Goloboff Pablo A, Mattoni Camilo I, Quinteros Andrés Sebastián
Consejo Nacional de Investigaciones Científicas y Técnicas, Instituto Miguel Lillo, Miguel Lillo 205, 4000 S. M. de Tucumán, Argentina.
Diversidad Animal I, F.C.E.F.y N., Universidad Nacional de Córdoba, Avenida Velez Sarsfield 299, 5000 Córdoba, Argentina.
Cladistics. 2006 Dec;22(6):589-601. doi: 10.1111/j.1096-0031.2006.00122.x.
Quantitative and continuous characters have rarely been included in cladistic analyses of morphological data; when included, they have always been discretized, using a variety of ad hoc methods. As continuous characters are typically additive, they can be optimized with well known algorithms, so that with a proper implementation they could be easily analyzed without discretization. The program TNT has recently incorporated algorithms for analysis of continuous characters. One of the problems that has been pointed out with existing methods for discretization is that they can attribute different states to terminals that do not differ significantly-or vice versa. With the implementation in TNT, this problem is diminished (or avoided entirely) by simply assigning to each terminal a range that goes from the mean minus one (or two) SE to the mean plus one (or two) SE; given normal distributions, terminals that do not overlap thus differ significantly (more significantly if using more than 1 SE). Three real data sets (for scorpions, spiders and lizards) comprising both discrete and quantitative characters are analyzed to study the performance of continuous characters. One of the matrices has a reduced number of continuous characters, and thus continuous characters analyzed by themselves produce only poorly resolved trees; the support for many of the groups supported by the discrete characters alone, however, is increased when the continuous characters are added to the analysis. The other two matrices have larger numbers of continuous characters, so that the results of separate analyses for the discrete and the continuous characters can be more meaningfully compared. In both cases, the continuous characters (analyzed alone) result in trees that are relatively similar to the trees produced by the discrete characters alone. These results suggest that continuous characters carry indeed phylogenetic information, and that (if they have been observed) there is no real reason to exclude them from the analysis.
数量性状和连续性状很少被纳入形态学数据的分支分析中；即便被纳入，也总是采用各种临时方法将其离散化。由于连续性状通常具有累加性，因此可以使用众所周知的算法进行优化，这样在恰当实施的情况下，无需离散化就能轻松进行分析。程序TNT最近纳入了用于分析连续性状的算法。现有离散化方法中被指出的一个问题是，它们可能会将不同的状态赋予差异并不显著的终端——或者反之亦然。在TNT中的实现方式是，通过简单地为每个终端分配一个范围（从均值减去一个（或两个）标准误到均值加上一个（或两个）标准误）来减少（或完全避免）这个问题；在正态分布的情况下，不重叠的终端因此具有显著差异（如果使用超过一个标准误则差异更显著）。分析了三个包含离散和数量性状的真实数据集（针对蝎子、蜘蛛和蜥蜴），以研究连续性状的表现。其中一个矩阵中的连续性状数量较少，因此单独分析连续性状只能生成解析度很差的树；然而，当将连续性状添加到分析中时，仅由离散性状支持的许多类群的支持度会增加。另外两个矩阵中的连续性状数量较多，因此可以更有意义地比较离散性状和连续性状的单独分析结果。在这两种情况下，连续性状（单独分析）产生的树与仅由离散性状产生的树相对相似。这些结果表明，连续性状确实携带系统发育信息，并且（如果已经观察到这些性状）没有实际理由将它们排除在分析之外。
