Ramsey Fred L, Usner Dale
Oregon State University, Corvallis, Oregon, 97311, USA.
Biometrics. 2003 Jun;59(2):332-40. doi: 10.1111/1541-0420.00040.
Biologists attach radio transmitters to animals so that the animals' movements through their preferred habitats can be followed. To analyze the resulting sequences of visited habitat classes, McCracken, Manly, and Vander Heyden (1998, Journal of Agricultural, Biological, and Environmental Statistics 3(3), 268-279) proposed an independent multinomial selections (IMS) model. Two issues that arise when using this approach are: (i) serial dependence possibly affects measures of uncertainty; and (ii) individual animals from the population studied may exhibit heterogeneity in their selection patterns. We develop two single-parameter extensions of the IMS model to address these issues. A Markov chain model allows for persistence in the habitat class previously visited. Heterogeneity is modeled by assuming the population of animal selection patterns follows a Dirichlet distribution, from which the study animals are a random sample. We show that these persistence and heterogeneity characteristics are present in the study by McCracken et al. (1998) of bear movements. Simulations demonstrate that failure to account for persistence or heterogeneity when either is present can seriously misrepresent measures of uncertainty.
生物学家给动物装上无线电发射器，以便追踪动物在其偏好栖息地中的活动。为了分析由此产生的所访问栖息地类别的序列，麦克拉肯、曼利和范德海登（1998年，《农业、生物和环境统计学杂志》3(3)，268 - 279）提出了一种独立多项选择（IMS）模型。使用这种方法时出现的两个问题是：（i）序列依赖性可能会影响不确定性的度量；（ii）所研究种群中的个体动物在其选择模式上可能表现出异质性。我们开发了IMS模型的两个单参数扩展来解决这些问题。一个马尔可夫链模型考虑了对先前访问过的栖息地类别的持续性。通过假设动物选择模式的种群遵循狄利克雷分布来对异质性进行建模，所研究的动物是从该分布中抽取的随机样本。我们表明，这些持续性和异质性特征存在于麦克拉肯等人（1998年）对熊活动的研究中。模拟表明，当存在持续性或异质性时未能加以考虑，可能会严重误代表不确定性的度量。
