Second Opinion

What it is

The distinction matters because the two are priced and arranged very differently, and because the gap between them is smaller than most people assume while still pointing one way.

The sharpest available measurement. Sara L. Jackson and colleagues had forty-nine pathologists interpret the same sixty breast specimens twice, at least nine months apart. On atypia, the same pathologist filed the slide under the same heading 53 per cent of the time; a different pathologist agreed 43 per cent of the time. Their own conclusion is the hinge: self-agreement on atypia was similar to observed rates of agreement between different pathologists.

So a second look is worth having, a second reader is worth more, and the difference is ten percentage points on the call where it matters most.

The rule that falls out of it is to ask what kind of call is being made first. In the same study, by the same forty-nine pathologists, invasive breast cancer was called the same way 92 per cent of the time by the same reader and 89 per cent by a different one. On a call professionals find straightforward there is almost nothing left for a second reader to resolve. On a call they split on, the second reader is the whole of the value.

And where the output is a number rather than a category, collect more than one. Gaurav Sood and Andrew Gelman put the consumer instruction in nine words: "If you are a customer, your optimal strategy is to get multiple quotes." That is the arithmetic of averaging arriving as advice, and it is the same reason Alan Boyd's groups of inspectors were more reliable than his individual ones.

The delay is not a detail. The book's own caveat is that asking an expert to judge the same case twice inside one session measures recall, which is why the studies above separated their sittings by months.

In effect

In medicine: the figures, with their limits attached

Joann G. Elmore and colleagues had pathologists re-read the same skin biopsy slides at least eight months later and found agreement of about 35 per cent on moderately dysplastic nevi, with an average of ten different diagnostic terms applied to each case across 187 pathologists and 8,976 interpretations. Carole A. Ridge and colleagues found six thoracic radiologists matching their own earlier classification of the same lung nodule in 77 per cent of cases, one reader in 45 per cent, with at least three weeks between sittings. These are agreement rates, not accuracy rates, and none of these studies measured an outcome for a patient. Nothing here is clinical advice.

In everyday life: the night's sleep against the colleague

The crowd-within result reported in Noise is that asking yourself the same quantity question twice, unwarned, and averaging gains about one tenth as much as a second opinion from somebody else, rising to about one third with three weeks between the guesses. A night's sleep is worth something and a colleague is worth more, and the practical rule is to get another person where you can.

In commerce: inside one firm, or across several

Multiple quotes work where the spread is noise inside one system, because there the identical case was held fixed. Across competing firms the spread is partly a market, so the cheapest quote carries information about pricing as well as about judgement. The question to ask of any striking spread is which of those two it was measured across. The most quoted figure in this field, a 55 per cent median difference between two randomly paired underwriters, was measured inside one firm, and it was published without peer review, method or participant counts.

What it does not say

It does not say a second reader is always worth the money. On the straightforward calls in the same study, agreement ran near ninety per cent either way, and there is little left to buy.

It does not say two readers agreeing makes either right. Every figure on this page is an agreement rate rather than an accuracy rate, and agreement can be high and wrong.

It does not say anything about outcomes. None of these studies followed a patient, and this page is about measurement rather than about care.

And the figures are bounded by how they were read. The headline comparison comes from an abstract read in full, with the full text behind a subscription, and the melanoma study carries a correction this publication has not read.

---