Strike rate and sample size

Strike rate and sample size

Strike rate is the percentage of selections that won. Back forty horses, ten win, that's a 25% strike rate. It's the single most quoted number in racing analysis and, on its own, close to useless.

A strike rate tells you nothing about price. A system that wins one race in four could be selecting 6/4 shots or 8/1 shots, and those two versions of 25% describe completely different pictures. One is probably making little sense of the market. The other might be finding something the odds haven't fully priced in.

So the question isn't really "what's the strike rate". It's "what's the strike rate, at what prices, over how many runs, with the losers included". Miss any one of those and the number is decoration.

Why 40% and 15% can mean the same thing

Picture two records. One shows a 40% strike rate, but every selection is odds-on or close to it, say 4/6 to 5/4. The other shows 15%, with selections mostly out at 5/1, 8/1, sometimes bigger. Read as raw percentages, the first sounds far stronger.

Work through the maths on paper and the gap narrows fast. Four winners at 4/6 from ten bets returns less than four winners at 8/1 would from the same ten runners, and it takes surprisingly few winners at bigger odds to outweigh a stack of short-priced ones. That's not a quirk. It's how fractional odds work: price times outcome, not outcome alone.

Neither record is automatically better. A high strike rate at odds-on can reflect solid, unspectacular judgement, correctly identifying horses the market has already got right. A lower strike rate at bigger prices might reflect the opposite: fewer correct calls, but calls the market hadn't caught up with yet. You can't rank the two without knowing the prices attached to every result, winners and losers both.

Why small samples mislead

Twenty bets is not a sample, it's a coin flip with extra steps. Horse racing has enough randomness in it, a stumble at the start, a wide draw, a rider's split-second call, that short runs bounce around wildly regardless of how sound the method behind them is. A cold spell of ten losers can happen to a good process. So can a hot streak of eight winners from a poor one.

Variance shrinks as the sample grows, but it doesn't vanish, and it shrinks slower than most people expect. A few hundred selections starts to mean something. A few thousand, spread across different tracks, distances, and going conditions, starts to look like a genuine record rather than a lucky or unlucky run.

This cuts both ways. Don't write off a method because it had a rough month, and don't trust one because it had a hot one. Ask how many selections sit behind the number before deciding whether the number is telling you anything at all.

What a proper published record actually shows

A record worth reading is dated. Every selection has a date attached, so you can see when it was made and check it wasn't quietly rewritten after the result was known. Racing has a long history of records that look sharper in hindsight than they ever were in real time.

It's complete. Every selection is in there, not a curated highlights reel of the winners with the losers trimmed off the bottom. A record that only shows what came in tells you nothing about what the method actually does when it's wrong, and it's always wrong some of the time.

It's settled at a real, obtainable price, not the best price of the day picked out after the fact, and not starting price alone if that's not what was available when the selection went up. And it separates strike rate from return, showing both, because either number by itself hides half the story. The Racing Bot publishes its record this way, dated and complete, because a record that's easy to check is the only kind worth trusting.

Reading a record like this

Next time a strike rate gets quoted at you, ask three questions before you take it anywhere. What were the prices? How many selections is it built on? And is the losing run in there alongside the winning one?

A 15% strike rate at long odds and a 40% strike rate at odds-on aren't rivals, they're different sports wearing the same number. Sample size decides whether either figure is worth listening to in the first place, and completeness decides whether you're looking at a record or a highlights reel dressed up as one.