Statistical analyses
To investigate the effect of ecological distance on insect biodiversityper se , we calculated generalized linear models of the formbiodiversity ~ distance * host-tree-genera + RDA1 and a negative binomial distribution. To account for effects of local, environmental conditions we calculated a constrained redundancy analysis (RDA) conditioned on the tree genera for each Heteroptera, Coleoptera and total insects respectively. We used the Hellinger transformed insect diversity matrix as dependent variable and diameter at breast height, total tree height and the amount of congenera within a ten-meter radius as independent variables. This RDA represents the explained variance of the insect diversity matrix with respect to the tree individual fitness (in form of growth parameters) and the proximity of congeners, which both might influence the diversity and abundance of insects. We use the first axis of the RDA (RDA1) to account for these local environmental effect in the analysis. We removed all individuals with the respective biodiversity measure of zero prior to calculations of the generalized model. For the models with Shannon index we used a gaussian distribution and for models of number of species and number of individuals we used a negative binomial distribution. The chosen distributions reflect common distributions in biodiversity analysis for the respective parameters. We report marginal r² values, Pearsons r² for models with gaussian distribution and trigamma calculated r² for models with negative binomial distribution following Nakagawa et al. (2017). To compare if the distance effects are stronger for phytophageous then non-phytophageous insects, we followed a similar procedure as described above, though including insect genera as covariate within the generalized linear model: biodiversity~ distance * host-tree-genera * insect-genera + RDA1 . Model assumptions were assessed visually (QQplots) and tested for overdispersion using package DHARMa(Hartig 2022). All analysis and visualization were done using R and the packages tidyverse (Wickham 2023), MASS (Venables & Ripley 2002), DHARMa (Hartig 2022), effects (Fox et al. 2022), lme4 (Bates et al. 2023), MuMIn (Barton 2020).