Readings › Mayo 1996 › Extract

Extract · a self-contained text inside a work

The severity requirement

Mayo, Error and the growth of experimental knowledge (1996), ch. 6, §6.2, pp. 178–183.

Extract published · 0 claims drawn

What it says, in one sentence

Evidence counts for a hypothesis only if it fits the hypothesis and the test that produced it would very probably have gone the other way, had the hypothesis been false.

The prospectus

Each line restated in my notation. P(… ; H false) reads “the probability of …, in trials where H is false”: a frequency of the test, not a degree of belief in H.

Two requirements

aFitH fits e

The minimum every approach shares: the outcome is what H would lead you to expect. How “fit” is measured can vary with the problem.

bGood teste is evidence for H ⇔ H fits e ∧ T is severe for H

Having evidence for H and having passed a good test of H are the same thing. So “a poor test, but still evidence” is ruled out.

The severity criterion, four equivalent forms

1aPass, highP(no pass as good as e ; H false) ≈ 1

Had H been false, the test would very probably not have passed it this well.

1bPass, lowP(a pass as good as e ; H false) ≈ 0

The same, stated as the improbability of the pass.

2aFit, highP(a worse fit ; H false) ≈ 1

Restated in terms of fit instead of pass.

2bFit, lowP(so good a fit ; H false) ≈ 0

The a forms speak of the test’s power to catch a false H; the b forms of how rarely it misses.

The simplest case

SC*Pass–fail testP(T fails H ; H false) ≈ 1

With a cut-off, every test becomes pass or fail. Too coarse to judge one result, but the form to keep in mind: re-cut the scores at the one obtained and the general criterion follows.

The scale

Severity runs from 0 to 1. The two ends are limits; the exam is Mayo’s own example.

0Zero severity
would pass H even if false: no test at all
.40Exam T2
fails 40% of under-prepared students
.99 · 1Exam T1 · maximum
passing it is good evidence; at 1, passing entails H

Passing a zero-severity test does not make H false. It only gives no reason to accept it: one can be right for the wrong reasons.

Where it sits in the work

177178179180181182183

pp. 177–183 · 14 marks

Claims drawn from this extract

No claims drawn yet.

Paraphrase and notation CC BY-NC 4.0, Claudio Cammarano. The formulations restate the author’s; they are not quotations. Back to the card of the work.