Abstract Understanding the drivers of change in communities is a major goal of paleoecology and community ecology, but statistical inference from multivariate time series is challenged by relative (rather than absolute) abundance data, observation uncertainty and missing data due to uneven sampling through time. We present a new state‐space approach for modelling multinomially distributed community time series, such as the proportional representation of pollen in sediment cores or species counts in long‐term studies. The approach includes separate models for process and observation error, and simultaneously estimates coefficients of environment‐species relationships and species‐species interactions, thereby decomposing sources of change in community composition. Our approach is implemented by the R package multinomialTS, which is specifically designed to test hypotheses about the drivers of ecological change through time. We demonstrate the capabilities of the multinomial state‐space model by applying it to a recently collected fossil pollen record from Sunfish Pond, PA, USA and then under simulated conditions to assess model performance. At Sunfish Pond, the analyses support the hypotheses that water availability was an important driver of past forest dynamics, and that species‐species interactions among Pinus strobus , Quercus spp. and Fagus grandifolia drive significant changes in relative abundances. In the simulation scenarios, the multinomial state‐space model successfully recovered the true coefficient values for species‐environment interactions and interactions within and among taxa. Our multivariate model complements currently available state‐space approaches by being explicitly designed for time series of proportional species abundances often used in community ecology and paleoecology. The R package multinomialTS thus provides tools for testing hypotheses about the extrinsic and intrinsic drivers of ecological change through time.
Asena et al. (2026) studied this question.