Data analysis
All the analyses were conducted with R, version R 4.0.2 (R Core Team,
2020).
Daily temporal patterns of flight activities - To
understand how flight performance of both honey bee and hornet vary over
the time, we used Generalized Additive Models (GAMs, using themgcv R-package) of the following flight parameters: the number of
trajectories per unit of time, the maximum flight speed, the flight
trajectory curvature and the percentage of time spent hovering for each
flight trajectory of bees and hornets.
Flight performance – To compare flight performances
between honey bees entering or leaving and hornets, we analysed their
distribution of maximum speed, curvature and hovering percentage. We
checked for normality using an Anderson-Darling normality test (adapted
for large datasets >5000pts, using the nortestR-Package) and for variance homogeneity with a Levene test (using thecar R-Package). We then used Kruskal-Wallis rank sum tests to
assess differences in flight parameters between the three types of
flight (hornets, honey bees entering, honey bees leaving the hive,
followed by a pairwise Wilcoxon test as a Post-Hoc test with a
Bonferroni P value adjustment method (using the stats R-Package).
Overlapping flight performances and predation success
behaviour - To assess which of the global parameters best explained
the hornet predation success (response variable), we ran a binomial
Generalized Linear Mixed Model (GLMM, using the GLMM R-Package)
with fixed parameters being the number of hornets present, the hour of
the day, the number of honey bees present, the interaction between all
those parameters, and the quadratic parameters of hornets and hour. To
select parameters of interest, we ran a multimodel inference procedure
by AIC comparison (using the MuMIn R-Package). In order to
extract specific behavioural patterns linked with the hornet predation
success, we analysed the distribution of maximum flight speed, flight
curvature and hovering percentage for the prey (pooling honey bees
entering and leaving their hive) and predator, in case of predation
success or failure. To test for statistical differences between
predation success and failure, we checked for variance homogeneity as
described above. We then used a Kruskal-Wallis rank sum tests to assess
differences in flight parameters between them, followed by a pairwise
Wilcoxon test as a Post-Hoc test with a Bonferroni P value
adjustment method (using the stats R-Package).
Impact of hornet density on bees and hornets flight
performance and predation success – To assess the impact of hornet
density (Log10 number of hornets) on bee and hornet
flight performance traits (i.e. flight speed, curvature and hovering),
we used Linear Models (LM, using the stats R-Package) on hornets,
honey bees leaving the hive or honey bees entering the hive. Using the
same statistical technique, we analysed whether the density of hornets
impacted the coefficient of variation in these three flight performance
traits in hornets and in honey bees leaving the hive and entering the
hive.
Results