Royal Statistical Society Discussion on Randomization vs Model Based Inference July 1, 2026

Senn’s discussion of marginal versus conditional inference assumes, implicitly, that the trial population already represents a causally coherent target. He largely accepts the gate-defined population as the object of inference and then optimizes estimation within it. The entire RCT enterprise generally does the same thing ignoring the generation of the third layer estimand. This allows unlimited trials based on synthetic data generating processes and false RCT transport, as was observed during the COVID pneumonia ventilator crisis. Given the magnitude of risk of false RCT transport, no broad discussion of randomization and covariate consideration should proceed without addressing the fundamental flaw which has produced worldwide harm, specifically randomization of a synthetic mixture of diseases and mechanisms. Given the magnitude of this field error, such discussions should begin by stating “we start by assuming a causally coherent pathway”. Without that, this leaves unexamined a prior and potentially more fundamental question: what biological pathway or disease process defines the enrolled population? Without this the entire design process has often proven to be a harmful mathematical embellishment of an apical error. It was an unstated assumption (which was false) that caused the decades old error, so assumptions must be stated lest another generation continue to bypass the analysis of the gate.