Mwalili Samuel M, Lesaffre Emmanuel, Declerck Dominique
Statistics and Actuarial Sciences, Jorno Kenyatta University of Agriculture and Technology, Kenya.
Stat Methods Med Res. 2008 Apr;17(2):123-39. doi: 10.1177/0962280206071840. Epub 2007 Aug 14.
Zero-inflated models for count data are becoming quite popular nowadays and are found in many application areas, such as medicine, economics, biology, sociology and so on. However, in practice these counts are often prone to measurement error which in this case boils down to misclassification. Methods to deal with misclassification of counts have been suggested recently, but only for the binomial model and the Poisson model. Here we look at a more complex model, that is, the zero-inflated negative binomial, and illustrate how correction for misclassification can be achieved. Our approach is illustrated on the dmft-index which is a popular measure for caries experience in caries research. An extra problem was the fact that several dental examiners were involved in scoring caries experience. Using our example, we illustrate how a non-differential misclassification process for each examiner can lead to differential misclassification overall.
用于计数数据的零膨胀模型如今越来越受欢迎，并且在许多应用领域都有发现，比如医学、经济学、生物学、社会学等等。然而，在实际中这些计数往往容易出现测量误差，在这种情况下可归结为错误分类。最近已经有人提出了处理计数错误分类的方法，但仅适用于二项式模型和泊松模型。在这里，我们研究一个更复杂的模型，即零膨胀负二项式模型，并说明如何实现对错误分类的校正。我们的方法通过dmft指数进行了说明，dmft指数是龋齿研究中用于衡量龋齿经历的常用指标。另一个问题是有几位牙科检查人员参与了龋齿经历的评分。通过我们的例子，我们说明了每个检查人员的非差异性错误分类过程如何会导致总体上的差异性错误分类。
