Testing hypotheses of correlated evolution using phylogenetically independent contrasts: Sensitivity to deviations from Brownian motion
Testing hypotheses of correlated evolution using phylogenetically independent contrasts: Sensitivity to deviations from Brownian motion
复制标题
DOI：
10.2307/2413510
复制
发表时间：
1996-03-01
影响因子：
6.5
通讯作者：
Garland, T
中科院分区：
文献类型：
--
作者：
DiazUriarte, R;Garland, T
我们检查了 Felsenstein (1985, Am. Nat. 125:1-15) 系统发育独立对比的比较方法的统计性能（就 I 类错误率而言），以检验有关连续值特征的进化相关性的假设。我们模拟了两种不同系统发育的数据，一种针对 15 种无胸蝾螈，另一种针对 49 种食肉目和有蹄类动物。我们实现了 15 种不同的角色进化模型，其中 14 种偏离了布朗运动，这实际上是该方法假设的。研究的模型包括奥恩斯坦-乌伦贝克过程和间断平衡（每个分叉处只允许一个子代发生变化），有或没有趋势以及表型进化程度的限制。正如之前的几项模拟研究所示，物种平均值的非系统发育皮尔逊相关性在大多数模型（包括简单的布朗运动模型）下都会产生夸大的 I 型错误率。独立对比在布朗运动下（以及在与该模型略有偏差的初步研究中）产生了可接受的 I 型错误率，但在大多数其他模型下它们被夸大了。这一新结果证实了独立对比的模型依赖性。然而，当检查和转换分支长度时，独立对比的 I 类错误率就会降低。此外，观察到的最大 I 型错误率从未超过 (α = 0.05) 名义 P 值的两倍。相比之下，非系统发育相关性往往会产生极度膨胀（且高度可变）的 I 型错误率。这些结果再次证明了基于系统发育的统计方法相对于非系统发育方法的总体优越性，即使在与布朗运动模型存在极大偏差的情况下也是如此。这些结果还表明有必要检查统计比较方法的假设，并表明诊断检查和补救措施可以显着提高了独立对比方法的性能。
We examined the statistical performance (in terms of type I error rates) of Felsenstein's (1985, Am. Nat. 125:1-15) comparative method of phylogenetically independent contrasts for testing hypotheses about evolutionary correlations of continuous-valued characters. We simulated data along two different phylogenies, one for 15 species of plethodontid salamanders and the other for 49 species of Carnivora and ungulates. We implemented 15 different models of character evolution, 14 of which deviated from Brownian motion, which is in effect assumed by the method. The models studied included the Ornstein-Uhlenbeck process and punctuated equilibrium (change allowed in only one daughter at each bifurcation) both with and without trends and limits on how far phenotypes could evolve. As has been shown in several previous simulation studies, a nonphylogenetic Pearson correlation of species' mean values yielded inflated type I error rates under most models, including that of simple Brownian motion. Independent contrasts yielded acceptable type I error rates under Brownian motion (and in preliminary studies under slight deviations from this model), but they were inflated under most other models. This new result confirms the model dependence of independent contrasts. However, when branch lengths were checked and transformed, then type I error rates of independent contrasts were reduced. Moreover, the maximum observed type I error rates never exceeded twice the nominal P value at (alpha = 0.05. In comparison, the nonphylogenetic correlation tended to yield extremely inflated (and highly variable) type I error rates. These results constitute another demonstration of the general superiority of phylogenetically based statistical methods over nonphylogenetic ones, even under extreme deviations from a Brownian motion model. These results also show the necessity of checking the assumptions of statistical comparative methods and indicate that diagnostic checks and remedial measures can substantially improve the performance of the independent contrasts method.