Applications of phylogenetically independent contrasts : a mixed progress report

Phylogeny embodies the historical development of the contemporary diversity of life. Studies of morphological adaptation and life-history evolution often portray this diversity by variation in quantitative, continuously distributed measurements of character traits, such as size and development time. In comparative analyses of such data, statistical interpretations of regression parameters and correlations are confounded by lack of independence among data points resulting from common ancestry of taxa (Felsenstein 1985, Burt 1989, Grafen 1989, Harvey and Pagel 1991). Recently, several techniques have been proposed to circumvent this problem of non-independence. The most widely used approach is to calculate phylogenetically independent contrasts, or PICs (Felsenstein 1985, Harvey and Pagel 1991 [who use the term "independent comparisons"], Martins and Garland 1991, Garland et al. 1992, 1993, Purvis and Garland 1993, Purvis and Rambaut 1995). As acceptance and application of such methods broadens, it becomes increasingly important to compare results obtained from phylogenetically based approaches with traditional analyses based on correlation and regression of trait values for individual taxa. Such comparisons are needed to judge whether conclusions from earlier, non-phylogenetic analyses are likely to be valid, and whether non-phylogenetic comparisons are justifiable today when phylogenetic information is unavailable or in analyses for which independent contrasts applications have not yet been developed. We undertake such a comparison here and comment upon the relative merits of analyzing independent contrasts versus trait values. PICs are calculated as differences in trait values between adjacent pairs of nodes or terminal taxa in a phylogenetic tree. Because no two contrasts share the same branches of a tree, they are statistically independent samples of evolutionary change within a lineage. PIC analysis requires a phylogenetic hypothesis. Thanks to cladistic methods and molecular data, phylogenetic trees are increasingly available. It is no surprise, therefore, that a growing number of comparative studies have used PIC and other phylogenetically based approaches and that these techniques are becoming standard. However, although phylogenetic independence is statistically desirable, phylogenetically based analyses are weakened by increased error variances inherent to the methods themselves. In addition, PIC and other phylogenetic analyses do not easily resolve shifts in the relative rates of evolution of traits within lineages because branches for each contrast span a wide range of time periods. We argue that it is premature at this point to insist upon particular phylogenetically based approaches to comparative analysis, even though pressure is mounting to reject conventional practices of calculating regressions and correlations from trait values of taxa (TIPs of a phylogenetic tree) in favor of PICs. In the minds of many authors, phylogenetically based comparison is the only acceptable approach. Statements such as "Simple correlational studies across an array of species are irredeemably flawed: not only is the number of degrees of freedom undefined, but the results are likely to be biased" (Devoogd et al. 1993) are becoming commonplace in the literature. On one side, Harvey (1996) remarks that "ecologists generally do not use phylogenetically based methods for analysing their comparative data" leading to "conclusions that are potentially misleading". On the other side, Westoby et al. (1995a) complain about the "hegemony of phylogenetic correction" (PC), stating that "...the belief seems to have become very widespread that PC should be automatic and obligatory". Weathers and Siegel (1995) make the same point. Finding no difference between results based on raw data and independent