Statistical analyses
Data analysis was done in R (version 3.6.1). We checked distributional assumptions using the fitdistrplus package (Delignette-Muller & Dutang 2015) and inspected model residuals for constant variance.
We used generalized linear mixed-effects models (package glmmTMB, Brooks et al., 2017) with tweedie errors and management nested in blocks as random effects to analyze data on flower visitor diversity and total arthropod diversity, which is expressed as Shannon’s entropy (Jost 2007) and its numbers equivalent (exponential of Shannon’s diversity) in the whole manuscript. This index was calculated using the package vegan (Oksanen et al. 2019). As fixed effects we used 1) crop diversity*management and 2) crop mixture*management in separate models.
We generated plant-flower visitor networks using the bipartite package (Dormann et al. 2009) for all crop mixtures and management intensities separately (N=4). For calculations of network metrics, we pooled our data per block, crop mixture and management intensity, and then analyzed effects of crop mixture and management on network metrics using the same model structure as above with generalized poisson and compois errors. We used the number of interactions, number of flower visitor species, and Shannon’s diversity of interactions as metrics to describe the networks, as more complex indices need a minimal network size to function reliably (Dormann et al. 2009), which was not the case in this study.
The number of flower visits and the total number of arthropod individuals were analyzed as described above with generalized poisson errors. Crop biomass was analyzed using tweedie errors. Means were compared using Type Ⅱ Wald chisquare tests from the car library (Fox & Weisberg 2019).