Sickness behaviour reduces network centrality in wild vampire bats

Simon P Ripperger; Sebastian Stockmaier; Gerald G Carter

linear mixed-effects model with treatment (LPS, saline) and day (1, 2, 3) as fixed effects, bat as 113 a random effect, and the network centrality measure (degree, strength, and eigenvalue, 114 . CC-BY-NC-ND 4.0 International license author/funder. It is made available under a The copyright holder for this preprint (which was not peer-reviewed) is the . https://doi.org/10.1101/2020.03.30.015545 doi: bioRxiv preprint respectively) as a response. We then extracted the standardized model coefficients for the 115 treatment effect and the interaction between treatment effect and day. If we detected an 116 interaction, we also fit a linear model for the observations within the first and last day separately 117 and extracted those standardized treatment effect coefficients. To get two-sided p-values, we 118 created 10,000 null datasets where the treatment was re-assigned randomly among bats at the 119 start of the study, then measured the proportion of the null coefficients that were greater than 120 the observed coefficients, and then doubled those one-sided p-values. This procedure creates a 121 null model accounting for the non-independent and non-normal structure of the network data 122 [11] . To assess assortativity of sick and control bats over time, we calculated for each hour the 123 association probability (proportion of possible pairs that were associated) and the mean 124 association time (total seconds per period) for three dyad types: control-control, control-sick, 125