Content area
Phylogenetic comparative methods are essential for analyzing cross-species data while accounting for evolutionary relationships. Traditional methods, such as phylogenetically independent contrasts (PIC) and phylogenetic generalized least squares (PGLS), often rely on parametric assumptions that may not hold under abrupt evolutionary shifts or deviations from Brownian motion (BM) models. Ordinary least squares (OLS) regression, when applied to PIC, forms the basis of PIC-OLS, a commonly used approach for analyzing trait correlations in evolutionary studies. Mathematically, PIC-OLS is equivalent to Pearson correlation analysis of PIC values, providing a computationally simpler yet directionally and statistically identical alternative to the regression-based method. We introduce a hybrid framework for phylogenetic correlation analysis tailored to dataset size, designed specifically for analyzing PIC values: outlier-guided correlation (OGC) for large datasets and outlier- and distribution-guided correlation (ODGC) for small datasets, collectively referred to as O(D)GC. OGC dynamically integrates Pearson and Spearman correlation analyses based on the presence of outliers in PIC values, while ODGC further incorporates normality testing to address the increased sensitivity of parametric methods to non-normality in small samples. This adaptive and dynamically adjusted approach ensures robustness against data heterogeneity. Using simulations across diverse evolutionary scenarios, we compared PIC-O(D)GC with a comprehensive range of methods, including eight robust regression approaches (PIC-MM, PIC-L1, PIC-S, PIC-M, and their PGLS counterparts); PGLS optimized using five evolutionary models: BM, lambda, Ornstein-Uhlenbeck random (OU-random), OU-fixed, and Early-burst; Corphylo (an OU-based method); PIC-Pearson; and two advanced models, phylogenetic generalized linear mixed models (PGLMM) and multi-response phylogenetic mixed models (MR-PMM). Our results demonstrate that under conditions with evolutionary shifts, PIC-O(D)GC and PIC-MM consistently outperform other methods by minimizing false positives and maintaining high accuracy. In no-shift scenarios, PIC-O(D)GC and PIC-MM often rank among the best-performing methods, though distinctions between methods become less pronounced. PIC-O(D)GC not only offers a more accurate tool for analyzing phylogenetic data but also introduces a novel direction for dynamically adjusting statistical methods based on dataset characteristics. By bridging the gap between computational simplicity and methodological robustness, PIC-O(D)GC emerges as a scalable and reliable solution for trait correlation analyses, effectively addressing the complexities inherent in both stable and dynamic evolutionary contexts.
Competing Interest Statement
The authors have declared no competing interest.
Footnotes
* Extensive revisions have been made to the title, introduction, materials and methods, results, and discussion.