Hindsight bias in horse racing: why every result looks obvious afterwards

The five minutes after a race

Listen to what gets said, in roughly this order.

Bad draw. Never travelled. Too keen early. Didn't get home. Ground was softer than they said. The pace collapsed. The market knew, it was weak all morning. The yard has been quiet for a fortnight.

Every one of those is a real racing factor. Several of them may even be true. The problem is not that the explanations are wrong. The problem is how many of them were available before the gates opened, and how confidently they are offered now.

Before the race there were perhaps a dozen live considerations pointing in different directions, and no clear way to weight them. Afterwards there is one, and it points exactly where the race went.

What the research found

Baruch Fischhoff published the foundational work in 1975, in the Journal of Experimental Psychology: Human Perception and Performance, under the title Hindsight is not equal to foresight. He gave people descriptions of events and told some of them how things turned out, then asked how likely each possible outcome had been.

Receiving outcome knowledge raised the likelihood people assigned to the outcome they had been told about. It changed which pieces of the background they treated as relevant. And, most usefully for anyone trying to guard against it, the participants were largely unaware that any of this was happening. They overestimated both what they would have known without being told, and what other people knew without being told.

Fischhoff called the effect creeping determinism: the tendency, once you know the ending, to perceive it as having been relatively inevitable.

Jonathan Baron and John Hershey identified a separate and equally awkward effect in 1988 in the Journal of Personality and Social Psychology. People judge the quality of a decision by how it turned out. A sensible choice that ended badly gets marked down; a reckless one that came off gets marked up. Neal Roese and Kathleen Vohs reviewed the whole hindsight literature in Perspectives on Psychological Science in 2012.

A good decision and a good outcome are different things. In a sport decided by a length, treating them as the same is a way of learning nothing at all.

Why this is expensive in racing specifically

Racing is close to a worst case for these effects. Outcomes are noisy, they arrive quickly, they are emotionally vivid, and there is always an explanation to hand because so many plausible factors are present in every race.

That combination lets a bettor accumulate what feels like a great deal of experience while learning very little. Each race produces a confident explanation. The explanations are never scored against anything. The sense of understanding grows and the ability to forecast does not move.

It also quietly corrupts the record. A punter who remembers being doubtful about the favourite, having actually been fairly keen, is not lying. Fischhoff's participants were not lying either. The memory has simply been rewritten by the result, and there is nothing to check it against.

We are describing a general human tendency measured in laboratory studies, not diagnosing any individual bettor. Nobody is exempt from it, including us, which is the reason for everything in the next section.

What we do about it

There is no known way to think your way out of hindsight bias. Being aware of it helps far less than people expect, which is precisely what Fischhoff found. The only defence we trust is a mechanical one: write the read down before the race, timestamp it, and never touch it afterwards.

Our daily card is frozen before the first race and published with the time it was locked. The market state that accompanies it is derived from price snapshots taken at fixed clock times before the off, not reconstructed afterwards from a price that has already been contaminated by the result.

When a read is wrong, it stays on the page wrong. When one of our own ideas fails, it goes in the audit as a failure. Of 227 systems and theories examined, 0 have cleared our validation bar, and that number is published rather than buried.

On a single sample card of 57 races, 41 were published as No Read, which is the honest output when the structural evidence is not there. A method that produces a confident story about every race is not reading racing. It is doing what the room does in the five minutes after the line.

The useful version of looking back

None of this means the result is uninformative. Results are the only thing that eventually settles anything. The distinction is between using the result to score what you said, and using it to edit what you said.

The first is how a method improves. The second is how a method stops being a method.

The result can teach you something. It should not be allowed to rewrite what you knew beforehand.

Why favourites lose is the study behind the race everybody remembers. Everyone remembers the beaten favourite is the shorter version of this argument. The evidence room and the system audit are where our own record lives.

See today's read, locked before the first race.