Local Polynomial Regression for Binary Response
Local Polynomial Regression for Binary Response
复制标题
二元响应的局部多项式回归
DOI：
--
复制
发表时间：
1997
期刊：
影响因子：
--
通讯作者：
N. Altman
中科院分区：
文献类型：
--
作者：
A. Aragaki;N. Altman
TL-1379-M非参数回归方法可以为非参数建模、参数模型的选择和诊断工具的选择提供指导。这对于二元回归尤其重要，因为缺乏用于数据探索的简单图形工具。本文讨论了局部多项式回归在二元回归问题中的应用。我们证明了局部多项式回归是一致的，是广义光滑模型的一种更简单的替代。良好的小样本性能的带宽选择仍然是个问题。我们通过模拟表明，交叉验证和“插入式”估计器等方法对连续响应表现较好，但对二进制数据表现较差。然而，引导带宽选择虽然非常耗费计算机，但似乎对二进制响应很有效。
tL- 1379- M Nonparametric regression methods can provide nonparametric modeling, guidance in selection of parametric models and diagnostic tools. This is particularly important for binary regression due to the lack of simple graphical tools for data exploration. In this article, we discuss the application of local polynomial regression to the binary re gression problem. We show that local polynomial regression is consistent, and is a simpler al ternative to generalized smooth models. Bandwidth selection for good small sample perfor mance remains problematic. We show by simulation that methods such as cross-validation and "plug-in" estimators, which perform well for continuous response, do poorly for binary data. However, bootstrap bandwidth selection, although very computer-intensive, appears to work well for binary response.