Statistical analysis
To compare the palatability of the species, we used a logistic regression model to know whether the tadpoles were depredated for the naiads and we used each species of tadpole as a dependent binomial variable. We also used the glm function with the ”logit” link function in ggplot2 package (Friendly and Meyer 2015).
Differences in the conspicuousness of the three species were analyzed with an analysis of variance (ANOVA). To estimate the additional effect of body size, we used an analysis of covariance (ANCOVA). In this case, we used only the conspicuousness values of the dorsal area and the interocular patch of the larvae, which represented the highest contribution percentage in the variance.
To estimate a toxicity score, we conducted principal component analysis (PCA), using 68 variables Ethovision software quantified. We chose variables that were better represented in the PCA and with these, we conducted a discriminant function analyses (DFAs) using as a group discriminating the injection type the mice received: mice injected with skin extracts versus mice injected with saline solution. The values of the first DFA were used as the toxicity score of each tadpole.
Finally, to evaluate the relationship between conspicuity and toxicity score, we used ANCOVA with three species of tadpoles as the covariate. The conspicuousness values were obtained through a PCA. The first PCA was represented by all dorsal colorations of the larvae and the second PCA by the coloration of the interocular patch, which was the most conspicuous and provided the largest contribution in the explained variance.