Data analysis
In this study, we used differences between sunset time and first emergence time of bats, and between sunset time and mid-emergence time, as two variables to quantify the time of emergence of bats. Moreover, we also used differences between sunrise and final return time as a variable to quantify the return time of bats. The three variables were tested for normality by Shapiro-Wilk tests. We found that only differences between sunrise and final return time did not follow a normal distribution (P < 0.001). In this case, the logarithmic transformation was used to make the data meet a normal distribution (P = 0.407).
Here, in order to state and use statistics more easily, we use the following abbreviations for independent variables measured at dusk, which were used to construct models to test the effects of these variables on emergence times of bats: HFE, Humidity at first emergence of bats; HME, Humidity at median emergence of bats; TFE, Temperature at first emergence of bats; TME, Temperature at median emergence of bats; SST, Sunset time; LISS, Light intensity of sunset; LIFE, light intensity of first emergence; LS, Lactation stages (lactation or post-lactation); PAPDD, Presence or absence of predators during dusk; LME, light intensity of mid-emergence; PHNHM, Predators were hunting or not hunting in mid-emergence. Additionally, variables measured at dawn were considered independent variables for the analysis of the effects of factors on final return events: HFR, Humidity at final return of bats; TFR, Temperature at final return of bats; SRT, Sunrise time; LISR, Light intensity at sunrise; LS, Lactation stages (lactation or post-lactation); PAPDA, Presence or absence of predators during dawn; LIFR, Light intensity at final return.
To test the effects of environmental factors, reproduction, and predation on the emergence time of V. sinensis, we selected optimized linear models using the dredge function in the package “MuMin” in R v. 3. 5.1 (R Core Development Team, 2018). In first model, we used the difference between sunset time and first emergence time of bats as a dependent variable; HFE, TFE, SST, LISS, LS, PAPDD and LIFE as independent variables; and year as a random variable. In the second model, we used the difference between sunset time and mid-emergence time as a dependent variable; HME, TME, ST, LISS, LIFE, LIME, LS and PHNHM as independent variables; and year as a random variable. Additionally, to test the effects of environmental factors, reproduction, and predation on the final return time of V. sinensis, we also selected optimized linear models using the dredge function in the package “MuMin” in R v. 3.5.1. In the third model, we used the difference between sunrise time and final return time as a dependent variable; HFR, TFR, SRT, LISR, LS, PAPDA and LIFR as independent variables; and year as a random variable. To test the multicollinearity of variables, the Variance Inflation Factor (VIF) of each predicted factor was calculated to determine which predictors could not be used for subsequent analysis. If VIF was < 5, the corresponding predictor variables were included in the models (Lin. & Feng, 2008). We compared the models using the Akaike information criterion corrected for small sample size (AICc). The model with the lowest AICc indicates the best-fitting model. We caculated ΔAICc as the difference of AICc of each model between the AICc of the best-fitting model. A difference of AICc > 2 (ΔAICc > 2) indicated that the model with the lower AIC value had better explanatory power (Anderson & Burnham, 2002). If the ΔAICc was ≤ 2, multimodel inference was performed using the function model.avg in the package ‘MuMIn’ (Bartoń., 2014) in R v. 3.5.1. We also calculated Akaike weights (Wi ) to assess the relative likehood of a model compared with other models. Additionally, we conducted a hierarchical partitioning analysis in the ‘hier.part package’ (Walsh. & Nally., 2020) in R v. 3.5.1 to estimate the independent effect of each predictor variable (Chevan. & Sutherland., 1991).
Multiple linear regression analyses were used to assess the relationships between sunset time and first emergence time and LIFE, between sunset time and first emergence time and LISS, between between sunset time and mid-emergence time and LIFE, between between sunset time and mid-emergence time and LISS, between between sunrise and final return time and LIFR, and between sunrise and final return time and LISR. During the analysis, if the relationships between dependent variables and independent variables were better simulated by a curve, a curvilinear regression analysis was performed. In this study, we used curvilinear regression analyses to assess the relationships between sunset time and first emergence time and the variables LIFE, between sunset time and mid-emergence time and LIFE, and between sunrise and final return time and LIFR. All statistical analyses were conducted in R v. 3.5.1.