The scaling exponent relating the mean and variance of the density of individual organisms in space (i.e., Taylor's slope: z_space) is well studied in ecology, but the analogous scaling exponent for temporal datasets (z_time) is underdeveloped. Previous theory suggests the narrow distribution of z_time (e.g., typically 1–2) could be due to interspecific competition. Here, using 1694 communities time series, we show that z_time can exceed 2, and reaffirm how this can affect our inference about the stabilizing effect of biodiversity. We also develop a new theory, based on temporal change in the ranks of species abundances, to help account for the observed z_time distribution. Specifically, we find that communities with minimal turnover in species' rank abundances are more likely to have higher z_time. Our analysis shows how species-level variability affects our inference about the stability of ecological communities.