Haukisalmi Voitto, Henttonen Heikki
Department of Ecology and Systematics, Division of Population Biology, PO Box 17, FIN-00014 University of Helsinki, Finland e-mail:
Vantaa Research Center, Finnish Forest Research Institute, PO Box 18, FIN-01301 Vantaa, Finland, , , , , , FI.
Oecologia. 1998 Oct;116(4):565-574. doi: 10.1007/s004420050622.
The purpose of the present study was (1) to test the ability of six alternative methods to detect random and non-random patterns of overall association in artificial presence/absence data sets, and (2) to analyse overall associations and effects of sampling heterogeneity in four empirical presence/absence data sets of helminths of the common shrew Sorex araneus. In the null model, the expected distribution was created by means of a randomisation procedure. Application of methods on artificial data sets indicated a generally low probability of type I statistical error. All methods were more likely to detect positive non-randomness than negative non-randomness of comparable strength, which may partly explain the predominance of positive overall associations in empirical data sets. The analyses based on artificial data sets indicated slight differences between methods in their ability to detect non-randomness of known strength (type II error). However, some of the methods failed to detect strong overall association when the artificial assemblages consisted of roughly equal numbers of positive and negative pairwise interactions. The structure of the artificial data sets always disappeared when the expected distribution was constrained to account for "sampling heterogeneity", i.e. varying prevalence of species among subsamples. The patterns of overall association in real helminth communities were variable, depending on the locality and association method used, but not usually on the simulation constraint used. Of the four empirical data sets analysed, one showed an unequivocal positive structure, in one the structure depended on the method used, and two data sets from the same locality were unequivocally unstructured (random). We discuss the applicability of various association measures, and the possible causes of positive overall associations in parasites.
（1）测试六种替代方法在人工存在/不存在数据集中检测总体关联的随机和非随机模式的能力；（2）分析普通鼩鼱（Sorex araneus）蠕虫的四个经验性存在/不存在数据集中的总体关联以及抽样异质性的影响。在零模型中，通过随机化程序创建预期分布。在人工数据集上应用这些方法表明，I型统计错误的概率通常较低。所有方法检测到正的非随机性的可能性都比检测到强度相当的负的非随机性的可能性更大，这可能部分解释了经验数据集中正的总体关联占主导地位的原因。基于人工数据集的分析表明，各方法在检测已知强度的非随机性（II型错误）的能力上存在细微差异。然而，当人工组合由大致相等数量的正、负成对相互作用组成时，一些方法未能检测到强的总体关联。当预期分布被约束以考虑“抽样异质性”，即子样本中物种的不同患病率时，人工数据集的结构总是会消失。实际蠕虫群落中的总体关联模式各不相同，这取决于所使用的地点和关联方法，但通常不取决于所使用的模拟约束。在分析的四个经验数据集中，一个显示出明确的正结构，一个结构取决于所使用的方法，来自同一地点的两个数据集则明确无结构（随机）。我们讨论了各种关联度量的适用性以及寄生虫中总体正关联的可能原因。
