Some Comments on the Use of Response Variable Transformations in Empirical Modelling
本文通过高密度聚乙烯焊接的实例，说明响应变量变换在实证建模中可能导致模型拟合差的问题，并展示通过数据检查和子集分析可得到更优模型。
This paper illustrates some practical difficulties that can arise when response transformations are used in empirical modelling. The ideas are discussed in the context of an investigation into the welding of high density polyethylene. Transformation of the response seems to be strongly indicated but leads to a model with a rather poor fit; this is attributed partly to the effects of the rather extreme variance function implied by the application of the transformation to the observed responses. Informal inspection of the data, together with the analysis of subsets, leads to a more meaningful and better fitting model. The paper concludes with a brief discussion of the effect of transformation on the ease of interpretation of a model.