.nan <- glmer(occ_Betula_nana ~ isotherm + Trange + Tsumm + Psumm +
> ddeg +
>
> sri + slope + mosses +
>
> bet.nan.bio + alt
>
> (1|fsite) + (1|fplotgr) + (1|fplot),
>
> data=env.sd.ran, family="poisson")
>
>
>
> The alternative is to include altitude both as fixed effect and as random
> effect due to the nested structure of the data
>
>
>
> GLMM.b.nan <- glmer(occ_Betula_nana ~ alt + isotherm + Trange + Tsumm +
> Psumm + ddeg +
>
> sri + slope + mosses +
>
> bet.nan.bio +
>
> (1|fsite) + (1|falt) + (1|fplotgr) +
> (1|fplot),
>
> data=env.sd.ran, family="poisson")
>
>
>
> The response: occ_Betula_nana, is count data i.e., the number of
> individuals registered in a pin point frame in each plot (ranges from 0-25)
>
> The fixed effects are represented by climate predictors: isotherm +
> Trange + Tsumm + Psumm + ddeg (isothermality, annual temperature range,
> summer temperature, summer precipitation and growing degree days) as well
> as sri (solar radiation index), slope, mosses (occurrence of mosses in the
> plots), bet.nan.bio ( a species-specific biotic-interaction variable we
> have computed), and finally alt (altitude m. asl).
>
>
>
> The random effects are site (fsite), plot group (fplotgr) and plots
> (fplot) and perhaps altitude (falt), but this is where we would appreciate
> some help to decide whether altitude should be include as fixed effect only
> , as random effect only or alternatively be included both as random and
> fixed effects. All random effects have been named individually which is why
> I haven’t used the code (1|fsite/fplotgr/falt), but they are nested.
>
>
>
> Site is named: S1-S5
>
> Alt is named according to site and altitude e.g., 1_20 (for site 1 at
> altitude 20) – NOTE: not all altitudes are present at each site!
>
> Plot group is named according to site, altitude and which of the three
> repetitions it represents e.g., 1_1_20 (for site 1, plot group no 1 at
> altitude 20)
>
> Plots are names (P1-P414)
>
> In total we have 5 sites x a varying number of altitudes per site x 3 plot
> groups per altitude x 6 plots = 414 plots in the entire data set.
>
>
>
> I hope this makes the issue a bit more clear
>
>
>
> Best regards,
>
>
>
> Anne
>
>
>
> *From:* Thierry Onkelinx [mailto:thierry.onkelinx op inbo.be
> <thierry.onkelinx op inbo.be>]
> *Sent:* 19 November 2015 15:01
> *To:* Anne Blach Overgaard
> *Subject:* Re: [R-sig-ME] Data structure issue for GLMM models
>
>
>
> Dear Anne,
>
>
>
> Can you provide the syntax of your base model and/or those of the models
> you are thinking about? And add a clear specification to the variable
> names. That would make your question much more clear and easier to answer.
>
>
>
> 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
>
>
>
> 2015-11-19 12:42 GMT+01:00 Anne Blach Overgaard <anne.overgaard op bios.au.dk
> >:
>
>
> Dear List,
> I hope that some of you may be able to help us with a data structure issue.
> We have collected plant cover data (count data) for selected species along
> a climatic gradient in random stratified sampling plots. The hierarchical
> structure of the data is as follows:
> We have sampled at five sites placed along a large-scale climatic
> gradient. Within each of the five sites we placed three plot groups 500
> meters apart on each of the altitudes 20 m 100, 200, 300, 400 and 500 m
> above sea level, whenever possible, as not all isoclines were present at
> each site. Each plot group consisted of six plots that were placed 10
> meters apart.
> In total we have 5 sites x a varying number of altitudes per site x 3 plot
> groups per altitude x 6 plots = 414 plots in the entire data set.
> Overall we would like to assess the relative importance of different
> predictor groups (altitude, climate, and biotic interactions) on the
> variation in cover per species. We are including the predictor groups as
> fixed effects in our models using lme4::glmer (family = poisson). We
> include site and plot group as nested random effects and plots as an
> observation-level random factor due to overdispersion in the data.
> Our question is whether altitude should be entered as a random factor, as
> a fixed effect, or possibly as both a fixed and a random effect. Altitude
> is a part of the nested structure of the data, but we also have an interest
> in including it as a fixed effect to assess how much of the variation in
> the data is due to altitude.
> We hope that some of you can guide us how to deal with altitude in this
> data analysis.
> Thanking you in advance.
> Best regards,
> Anne & co-workers
>
>
>
>
>
> [[alternative HTML version deleted]]
>
> _______________________________________________
> R-sig-mixed-models op r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>
>
>
[[alternative HTML version deleted]]
More information about the R-sig-mixed-models
mailing list