The ability to predict microbial community dynamics lags behind the quantity of data available in these systems. Most predictive models use only environmental parameters, although a long history of ecological literature suggests that community complexity should also be an informative parameter. Thus, we hypothesize that incorporating information about a community's complexity might improve predictive power in microbial models. Here, we present a new metric, called community "cohesion," that quantifies the degree of connectivity of a microbial community. We validate our approach using long-term (10+ year) phytoplankton datasets, where absolute abundance counts are available. As a case study of our metrics' utility, we show that community cohesion is a strong predictor of Bray-Curtis dissimilarity (R2 = 0.47) between phytoplankton communities in Lake Mendota, WI, USA. Our cohesion metrics outperform a model built using all available environmental data collected during a long-term sampling program. The result that cohesion corresponds strongly to Bray-Curtis dissimilarity is consistent across the five lakes analyzed here. Our cohesion metrics can be used as a predictor for many community-level properties, such as phylogenetic diversity, nutrient fluxes, or ecosystem services. We explain here the calculation of our cohesion metrics and their potential uses in microbial ecology.