I suspect that “Climbing” and “Rainforest” are correlated with each other, so one of both lose significance when both are included in the model.
Also, the Wald-type tests in logistic regression are not exact, and they typically give a different p-value than a likelihood ratio test. Both Wald tests and LRT are approximate, but the Wald-type p-values are typically worse approximations. Cécile On Sep 2, 2019, at 11:28 AM, Søren Faurby <soren.fau...@bios.au.dk<mailto:soren.fau...@bios.au.dk>> wrote: Dear list, I am currently working on a project looking at the predictors of spinyness in palms Spinyness is coded binary. I have a number of potential predictors and want to identify the best set of predictors. I am using "phyloglm" from the "phylolm" package. I am currently using the method=�logistic_MPLE� method which as I describe below gives odd results. method=�logistic_IG10� fails to converge. AIC and the estimated significance of predictors from Walds test seems oddly inconsistent, one predictor produced a much lower AIC but is found to be a non-significant predictor. Another is found to be highly significant but produces a much higher AIC. I am aware that Walds test are only expected to be approximately correct but the degree of difference surprise me. What is the cause of this inconsistent behavior and more importantly, how can I under these circumstances see what is the best predictor(s) in my data? My problem is illustrated by four potential models M1= phyloglm(Spines~1, dat=DATA, phy=TREE, method="logistic_MPLE") M2= phyloglm(Spines~ Climbing, dat=DATA, phy=TREE, method="logistic_MPLE") M3= phyloglm(Spines~ Rainforest, dat=DATA, phy=TREE, method="logistic_MPLE") M4= phyloglm(Spines~ Climbing + Rainforest, dat=DATA, phy=TREE, method="logistic_MPLE") AIC(M1)= 568.17; AIC(M2)= 546.94; AIC(M3)= 528.68; AIC(M4)= 531.98 Based on AIC it is thus clear that the best model is M3, but I get surprised when I compare the results of the models summary(M2) Estimate StdErr z.value p.value (Intercept) 0.90537 1.43428 0.6312 0.5278839 Climbing 0.39713 0.11884 3.3417 0.0008327 *** Note: Wald-type p-values for coefficients, conditional on alpha=0.004616504 summary(M3) Estimate StdErr z.value p.value (Intercept) 1.065304 1.493856 0.7131 0.47577 Rainforest -0.172741 0.090467 -1.9094 0.05621 . Note: Wald-type p-values for coefficients, conditional on alpha=0.004046622 Summary(M4) Estimate StdErr z.value p.value (Intercept) 0.66313 0.87768 0.7556 0.449916 Climbing 0.29763 0.10801 2.7555 0.005861 ** Rainforest 0.27519 0.18470 1.4899 0.136257 Note: Wald-type p-values for coefficients, conditional on alpha= 0.004563691 Hope someone can help me, S�ren - S�ren Faurby http://antonelli-lab.net<http://antonelli-lab.net/> [[alternative HTML version deleted]] _______________________________________________ R-sig-phylo mailing list - R-sig-phylo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-phylo Searchable archive at http://www.mail-archive.com/r-sig-phylo@r-project.org/ [[alternative HTML version deleted]] _______________________________________________ R-sig-phylo mailing list - R-sig-phylo@r-project.org https://stat.ethz.ch/mailman/listinfo/r-sig-phylo Searchable archive at http://www.mail-archive.com/r-sig-phylo@r-project.org/
