Abstract:Phylogenetic autocorrelation refers to the similarity in traits among closely related species due to shared evolutionary history, violating the assumption of independence in conventional statistical analyses. Phylogenetic eigenvector regression (PVR) addresses this issue by incorporating phylogenetic eigenvectors derived from phylogenetic trees to control for shared ancestry and isolate trait correlations. In this study, we investigated how different selections of dependent and independent variables affect eigenvector identification and trait correlation results. We used simulated data to analyze both fixed-balanced and random phylogenetic trees, which allowed us to record internal node data. This enabled us to determine the true correlation between traits using traditional Pearson and Spearman rank correlations, providing a gold standard for comparison. We found that the eigenvectors selected differed significantly when switching between dependent variables ( and ), which led to inconsistent PVR correlation results. We assessed three criteria- , Pagel's λ, and AIC-to determine which dependent variable produced more accurate results. Our results indicate that, under scenarios with significant phylogenetic autocorrelation and larger sample sizes (128 and 1024 species), the criterion of maximizing consistently provided the more accurate model. This was true across multiple evolutionary models, tree structures, and regression estimators, including L2 (OLS), MM, and L1. However, in smaller samples (16 species), Pagel's λ was more effective as a selection criterion. These findings emphasize the importance of using appropriate criteria, such as or λ, for selecting dependent variables to enhance the accuracy and consistency of evolutionary inferences. Additionally, our preliminary findings suggest that similar issues of variable selection may affect other spatial statistical tools, such as spatial eigenvector mapping, conditional autoregressive models, simultaneous autoregressive models, and generalized additive models. These insights extend beyond phylogenetic analyses to other contexts where spatial autocorrelation exists or coexists with phylogenetic autocorrelation. Future work is needed to address how to appropriately select dependent and independent variables in these contexts, which remains a key challenge for understanding evolutionary and ecological relationships.

Optimizing Variable Selection in Phylogenetic Eigenvector Regression for Trait Correlation Analysis

Analyses of the variable selection using correlation methods: An approach to the importance of statistical inferences in the modelling process

Variable selection method for the identification of epistatic models.

Pearson's correlation under the scope: Assessment of the efficiency of Pearson's correlation to select predictor variables for linear models

Power Analysis of Principal Components Regression in Genetic Association Studies.

Extending phylogenetic regression models for comparing within‐species patterns across the tree of life

Variable-Selection Emerges on Top in Empirical Comparison of Whole-Genome Complex-Trait Prediction Methods

Dissecting genetic correlation and pleiotropy through a genetic cross

Importance of spatial predictor variable selection in machine learning applications -- Moving from data reproduction to spatial prediction

Phylogenetic ANOVA: Group‐clade aggregation, biological challenges, and a refined permutation procedure

Unifying approaches from statistical genetics and phylogenetics for mapping phenotypes in structured populations

Analysis of comparative data with hierarchical autocorrelation

Spatial Knockoff Bayesian Variable Selection in Genome-Wide Association Studies

Multivariate Bayesian variable selection with application to multi-trait genetic fine mapping

A tale of too many trees: a conundrum for phylogenetic regression

Model Selection with Misspecified Spatial Covariance Structure

A Two-Stage Variable Selection Approach for Correlated High Dimensional Predictors

Assessing phenotypic correlation through the multivariate phylogenetic latent liability model

Evaluation of redundancy analysis to identify signatures of local adaptation

Partial least squares regression as an alternative to current regression methods used in ecology