Department of Ecology and Evolutionary Biology, University of Connecticut, 75 North Eagleview Road, Storrs, Connecticut 06269, USA.
Ecology. 2011 Jun;92(6):1332-43. doi: 10.1890/10-1345.1.
We develop a novel statistical approach for classifying generalists and specialists in two distinct habitats. Using a multinomial model based on estimated species relative abundance in two habitats, our method minimizes bias due to differences in sampling intensities between two habitat types as well as bias due to insufficient sampling within each habitat. The method permits a robust statistical classification of habitat specialists and generalists, without excluding rare species a priori. Based on a user-defined specialization threshold, the model classifies species into one of four groups: (1) generalist; (2) habitat A specialist; (3) habitat B specialist; and (4) too rare to classify with confidence. We illustrate our multinomial classification method using two contrasting data sets: (1) bird abundance in woodland and heath habitats in southeastern Australia and (2) tree abundance in second-growth (SG) and old-growth (OG) rain forests in the Caribbean lowlands of northeastern Costa Rica. We evaluate the multinomial model in detail for the tree data set. Our results for birds were highly concordant with a previous nonstatistical classification, but our method classified a higher fraction (57.7%) of bird species with statistical confidence. Based on a conservative specialization threshold and adjustment for multiple comparisons, 64.4% of tree species in the full sample were too rare to classify with confidence. Among the species classified, OG specialists constituted the largest class (40.6%), followed by generalist tree species (36.7%) and SG specialists (22.7%). The multinomial model was more sensitive than indicator value analysis or abundance-based phi coefficient indices in detecting habitat specialists and also detects generalists statistically. Classification of specialists and generalists based on rarefied subsamples was highly consistent with classification based on the full sample, even for sampling percentages as low as 20%. Major advantages of the new method are (1) its ability to distinguish habitat generalists (species with no significant habitat affinity) from species that are simply too rare to classify and (2) applicability to a single representative sample or a single pooled set of representative samples from each of two habitat types. The method as currently developed can be applied to no more than two habitats at a time.
我们开发了一种新的统计方法，用于对两个不同生境中的广域种和特化种进行分类。该方法基于两个生境中估计的物种相对丰度的多项模型，最小化了由于两种生境类型的采样强度差异以及每个生境中采样不足导致的偏差。该方法允许对生境特化种和广域种进行稳健的统计分类，而无需先验排除稀有物种。根据用户定义的特化阈值，该模型将物种分为以下四类：（1）广域种；（2）生境 A 特化种；（3）生境 B 特化种；（4）由于置信度不足而无法分类的稀有物种。我们使用两个对比数据集来说明我们的多项分类方法：（1）澳大利亚东南部林地和石南灌丛鸟类丰富度；（2）哥斯达黎加东北部低地加勒比地区次生林（SG）和老林（OG）的树木丰富度。我们详细评估了树木数据集的多项模型。我们对鸟类的结果与之前的非统计分类高度一致，但我们的方法以统计置信度分类了更高比例（57.7%）的鸟类物种。根据保守的特化阈值和多重比较调整，全样本中 64.4%的树种置信度不足，无法分类。在分类的物种中，OG 特化种构成最大的类群（40.6%），其次是广域种（36.7%）和 SG 特化种（22.7%）。多项模型比指示值分析或基于丰度的 phi 系数指数更敏感，能够检测到生境特化种，也能统计检测到广域种。基于稀疏样本的特化种和广域种分类与基于全样本的分类高度一致，即使采样比例低至 20%。新方法的主要优势在于：（1）它能够区分生境广域种（对生境没有明显偏好的物种）和仅仅因为太稀有而无法分类的物种；（2）适用于来自两个生境类型的单个代表性样本或单个综合代表性样本。目前开发的方法一次最多只能应用于两个生境。
