Statistical analysis

Ecology and biology extensively vary among species (Table 1). To control the bias of unrelated sample size for each species at each capture site rather than specific prevalence, we used individual serological response. A multivariate logistic regression analysis was used with R [60] to infer the role of bioecological factors including diet, type of roots, type of colonies (monospecific vs. multispecific), size of colonies, and ecological plasticity [44,61], as well as the characteristics of capture sites, including vegetation type [41], landscape units [42] and disturbance index [40]. For vampire bats that were recaptured, only the serological result obtained at their first capture was considered. To limit the uncovered bias of likely temporal and geographic variations of serology (e.g., [22], this study), we considered the year of sampling and the capture sites as random effects. A two-step procedure was used in the logistic regression model to determine the most parsimonious multivariate model. First, the association of all factors and the positivity were assessed one by one in bivariate analyses with the year as random effect; a p-value < 0.1 allowed selecting candidate variables for the second step. In this second step, the same logistic regression model including all factors identified in the first step was applied, with a generalized linear mixed model fit by maximum likelihood. The best multivariate model was selected with AIC and BIC criteria.