The favourite-longshot bias is the tendency for long shots to return less per dollar staked than favourites. Their prices imply a better chance than they turn out to have, while the shortfall is smaller, or absent, at the short end. R. M. Griffith described it in American horse racing in 1949, in the American Journal of Psychology.
It is a pattern across thousands of bets, not a verdict on any one runner, and you can measure whether it holds in your races by price band.
What the bias looks like across price bands
Return per $1 is the measure: what $1 on every runner in a group would have returned, divided by the number of runners. With illustrative numbers for one season, every runner grouped by starting price (one price per band, for simplicity):
| Starting price | Runners | Winners | Implied chance: 1 / price | Strike rate: winners / runners | Return per $1: winners x price / runners |
|---|---|---|---|---|---|
| $2.00 | 500 | 238 | 50.0% | 47.6% | 95.2c |
| $4.00 | 600 | 138 | 25.0% | 23.0% | 92.0c |
| $8.00 | 700 | 77 | 12.5% | 11.0% | 88.0c |
| $16.00 | 800 | 42 | 6.25% | 5.25% | 84.0c |
| $34.00 | 1,000 | 23 | 2.94% | 2.30% | 78.2c |
Every band returns less than a dollar, because every price carries the margin, but the shortfall grows with the price: the $2.00 runners win 47.6% against an implied 50.0%, the $34.00 runners 2.30% against 2.94%. That widening gap is the bias, and roughie is the Australian racing word for the long end.
Note: Illustrative numbers, not measured results: the bias varies by market, code and period, so measure it on a dated set of results before relying on any figure.
Why long shots may be overbet
Researchers still disagree about the cause. Richard Thaler and William Ziemba's 1988 review in the Journal of Economic Perspectives found racetrack odds to be good estimates of winning chances overall, with this bias among the anomalies that stood out. Three explanations lead:
| Explanation | The idea | Who argued it |
|---|---|---|
| Small chances feel bigger than they are | Punters overrate unlikely results, so long shots draw more money than their chance deserves | Griffith in 1949; Erik Snowberg and Justin Wolfers (Journal of Political Economy, 2010) found misjudged chances fitted US racing data better than a taste for risk |
| Paying for the thrill | Some punters accept a worse average return for the chance of a big collect, as with a lottery ticket | The risk-loving account Snowberg and Wolfers tested against |
| Protection against informed money | A bookmaker who fears better-informed punters prices long shots shorter, where an informed bet would cost it most | Hyun Song Shin's models of bookmaker pricing (Economic Journal, 1992 and 1993) |
These may all play a part. Shin's idea also sits behind the Shin method of taking out the margin, which puts more of it on long shots.
How to measure it by price band
You can run the test on past races or on your own records:
- Pick one price type and keep to it: starting price, Betfair Starting Price, tote dividend, or the fixed price at a set time before the jump.
- Include every runner from a stated date range, not only the bets you placed.
- Group runners into bands, such as under $3.00, $3.00 to $5.99, $6.00 to $11.99, $12.00 to $25.99 and $26.00 or more.
- For each band, divide what $1 on every runner would have returned by the number of runners.
- Check each band's sample: at a single price, the standard error of return per $1 is about price x square root of (strike rate x (1 - strike rate) / runners).
Step 5 decides whether a band means anything. For the two ends of the table:
| Band | Standard error of return per $1 | Two standard errors either side |
|---|---|---|
| $2.00, 500 runners | 2.00 x square root of (0.476 x 0.524 / 500) = 4.5c | 95.2c plus or minus 2 x 4.5c: about 86c to $1.04 |
| $34.00, 1,000 runners | 34.00 x square root of (0.023 x 0.977 / 1,000) = 16.1c | 78.2c plus or minus 2 x 16.1c: about 46c to $1.10 |
So 1,000 runners at $34.00 prove nothing on their own. Cutting that error to 5 cents takes about 34 x 34 x 0.023 x 0.977 / (0.05 x 0.05) = 10,400 runners, which is why sample size in betting bites hardest at long prices.
Tote, fixed odds and the exchange
Where a bias can come from depends on who sets the price:
| Market | How the price is set | Where a bias can come from |
|---|---|---|
| Tote | The operator takes its commission from the pool and shares the rest among winning units, as a dividend declared after the result | Mainly how punters spread their money, since commission comes off the whole pool before it is shared; dividend rounding, below, leans the other way |
| Fixed odds | The bookmaker sets or follows each price and decides where its margin sits | Punters' money, plus any choice to load margin on long prices, or to shade a popular favourite instead |
| Exchange | Punters back and lay each other; commission comes from net winnings in a market | Punters' money, and thin markets with wide gaps between back and lay at long prices |
Under the Northern Territory's tote rules (as in force at 14 April 2020), for example, each dividend is rounded down to a multiple of 5 cents, which costs a short price proportionally more. An illustrative dividend worked out at $1.38 pays $1.35, 1 - 1.35 / 1.38 = 2.2% less, while $20.48 pays $20.45, 1 - 20.45 / 20.48 = about 0.15% less.
On exchanges, a 2006 study of UK racing in Economica by Michael Smith, David Paton and Leighton Vaughan Williams found significantly lower biases than in traditional bookmaker markets. That is UK evidence, so measure your own before assuming the same here.
Multis and exotics: where the shortfall compounds
In a multi the legs' prices multiply, and so do their average returns if the legs are independent. Using the illustrative band returns for each leg:
| Multi | Combined price | Average return per $1: leg returns multiplied |
|---|---|---|
| Two $2.00 legs | $4.00 | 0.952 x 0.952 = 90.6c |
| Two $16.00 legs | $256.00 | 0.84 x 0.84 = 70.6c |
| Three $8.00 legs | $512.00 | 0.88 x 0.88 x 0.88 = 68.1c |
On these numbers, the bigger the combined price, the worse the average return. The multi bet calculator multiplies the legs; a same game multi is priced differently again, because its legs are linked.
Exotics work through a pool: a trifecta (first three in order) pays from the net pool shared among winning units, so it is generous only when other punters underbet the winning combination. Whether long-shot combinations are overbet is a question for each pool's declared dividends.
A box spreads the stake across every order. A five-runner boxed trifecta is 5 x 4 x 3 = 60 combinations, and if two of the five are long shots, 60 - 3 x 2 x 1 = 54 of them include one. So 54 / 60 = 90% of the box's stake sits on long-shot combinations, where any overbetting in the pool would cut your return.
Risk: Betting involves risk. A multi or an exotic loses unless every part comes in, and a big dividend is rare by design. See responsible gambling for limits and support.
Are outsiders bad value?
Not as a rule. A long shot is value when its price beats its fair price, like any other bet, as value betting sets out; the bias matters most in working out that fair price.
Take an illustrative race priced $2.80, $4.00, $5.50, $8.00, $11.00, $15.00, $21.00 and $26.00. Its 1 / price figures sum to a market of 115.76%. Taking the margin off every runner in proportion gives the $26.00 outsider a fair price of 26.00 x 1.1576 = $30.10. The power method takes more off long prices. It raises each implied chance to the power that makes the field add to exactly 100%, here 1.0877, giving the outsider (1 / 26) to the power 1.0877 = 2.89%, a fair price of $34.60.
| Another bookmaker's price on the outsider | Edge against $30.10, proportional | Edge against $34.60, power method |
|---|---|---|
| $31.00 | 31.00 / 30.10 - 1 = +3.0% | 31.00 / 34.60 - 1 = -10.4% |
The same price is value on one method and a poor bet on the other. If the bias holds in your races, proportional scaling flatters long prices. How to remove the bookmaker margin compares the methods, the fair odds calculator runs both, and a race's market percentage shows how much margin there is to share out.
Risk: Betting involves risk. A fair price from any de-vig method is an estimate, and a long shot priced above it still loses most of the time. See responsible gambling for limits and support.
Using the bias without fooling yourself
Two uses for a betting strategy, each with its catch:
- De-vig races with a method that puts more margin on long shots, then check the result against the exchange near the jump.
- Laying long shots on an exchange takes the other side of the bias. But an illustrative $10 lay at $34.00 risks 10 x 33 = $330 to win $10 before commission, so one winner costs as much as 33 losers pay.
| Mistake | What it costs |
|---|---|
| Backing favourites because they lose less | They still lose on average: 95.2 cents per $1 in the illustrative $2.00 band |
| Reading a few hundred long-shot bets as proof | Even 1,000 runners at $34.00 leave a range of about 46 cents to $1.10 |
| Mixing price types in one band | Starting prices, BSP, tote dividends and early fixed prices are different prices, so a mixed band measures none of them |
Risk: Betting involves risk. A bias measured on past races is no guarantee of future results, a lay at a long price can cost many times its stake, and there is no guarantee of profit. See responsible gambling for limits and support.
Checking long prices on B337's Terminal
B337's Terminal shows prices from 40+ bookmakers across racing and sports side by side, with Betfair back and lay beside them, so you can see whether a bookmaker's long prices sit short of the exchange on the same runners. EV overlays compare each price with an estimated fair price: an estimate, not a forecast.
Prices are read on a repeating cycle, so confirm any price in your bookmaker account before you bet. A free account opens a limited view of the Terminal with live odds. Betfair is a trade mark of its owner, and B337 is not affiliated with it. Bookmaker names are trade marks of their owners. B337 is not affiliated with them.
For free and confidential support call 1800 858 858 or visit gamblinghelponline.org.au.
Risk: Betting involves risk. A long price that sits above the exchange can still lose, and usually does, and a price on screen can move before your bet is accepted. See responsible gambling for limits and support.