ABSTRACT
Statistics deals with errors, as cleaners do with dust and dustmen do with our domestic refuse. Indeed, hypothesis testing was a great statistical innovation in the 1950s, and its subsequent promotion into the mainstream statistical practice is one of the great success stories in science. First, the hypothesis and the alternative are treated asymmetrically; in the absence of evidence against it, the hypothesis is promoted, even though there is no evidence to support it. The hypothesis is treated as a default, applicable until it is contradicted by evidence. And, the probabilities handled in hypothesis tests are hypothetical, evaluated under temporarily adopted assumptions, an exercise abstract even to the most hardened theoretician. Statistical errors are not made equal; we all agree that they differ in magnitude and frequency, and the characterisation of the features, by standard errors, false discovery rates, and similar quantities is an important preoccupation in both theory and practice of statistics.
