Abstract

Local polynomial regression is a convenient method for smoothing scatterplots with readily available software. However, it is well known that variable amounts of bias are induced by the smoothing operation. This article proposes a simple visualization tool based on approximate confidence intervals which can alert the data analyst to regions of the regression function domain which might be susceptible to unacceptably large levels of bias, and possibly indicating a need for a less automatic smoothing approach. A simulation study verifies the accuracy of the confidence bands, and the method is illustrated with several real datasets.

中文翻译：

---

局部回归中的偏差评估

摘要

局部多项式回归是一种使用现成软件平滑散点图的便捷方法。然而，众所周知，平滑操作会引起可变量的偏差。本文提出了一种基于近似置信区间的简单可视化工具，它可以提醒数据分析师注意回归函数域中可能容易受到不可接受的大水平偏差影响的区域，并可能表明需要一种不太自动的平滑方法。模拟研究验证了置信带的准确性，并用几个真实数据集说明了该方法。