The baseline hazard is multiplied by an exponential function that expresses the multiplicative effect of the 1 to p covariates, multiplied by the corresponding

regression parameters \( \beta_i \). If a particular covariate \( x_i \) does not influence the observed hazard rate, then \( \beta_i \) does not differ significantly from 0. The estimates for the regression coefficients are used to compute a hazard ratio (HR), which describes the effect of the covariate (Kalbfleisch and Prentice 2002). Its significance is assessed with a Z score. Covariates used in the analysis NVP-BSK805 datasheet were coded as categorical since the measurements were unevenly spread over the ranges: temperature (°C; T), radiation (°C; R), cloudiness (% cloud cover; C), wind speed (m/s; W), gender (G; male Torin 1 mouse versus baseline female), and year (Y; 2007 versus baseline 2006; representing unmeasured factors changing between years,

e.g. food supply). Weather variables were clustered into ‘low’, ‘intermediate’, and ‘high’ categories to distinguish optimum or unidirectional effects of weather variables on the duration of bouts (Table 2). We based the clustering of covariates on Kaplan–Meier plots. A Kaplan–Meier survival curve is a step function that decreases from 1 (all individuals are still flying MEK162 supplier at time t) toward a minimum value >0 due to termination of flying bouts. Kaplan–Meier survival curves O-methylated flavonoid should be parallel for all covariate categories, i.e. should not cross (Kalbfleisch and Prentice 2002),in order to be able to assume proportionality estimating the effect size in Cox model(s). We plotted Kaplan–Meier survival curves for flying bouts for all covariate values separately, to see under what values curves do not cross (for an example see Appendix Fig. 4). Clustering was subsequently based on best Kaplan–Meier plot appearance. Next, we tested for pairwise differences in behavioural response under

low, intermediate and high weather categories. The effects of single weather variables were estimated simultaneously with other weather variables. We used R 2.7.0 software (Ihaka and Gentleman 1996) to perform the survival analysis. For P. argus, temperature, cloudiness, and wind speed were highly correlated, and differed strongly between years (see Appendix Table 8). Therefore, only radiation was used in the analysis, together with gender and year. Table 2 Clustering of weather variables into ‘low’, ‘intermediate’, and ‘high’ categories per species, resulting from Kaplan–Meier survival curves for flying bouts Weather variable Category C. pamphilus M.