Variable Dependence in Eigenvector Selection for Phylogenetic Eigenvector Regression

Abstract

AbstractIn evolutionary biology, accurate analysis of trait correlations is essential for understanding phylogenetic relationships and evolutionary processes. This study investigates the impact of dependent and independent variable selection on the outcomes of Phylogenetic Vector Regression (PVR) and explores the broader implications for similar statistical models. Using simulated data, we demonstrated how the choice of dependent and independent variables influences phylogenetic eigenvector selection and subsequently affects correlation results within PVR. Our findings reveal that models with higher R2values consistently provide more accurate detection of correlations, suggesting that R2is a reliable criterion for variable selection in PVR analysis. Additionally, while our primary analysis focused on PVR, preliminary investigations into other spatial statistical techniques, such as spatial eigenvector mapping, conditional autoregressive models and simultaneous autoregressive models, have also indicated similar discrepancies when swapping dependent and independent variables. These observations suggest a potentially widespread issue across various statistical models. However, to maintain focus and coherence, a detailed exploration of these findings will be the subject of future work.