Tip rate estimates can predict future diversification, but are unreliable and context dependent

Understanding the variability of processes leading to the emergence of new lineages is one of the major tasks of macroevolution as a scientific field. Recent years have seen the rise of rate-variable diversification models and metrics that estimate the rates of species diversification at the tips of phylogenetic trees and are thus potentially useful for predicting future evolutionary success of individual species. These methods use various assumptions about the variability and heritability of diversification rates. However, the general performance of rate-variable diversification methods have never been consistently tested against real world data. Here we explore the capacity of multiple rate-variable diversification methods to predict near-future diversification using temporal slices of empirical fossil and extant phylogenies. We do this using a newly developed approach similar to generalized linear models, allowing us to quantify the relationship between predictor tip rates and subsequent diversification rates derived from a probability distribution of numbers of daughter species. We find that tip rates estimated from current methods have non-zero but limited capacity to predict diversification in both fossil and extant phylogenies. The quality of the predictions depends not only on the methods used but also on the specific phylogeny, suggesting that diversification dynamics in some taxa may be more predictable in principle. Our results suggest that future cladogenesis can be, to a certain extent, predicted using existing tip rate methods, but the quality of such predictions is highly variable and depends on factors that are difficult to evaluate in practical applications.