Appendix 1. Model selection in the logistic and logit analyses
To judge the goodness of fit of a model, several approaches were used. The first was an overall test of significance using the log likelihood ratio chi-square test, which compares the full model to a model with just a constant. The second, a single effect test, which is the significance of each explanatory variable in the model judged individually by a log likelihood ratio chi-square test comparing the model with the model with just that variable deleted. This single effect test is recommended instead of the Wald chi-square test, which may show aberrant behaviour (Hosmer & Lemeshow 1989, Trexler and Travis 1993). The explanatory power of the model can be described by the uncertainty coefficient U, which is the ratio of the difference in the loglikelihoods of model and the model with just a constant to the loglikelihood of the model with just a constant. U takes values from 0 for no improvement to 1 for a perfect fit. Model selection was carried out using a hierarchical forward selection procedure. Note that I allowed the last explanatory variable to enter the model at P<0.06 for the single effect test. Sometimes, inclusion of a variable in a model would render a previously selected variable insignificant, but as that variable had already been decided to be the better of the two, in that case the primarily selected variable was kept, and the second was excluded, whether or not it had a significant single effect test. My selection procedure was much more stringent than recommended by Hosmer & Lemeshow (1989) due to the fact that the goal of these analyses was not to find the best model in a predictive or descriptive sense, but to pick out only strong relationships.
Appendix 2. The final logistic regressions models for presence-absence
Only species with a significant final model are listed. The P-value for the whole-model likelihood ratio test is given at WM-P. The intercept is given at I. For a binary variable the value pertains to the state given in brackets after the variable abbreviation. The value for the other state of the binary variable is zero. The 3-state nominal variable , TOP is modeled by two dummy variables, TOP[1–3] and TOP[2–3] with the values 1 and 0 for TOP=1, 0 and 1 for TOP=2 and -1 and -1 for TOP=3, respectively. The significance level of the effect likelihood test for each of the terms in the model are given by the symbol at the parameter for that term: <0.06 (†), < 0.05 (*), <0.01 (**), <0.001 (***), and <0.0001 (****).
Small palms:
Chamaedorea pauciflora: WM-P=0.0078, I=1.609, MC[1]=1,917**
Chamaedorea pinnatifrons: WM-P=0.0504, I=1.462, NN [1]=-0.902†
Geonoma cf. aspidiifolia: WM-P<0.0001, I=30.465, ALT=-0.135*, TOP[1–3]=-2.075**, TOP[2–3]=-0.574**
Geonoma macrostachys var. nov.: WM-P<0.0001, I=0.225, TOP[1–3]=1.913****, TOP[2–3]=0.184****, PDSM[1]=-1.807**
Geonoma stricta var. stricta: WM-P=0.0121, I=-1.344, INC=0.065*
Geonoma triglochin: WM-P=0.0198, I=0.474, TOP[1–3]=-0.029†, TOP[2–3]=0.698†, PDSM[1]=1.341*
Hyospathe elegans: WM-P<0.0001, I=2.417, TOP[1–3]=-1.906****, TOP[2–3]=0.740****
Prestoea schultzeana:WM-P<0.0001, I=-20.602, ALT=0.086*, PDSM[1]=-2.992***, MC[1]=1.300*
Medium-sized palms:
Astrocaryum murumuru var. urostachys: WM-P<0.0001, I=-25.928, ALT=0.113*, TOP[1–3]=-0.655***, TOP[2–3]=1.042***
Attalea indet.: WM-P=0.0031, I=0.283, TOP[1–3]=-0.225*, TOP[2–3]=0.745*, MC[1]=-1.125**
Geonoma maxima: WM-P=0.0005, I=1.909, TOP[1–3]=-1.399***, TOP[2–3]=-0.275***
Phytelephas tenuicaulis: WM-P=<0.0001, I=1.712, NN[1]=-2.330****, TOP[1–3]=0.773*, TOP[2–3]=0.180*
Large palms:
Astrocaryum chambira: WM-P=0.0395, I=0.751, NN[1]=0.983*
Euterpe precatoria var. precatoria: WM-P=0.0207, I=2.529, TOP[1–3]=-1.256*, TOP[2–3]=0.201*
Oenocarpus bataua var. bataua: WM-P=0.0008, I=-0.869, NN[1]=-1.120*, TOP[1–3]=-1.561**, TOP[2–3]=0.550**
Palm lianas:
Desmoncus spp.: WM-P=0.0368, I=3.478, INC=-0.087*
Palmoids:
Aechmea sp.: WM-P=0.0016, I=0.788, MC[1]=1.579**
Asplundia cf. alata: WM-P=0.0001, I=-2.280, PDSM[1]=1.715**, MC[1]=-1.634*
Cybianthus (Weigeltia) sp.: WM-P=0.0064, I=1.935, TOP[1–3]=1.499**, TOP[2–3]=-1.019**
Alsophila cuspidata: WM-P=0.0018, I=38.621, ALT=-0.161*, TOP[1–3]=2.316*, TOP[2–3]=-0.292*, PDSM[1]=-1.139*
Cyathea laesiosora: WM-P=0.0004, I=-26.192, ALT=0.118*, TOP[1–3]=-1.944***, TOP[2–3]=-0.165***
Appendix 3. The final ordinal logit and logistic regression models for abundance
Only species with a significant final model are listed. Explanatory variables are modeled as explained in Appendix 2. Abbreviations as symbols as in Appendix 2, except: I[a] is the intercept for being abundant, while I[p] is the intercept for just being present; the logistic regression models are marked †.
Small palms:
Geonoma cf. aspidiifolia: WM-P<0.0001, I[a]=25.907, I[p]=29.086, ALT=-0.116*, TOP[1–3]=-1.527**, TOP[2–3]=-0.828**, PDSM=1.117*
Geonoma macrostachys var. macrostachys†: WM-<0.0001, I=0.123, NN=-1.208**, INC=0.095**
Prestoea schultzeana: WH-P<0.0001, I[a]=-15.0117, I[p]=-10,240, NN[1]=-0.572***, NN[2]=-2.645***, ALT=0.064†, PDSM[1]=-2.338****, MC[1]=1.038*
Medium-sized palms:
Astrocaryum murumuru var. urostachys: WM-P<0.0001, I[a]=-26.357, I[p]=-23.332, ALT=0.115*,TOP[1–3]=-0.624***, TOP[2–3]=1.107***
Attalea indet.: WM-P=0.0008, I[a]=0.247, I[p)=3.172, NN[1]=0.248*, NN[2]=-1.710*, TOP[1–3]=-0.182*, TOP[2–3]=0.686*, MC[1]=-0.979*
Large palms:
Oenocarpus bataua var. bataua: WM-P<0.0001, I[a]=-2.158, I[p]=0.849, TOP[1–3]=-1.426****, TOP[2–3]=0.631****, PDSM[1]=0.925*