Occam's bias undermines inferences from phylogenetic linear models

Phylogenetic modelling has consolidated as the analytical standard to address hypotheses about the patterns and dynamics of biodiversity in inter-specific contexts. These analyses are traditionally performed implementing phylogenetic linear models where single outcomes are regressed against multiple predictors without explicitly modelling the relationships amongst predictors. A prevailing, yet largely overlooked consequence of neglecting these relationships is what we introduce as Occams bias - a statistical distortion arising where the model has fewer cause-effect connections than predicted by theory. Here, we propose that Occams bias is likely to have impacted a wide range of inferences about ecological and evolutionary processes made from phylogenetic linear models across the literature, and thus, that the adoption of approaches to address this bias are critical. We present an empirical test of the long-standing hypothesis that interspecific variation in life-history traits influences the likelihood of extinction risk across 13,949 species of terrestrial vertebrates to show the impacts of Occams bias in phenomenological inference. Our study calls for a re-evaluation of hypotheses tested using the traditional linear modelling structure and advocate the use and continued development of multi-response model structures that account for all causal pathways in phylogenetic analyses.