Robust model comparison and tests of interactions in phylogenetic path analysis: Revisiting "Why are telomeres the length they are?" (Benson et al. 2025).

Abstract

We discuss robust model comparison in phylogenetic path analysis and how to incorporate interactions in path analysis with d-separation. We revisit a study by Benson et al. (2025) to highlight the importance of carefully evaluating the causal models being compared to avoid erroneous conclusions. Our results show that there are biologically plausible models, not previously considered, that outperform those in the original study leading to quite different biological conclusions. We also use this work to demonstrate, using empirical data, how interactions can be incorporated in path analyses, and highlight how their presence affects the conditional independencies that must be tested for the model to be considered a plausible causal model. Our commentary highlights the importance of rigorously evaluating model space and verifying both the implied conditional independencies as well as the robustness of the assumed causal effects.