Dear Ludovico, "the likelihood of colonizing or becoming extinct" is not a binary response as such. For each of your plant species, you actually can run one of two models : "Species A persists on plot YX or goes extinct" or "Species B colonizes plot xy or is still absent". Otherwise you would ignore that a species can be present during both surveys. Also, your "PROB_COL" is not really a probability. It is just the number of species with a certain trait that showed up only during the second survey divided by 24. If you would calculate this for each species separately, you would get a fraction (still not a probability). To me your data look like it should be analysed with some other method. GLMs are more convincing and easier to interpret/plot if your predictors are continuous (e.g., area, elevation, site age, pH, vegetation cover, etc) and not a factor with 12 levels when you have only 24 plots. More importantly, don't forget that the probability of a species colonizing a plot not only depends on which life form that particular species has, but also on how many species with the same life form are the regional species pool, i.e. in the draw. HTH, Claas Claas Damken, Postdoctoral Fellow Institute for Biodiversity and Environmental Research Universiti Brunei Darussalam Brunei Darussalam email: c.damken at auckland.ac.nz ________________________________________ Today's Topics: 1. Binomial GLM: Is it the right way? (Ludovico Frate) ---------------------------------------------------------------------- Message: 1 Date: Fri, 31 Oct 2014 21:49:26 +0100 From: Ludovico Frate <ludovicofrate at hotmail.it> To: "r-sig-ecology at r-project.org" <r-sig-ecology at r-project.org> Subject: [R-sig-eco] Binomial GLM: Is it the right way? Message-ID: <DUB126-W76C71CD937F919A537C3ECD69A0 at phx.gbl> Content-Type: text/plain; charset="UTF-8" Dear List, I'am trying to fit a GLM with a binomial error distribution but I am a little bit confused about the model. I have 24 vegetation permanent plots, randomly distributed, sampled two times: 1962 and 2009. I am looking for any changes in vegetation composition and in particular based on plant traits (life forms).For each species, I assessed whether it had colonized (i.e. been in 1962 absent but in 2009 present) or become locally extinct (i.e. been in 1962 present but recently absent) in the plot.I would like to analyze whether a species trait affected the likelihood of colonizing or becoming extinct on a summit. example: Colonizing species Traits TOTAL COLONIZING PROB_ COLChF 24 0 0 ChC 24 1 0,041666667 ChR 24 11 0,458333333 ChSC 24 1 0,041666667 ChS 24 18 0,75 Ph 24 2 0,083333333 Geo 24 18 0,75 HeB 24 1 0,041666667 HeC 24 36 1,5 HeR 24 47 1,958333333 HeS 24 161 6,708333333 Th 24 28 1,166666667 Traits is the plant trait analyzed, TOTAL is the number of plots and COLONIZING represents the number of new species found in 2009 per plot, PROB_COL is the probability of colonization (COLONIZING/TOT) However PROB_COL is >1, so I think that TOTAL should be different (i.e. total number of colonizing species)? Any suggestions?Thank you,Ludovico Ludovico Frate PhD student (University of Molise - Italy) Environmetrics Lab http://www.distat.unimol.it/STAT/environmetrica/organico/collaboratori/ludovico-frate-1 Department of Biosciences and Territory - DiBT Universit?el Molise. Contrada Fonte Lappone, 86090 - Pesche (IS) ITALIA. Cel: ++39 3333767557 Fax: ++39 (0874) 404123 E-mail ludovico.frate at unimol.it ludovicofrate at hotmail.it ------------------------------ _______________________________________________ R-sig-ecology mailing list R-sig-ecology at r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-ecology End of R-sig-ecology Digest, Vol 80, Issue 1 ********************************************
R-sig-ecology Digest, (80/1): Binomial GLM: Is it the right way?
3 messages · Claas Damken, Ludovico Frate, Abdoul Dia
Dear Class, thank you for your reply. Maybe I have not been very clear in the exposition. I'll try to reformulate. "the likelihood of colonizing or becoming extinct" is a proportion so the binomial distribution should be still ok! Considering only colonizing species, or arrivals, for a given plot. This means that a species is not new in overall sampled flora but only for that plot (i.e. *local* colonizer). PLOTID;TRAITS;NUMBER OF NEW COLONIZERS;NUMBER OF SUCCESS;PROPORTION 1; ChF; 4; 0; 0 1; ChC; 4; 0; 0 1; ChR; 4; 0; 0 1; ChSC;4; 0; 0 1; ChS; 4; 1; 0,25 1; Ph; 4; 0; 0 1; Geo; 4; 0; 0 1; HeB; 4; 0; 0 1; HeC; 4; 0; 0 1; HeR; 4; 0; 0 1; HeS; 4; 3; 0,75 1; Th; 4; 0; 0 2; ChF; 17; 0; 0 2; ChC; 17; 0; 0 2; ChR; 17; 1; 0,058823529 2; ChSC;17; 0; 0 2; ChS; 17; 2; 0,117647059 2; Ph; 17; 0; 0 2; Geo; 17; 1; 0,058823529 2; HeB; 17; 0; 0 2; HeC; 17; 0; 0 2; HeR; 17; 2; 0,117647059 2; HeS; 17; 7; 0,411764706 2; Th; 17; 4; 0,235294118 3; ChF; 16; 0; 0 3; ChC; 16; 0; 0 3; ChR; 16; 2; 0,125 3; ChSC;16; 0; 0 3; ChS; 16; 0; 0 3; Ph; 16; 0; 0 3; Geo; 16; 2; 0,125 3; HeB; 16; 0; 0 3; HeC; 16; 0; 0 3; HeR; 16; 1; 0,0625 3; HeS; 16; 9; 0,5625 3; Th; 16; 2; 0,125 . . . . . . . . . . . . 24 . . . for example, in the table, the plot 1 has 4 new species found of which 1 is a ChS and 3 are HeS with a proportion of 0.25 and 0.75 respectively. In a GLM model, the response variable should be PROPORTION (that I have improperly called PROBABILITY) An hypothetical model could be: M0<-glm(PROPORTION~TRAITS, family = binomial, weights = NUMBER OF NEW COLONIZERS, data = ...) Does it make sense? For the second question, I agree with you that "the probability of a species colonizing a plot not only depends on which life form that particular species has". Regards, Ludovico -- View this message in context: http://r-sig-ecology.471788.n2.nabble.com/Re-R-sig-ecology-Digest-80-1-Binomial-GLM-Is-it-the-right-way-tp7579191p7579192.html Sent from the r-sig-ecology mailing list archive at Nabble.com.
Dear Frate, I don't know if I do understand well your problem but Binomial suppose that you have a binary response variable. For proportions, you should use other models otherwise you should specifiy proportions of success vs failure. Hope this could help. Cheers Abdoul
On 14-11-01 10:29 AM, Ludovico Frate wrote:
Dear Class, thank you for your reply. Maybe I have not been very clear in the exposition. I'll try to reformulate. "the likelihood of colonizing or becoming extinct" is a proportion so the binomial distribution should be still ok! Considering only colonizing species, or arrivals, for a given plot. This means that a species is not new in overall sampled flora but only for that plot (i.e. *local* colonizer). PLOTID;TRAITS;NUMBER OF NEW COLONIZERS;NUMBER OF SUCCESS;PROPORTION 1; ChF; 4; 0; 0 1; ChC; 4; 0; 0 1; ChR; 4; 0; 0 1; ChSC;4; 0; 0 1; ChS; 4; 1; 0,25 1; Ph; 4; 0; 0 1; Geo; 4; 0; 0 1; HeB; 4; 0; 0 1; HeC; 4; 0; 0 1; HeR; 4; 0; 0 1; HeS; 4; 3; 0,75 1; Th; 4; 0; 0 2; ChF; 17; 0; 0 2; ChC; 17; 0; 0 2; ChR; 17; 1; 0,058823529 2; ChSC;17; 0; 0 2; ChS; 17; 2; 0,117647059 2; Ph; 17; 0; 0 2; Geo; 17; 1; 0,058823529 2; HeB; 17; 0; 0 2; HeC; 17; 0; 0 2; HeR; 17; 2; 0,117647059 2; HeS; 17; 7; 0,411764706 2; Th; 17; 4; 0,235294118 3; ChF; 16; 0; 0 3; ChC; 16; 0; 0 3; ChR; 16; 2; 0,125 3; ChSC;16; 0; 0 3; ChS; 16; 0; 0 3; Ph; 16; 0; 0 3; Geo; 16; 2; 0,125 3; HeB; 16; 0; 0 3; HeC; 16; 0; 0 3; HeR; 16; 1; 0,0625 3; HeS; 16; 9; 0,5625 3; Th; 16; 2; 0,125 . . . . . . . . . . . . 24 . . . for example, in the table, the plot 1 has 4 new species found of which 1 is a ChS and 3 are HeS with a proportion of 0.25 and 0.75 respectively. In a GLM model, the response variable should be PROPORTION (that I have improperly called PROBABILITY) An hypothetical model could be: M0<-glm(PROPORTION~TRAITS, family = binomial, weights = NUMBER OF NEW COLONIZERS, data = ...) Does it make sense? For the second question, I agree with you that "the probability of a species colonizing a plot not only depends on which life form that particular species has". Regards, Ludovico -- View this message in context: http://r-sig-ecology.471788.n2.nabble.com/Re-R-sig-ecology-Digest-80-1-Binomial-GLM-Is-it-the-right-way-tp7579191p7579192.html Sent from the r-sig-ecology mailing list archive at Nabble.com.
_______________________________________________ R-sig-ecology mailing list R-sig-ecology at r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-ecology