Are the population dynamics of species mainly determined by direct interactions with predators, preys and conspecifics? Or, instead, are those dynamics more dependent on indirect feedbacks that ripple across the whole interaction network? Here we show that, from a basic spectral feature of the interaction network, we can predict the length of indirect interaction pathways that contribute to community-level dynamical patterns, such as the depth of a perturbation's reach, or the contribution of biotic processes to realized species niches. In doing so, we propose a measure of collectivity that integrates existing approaches to community complexity, collective integration and indirect interactions. By revisiting classic concepts of theoretical community ecology, our work proposes an original perspective on the question of to what degree communities are more than loose collections of species or simple interaction motifs. This perspective can help clarify when reductionist approaches, focusing on particular species and small interaction motifs, ought to suffice or fail when applied to ecosystems.