Cross-Validation, the Bayes Theorem, and Small-Sample Bias
研究交叉验证在模型选择中的贝叶斯解释，发现样本分割和重抽样方法会导致后验几率出现系统性偏差，并提出了修正方法。
Researchers in many fields of business are beginning to use cross-validatory techniques to choose between alternative models. In marketing, a Bayesian interpretation of the technique has been proposed. This article examines the Bayesian link in more detail and shows that cross-validatory posterior odds exhibit a systematic departure from exact posterior odds when sample-splitting procedures are employed. Resampling procedures also exhibit a systematic departure when the dimensions of the models differ. Methods of overcoming these biases are proposed.