Trait Dependent Diversification

Current understanding

Methods that test whether a binary character state is associated with shifts in speciation or extinction rates — BiSSE being the most widely used — are vulnerable to a specific and severe failure mode: elevated Type I error driven by background diversification heterogeneity in the phylogeny, entirely independent of any real trait effect. The problem is not subtle. On the empirical cetacean phylogeny, more than 77% of 400 neutral character datasets simulated with no state-dependent diversification at all returned a statistically significant BiSSE result (p < 0.05), and 58% cleared the p < 0.001 threshold (10.1093/sysbio/syu131, Finding 1). That phylogeny is not pathological — it has a well-documented dolphin radiation, the kind of diversification heterogeneity present in most large empirical trees.

The taxon-name-length experiment makes the case as cleanly as possible: across vertebrate subtrees, a character defined purely by the number of letters in a species’ binomial showed a significant correlation with speciation rate in more than 69% of trees, climbing toward 100% for ray-finned fishes (10.1093/sysbio/syu131, Finding 2). Name length cannot cause speciation. The false positives cannot be attributed to biological confounding. They reflect the mismatch between what BiSSE’s likelihood assumes — a single pair of diversification rates per character state — and the rate variation actually present in large, empirically estimated phylogenies.

The implication is that a large fraction of published BiSSE-based conclusions may be spurious, and any single significant result from an unmodified BiSSE analysis on a diverse clade warrants skepticism. Whether hidden-state extensions (HiSSE, MuSSE variants) adequately correct the problem, and whether model-adequacy tests can be made routine enough to be trusted across the breadth of empirical applications, remain open questions.

Supporting evidence

Contradictions / open disagreements

The cetacean result may overstate the typical false positive rate. The dolphin radiation creates unusually sharp diversification heterogeneity, and on phylogenies with more homogeneous background rates the error rate could be substantially lower. Additionally, the taxon-name-length character is not a perfectly clean null: congeners share name prefixes, so the character carries residual phylogenetic signal. Some portion of the inflated error rate in ray-finned fishes may trace to that structure rather than to diversification heterogeneity alone. Neither caveat undermines the core finding, but both mean the reported error rates should be treated as upper bounds for the specific clades tested rather than universal constants.

Tealc’s citation-neighborhood suggestions

Question copied. Paste it into the NotebookLM tab.