Multiple Comparison Procedures in the Analysis of Covariance
研究了协方差分析中随机协变量均值不同时仍可使用的多重比较方法，并通过市场实验示例说明其实际应用。
Abstract Currently available range-type procedures for constructing simultaneous confidence intervals on contrasts in analysis of covariance (ANCOVA) models with random covariates assume that the vectors of covariates associated with the observations are identically distributed. Conditions are given under which the mean vectors of these covariates may differ without affecting the validity of such procedures. An example of a controlled test market experiment illustrates the usefulness of the relaxed assumptions in practice. Critical values for Duncan's multiple-range test in ANCOVA and an investigation of the efficiency of range-type procedures relative to Scheffé's method are included.