The loss of information accompanying assessment of absolute fit of substitution models to phylogenetic data negatively affects the discriminatory power of previous methods and can make them insensitive to lineage-specific changes in the substitution process. As an alternative, I propose evaluating absolute fit of substitution models based on a novel statistic which describes the observed data without information loss and which is unlikely to become zero-inflated with increasing numbers of taxa. This method can accommodate gaps and is sensitive to lineage-specific shifts in the substitution process. In simulation experiments, it exhibits greater discriminatory power than previous methods. The method can be implemented in both Bayesian and Maximum Likelihood phylogenetic analyses, and used to screen any set of models. Recently, it has been suggested that model selection may be an unnecessary step in phylogenetic inference. However, results presented here emphasize the importance of model fit assessment for reliable phylogenetic inference.

中文翻译：

---

适应时间可逆和非时间可逆进化过程的绝对替代模型拟合评估

替换模型与系统发育数据的绝对拟合评​​估伴随的信息丢失会对先前方法的辨别力产生负面影响，并可能使它们对替换过程中谱系特异性的变化不敏感。作为替代方案，我建议基于一种新颖的统计数据来评估替代模型的绝对拟合度，该统计数据描述了观察到的数据而没有信息丢失，并且随着类群数量的增加而不太可能变得零膨胀。该方法可以适应间隙，并且对替换过程中谱系特异性的变化敏感。在模拟实验中，它比以前的方法表现出更大的判别力。该方法可以在贝叶斯和最大似然系统发育分析中实现，并用于筛选任何模型集。最近，有人认为模型选择可能是系统发育推断中不必要的步骤。然而，这里提出的结果强调了模型拟合评估对于可靠的系统发育推断的重要性。