ABSTRACT
The 18-year-old failure of the Berger–Exner test to increase the attention paid to selection bias in clinical trials indicates that we need a new approach to the problem. Using elements of the test from the start of a clinical trial rather than waiting until the outcome data are collected would be new. How this can be accomplished is considered under five subheadings: Replacing Blocks with Minimization, Response of Minimization to Selection Bias, Flexible Minimization, Prospective Search for Selection Bias, and More Efficient Algorithms. Flexible Minimization and monitoring for the required imbalances needed for successful selection bias makes simultaneous reductions of selection and accidental bias possible for the first time. It also holds promise of reducing the current diversity that exists in statisticians’ advice on conducting clinical trials.
