Matzke Nicholas J, Irmis Randall B
Division of Ecology and Evolution, Research School of Biology, The Australian National University, Canberra, Australian Capital Territory, Australia.
School of Biological Sciences, University of Auckland, Auckland, New Zealand.
PeerJ. 2018 Apr 6;6:e4553. doi: 10.7717/peerj.4553. eCollection 2018.
Tip-dating, where fossils are included as dated terminal taxa in Bayesian dating inference, is an increasingly popular method. Data for these studies often come from morphological character matrices originally developed for non-dated, and usually parsimony, analyses. In parsimony, only shared derived characters (synapomorphies) provide grouping information, so many character matrices have an ascertainment bias: they omit autapomorphies (unique derived character states), which are considered uninformative. There has been no study of the effect of this ascertainment bias in tip-dating, but autapomorphies can be informative in model-based inference. We expected that excluding autapomorphies would shorten the morphological branchlengths of terminal branches, and thus bias downwards the time branchlengths inferred in tip-dating. We tested for this effect using a matrix for Carboniferous-Permian eureptiles where all autapomorphies had been deliberately coded. Surprisingly, date estimates are virtually unchanged when autapomorphies are excluded, although we find large changes in morphological rate estimates and small effects on topological and dating confidence. We hypothesized that the puzzling lack of effect on dating was caused by the non-clock nature of the eureptile data. We confirm this explanation by simulating strict clock and non-clock datasets, showing that autapomorphy exclusion biases dating only for the clocklike case. A theoretical solution to ascertainment bias is computing the ascertainment bias correction (M), but we explore this correction in detail, and show that it is computationally impractical for typical datasets with many character states and taxa. Therefore we recommend that palaeontologists collect autapomorphies whenever possible when assembling character matrices.
末端定年法是一种越来越流行的方法，即在贝叶斯定年推断中将化石作为有年代的末端分类单元纳入其中。这些研究的数据通常来自最初为非定年分析（通常是简约分析）而构建的形态特征矩阵。在简约分析中，只有共享的衍生特征（共源性状）提供分组信息，因此许多特征矩阵存在确定偏差：它们忽略了自源性状（独特的衍生特征状态），认为这些特征没有信息价值。目前尚无关于这种确定偏差在末端定年法中的影响的研究，但自源性状在基于模型的推断中可能具有信息价值。我们预计排除自源性状会缩短末端分支的形态分支长度，从而使在末端定年法中推断的时间分支长度向下偏差。我们使用一个石炭纪 - 二叠纪真爬行动物的矩阵来测试这种影响，在该矩阵中所有自源性状都经过了刻意编码。令人惊讶的是，排除自源性状时年代估计实际上没有变化，尽管我们发现形态速率估计有很大变化，并且对拓扑结构和定年置信度有较小影响。我们推测对定年缺乏影响这一令人困惑的现象是由真爬行动物数据的非时钟性质导致的。我们通过模拟严格时钟和非时钟数据集证实了这一解释，表明排除自源性状仅在类似时钟的情况下会使定年产生偏差。确定偏差的一个理论解决方案是计算确定偏差校正（M），但我们详细探讨了这种校正，结果表明对于具有许多特征状态和分类单元的典型数据集，它在计算上是不切实际的。因此，我们建议古生物学家在组装特征矩阵时尽可能收集自源性状。
