A Simple Dynamic Graphical Diagnostic Method for Almost Any Model
提出一种基于二维控制器的动态图形诊断方法，通过探索参数拟合邻域来评估模型拟合效果，适用于多种参数模型，并以多元线性回归为例展示应用。
Abstract This article describes a simple dynamic graphical diagnostic method that can be used to explore fits to almost any parametric model. The technique is based on viewing diagnostic plots while fits in a neighborhood of a parametric fit are dynamically explored using an intuitively appealing and easy-to-use two-dimensional controller. The motivation for the controller comes from the 2 p -factorial designs used in response surface analysis. As with response surface analysis, exploration can proceed iteratively. When an interesting new fit is found, the controller can be recentered and exploration continued around the new fit. The technique integrates easily with other dynamic graphical techniques, such as brushing and the use of sliders. The dynamic graphical technique is developed in this article primarily in the context of multiple linear regression. An example of the application of the technique to the Brownlee stack loss data set is presented and compared to the extensive analysis given by Daniel and Wood. The method can also easily be extended to other fitting procedures, such as nonlinear regression.