Human disturbance but not predation risk is associated with increased vigilance in roe deer

This is the dataset used to analyse the vigilance behaviour of the roe deer. The behavioural data were collected using camera traps and direct observations across seven study areas in north-western Italy. We analysed vigilance behaviour, i.e. the proportion of time spent vigilant, as the dependent variable in relation to several categorical predictors: sampling method (camera trapping vs. direct observation), season (spring, summer, autumn, winter), time period (day, night, twilight), landscape type (natural, mixed, modified), wolf presence (stable vs. occasional), and management status (protected vs. hunting area). Continuous predictors included group size and the distance (in metres) from each observation to the nearest woodland, natural open area (e.g., grasslands, pastures, shrublands), cultivated land, and urban area. We modelled our response variable using beta regression with a logit link function. Starting from a global model including all the predictors previously described, we generated a set of models using all possible combinations of uncorrelated variables; model selection was performed using an AICc-based dredge procedure and models with Δ AICc less than 2 were retained for model averaging. All analyses were conducted in R version 4.5 (R Core Team, 2025) using the 'stats', 'MuMIn' (Bartoń, 2023), and 'betareg' (Cribari-Neto and Zeileis, 2010) packages.