Phylogenetic Analysis and Comparative Data: A Test and Review of Evidence

TLDR

Researchers question whether controlling for phylogenetic associations is statistically necessary in comparative studies. The study investigates whether Pagel’s lambda should be used to detect and correct for phylogenetic dependence in comparative data, and reviews 26 published datasets to assess the prevalence of such associations. The authors employ Pagel’s lambda, a metric ranging from 0 to 1, to quantify phylogenetic correlation and analyze 103 traits from 26 comparative datasets selected from ecological literature. Simulations confirm lambda is a powerful, low‑error, and robust index for detecting phylogenetic dependence, and analysis of the 26 datasets shows 88 % contain at least one trait with significant dependence, 60 % of all traits exhibit dependence, while 16 % are equivocal—leading the authors to recommend routine lambda estimation for optimal correction.

Abstract

The question is often raised whether it is statistically necessary to control for phylogenetic associations in comparative studies. To investigate this question, we explore the use of a measure of phylogenetic correlation, lambda, introduced by Pagel (1999), that normally varies between 0 (phylogenetic independence) and 1 (species' traits covary in direct proportion to their shared evolutionary history). Simulations show lambda to be a statistically powerful index for measuring whether data exhibit phylogenetic dependence or not and whether it has low rates of Type I error. Moreover, lambda is robust to incomplete phylogenetic information, which demonstrates that even partial information on phylogeny will improve the accuracy of phylogenetic analyses. To assess whether traits generally show phylogenetic associations, we present a quantitative review of 26 published phylogenetic comparative data sets. The data sets include 103 traits and were chosen from the ecological literature in which debate about the need for phylogenetic correction has been most acute. Eighty-eight percent of data sets contained at least one character that displayed significant phylogenetic dependence, and 60% of characters overall (pooled across studies) showed significant evidence of phylogenetic association. In 16% of tests, phylogenetic correlation could be neither supported nor rejected. However, most of these equivocal results were found in small phylogenies and probably reflect a lack of power. We suggest that the parameter lambda be routinely estimated when analyzing comparative data, since it can also be used simultaneously to adjust the phylogenetic correction in a manner that is optimal for the data set, and we present an example of how this may be done.