Biological diversity is fundamental for the functioning of ecosystems on our planet and thus for the survival of humanity. Today, we are experiencing a global biodiversity crisis that severely disrupts ecosystems worldwide, but surprisingly does not lead to species loss on the local scale. Rather, local species assemblages experience shifts in their composition, such as changes in dominance patterns. The effects of compositional shifts compared to the complete loss of species are less well studied, but first experiments show that negative effects are comparable. Many experiments about the influence of biodiversity on ecosystem functioning are done in highly artificial settings, which makes extrapolating results to real world ecosystems difficult. In my doctoral thesis, I describe the results of biodiversity experiments with increased realism in their set-up. I achieved this by using natural species assemblages or studying species directly in the field, considering multiple spatial and temporal scales, including several measures of biodiversity, and exposing the communities to natural stressors. My results show that increasing realism in experimental ecology changes the response of communities to simulated climate stressors. In addition, I found that initial species composition is crucial for determining community responses to stress and that connectivity between local patches can lead to the maintenance of ecosystem functioning in some communities, but not always, especially in response to regional stressors. I conclude that the move to more realism in experimental ecology is vital for understanding changes in natural community dynamics which enables us to give better recommendations for practical biodiversity conversation.