A General Method for Constructing Simultaneous Confidence Intervals

Abstract

Abstract A general method is proposed for constructing simultaneous confidence intervals that gives special emphasis to any preselected finite spanning subset in a linear space of estimable functions. The method has a wide range of application, including regression analysis. The length of a confidence interval is determined as the solution to a linear programming problem. The studentized maximum modulus distribution can be used to choose critical values for conservative control of the overall confidence level. The method is compared to Scheffé's procedure.