Each horse has six hidden ratings: top speed, acceleration, stamina, consistency, late finish and quick start. When you press the race button, the result is worked out in one go by a short simulation. The animation you watch afterwards is a replay of that result. Nothing you do during the animation can change it.
The luck term means the best horse does not always win. Over many races the stronger horses win more often, but any single race can go to an outsider. That is what gives the odds their meaning.
Sprints reward speed and acceleration. Marathons reward stamina and a strong finish. Switching distance changes which horse is favourite, sometimes by a lot.
Before each race the same simulation is run 3,000 times and the wins for each horse are counted. That count gives the "chance" shown on each button. The payout is the fair price for that chance, less a 10% margin.
No. Because the same margin is taken from every price, every horse returns about 90 coins for each 100 staked on average. A favourite pays little and wins often. A long shot pays a lot and rarely wins. The average return is the same. So picking between them is about how much risk you like, not about finding a bargain. That is the point of the first hypothesis.
Each race uses a pseudo-random number generator (mulberry32) seeded from the clock and the trial number, so every race gets a fresh draw. The odds use a fixed set of 3,000 seeds, so the same field and distance always give the same prices. The engine never reads your pick or your stake before working out the result.
If a horse truly wins one race in four, a fair payout would be 4 times your stake. This game pays 3.6 times. The 10% difference is the margin. In the short run you can still win, sometimes by a lot. Over a long run the margin pulls the balance down, which is how real betting works too.
With 3,000 simulations the chance shown for each horse is accurate to within about 1.6 percentage points (95% interval) of its true value.
The green bar is the chance implied by the price on offer. The red mark is the chance from the simulation. The green bar always runs slightly past the red mark. That gap is the margin.
When the odds are fair apart from a known margin, and the true chances are shown on screen, does a person still bet in the ways prospect theory and the gambling literature predict? A risk-neutral bettor with full information would have no reason to prefer long shots, would not change their stake depending on the last result, and would treat winnings and starting coins the same. This study measures how far one person's play moves away from that.
This is a single-case, within-subject design (Barlow et al., 2009). There is no control group. Each hypothesis compares your own trials with each other, for example your stakes after a win against your stakes after a loss. The conditions being compared (win or loss, ahead or behind, close second or clear loss) are not set by the researcher. They come out of the race engine. Once you have chosen a horse, whether it wins is decided by the random draw, so the previous outcome works as a natural random assignment. Your choice of horse still shapes the odds of winning, which is why this is described as a quasi-experimental contrast and not a true experiment.
One adult (18+) per session. People with a history of problem gambling are asked not to take part.
Everyone starts with 100 coins. On each trial you choose a distance (optional), a horse, and a stake as a share of your current balance (5%, 10%, 25%, 50% or all in). The stake has to be chosen fresh each trial, so it cannot just carry over from the last race by default. The session ends at 60 trials, when the balance reaches zero, or when an all-in bet loses. The trial count was fixed in advance so that the report is not affected by stopping when the results look good.
Before the race: distance, the chance of every horse in the field, the horse picked, its odds, the stake, the balance, the stake as a share of balance, and how long the choice took. After the race: whether you won, where your horse finished, the margin in lengths, and the new balance. All of it can be downloaded as a CSV from the results screen.
No account is needed and no data about your play is sent anywhere. Everything is stored in this browser and can be deleted with the new session button. The CSV download is the only way the data leaves the device, and only if you choose to share it. There is no real money involved.
Support. In Aotearoa New Zealand, the Gambling Helpline is free on 0800 654 655. Gambling Therapy (gamblingtherapy.org) offers free support worldwide.
Barlow, D. H., Nock, M. K., & Hersen, M. (2009). Single case experimental designs: Strategies for studying behavior change (3rd ed.). Pearson.
Clark, L., Lawrence, A. J., Astley-Jones, F., & Gray, N. (2009). Gambling near-misses enhance motivation to gamble and recruit win-related brain circuitry. Neuron, 61(3), 481–490.
Edgington, E. S., & Onghena, P. (2007). Randomization tests (4th ed.). Chapman & Hall/CRC.
Holm, S. (1979). A simple sequentially rejective multiple test procedure. Scandinavian Journal of Statistics, 6(2), 65–70.
Kahneman, D., & Tversky, A. (1979). Prospect theory: An analysis of decision under risk. Econometrica, 47(2), 263–291.
Reid, R. L. (1986). The psychology of the near miss. Journal of Gambling Behavior, 2(1), 32–39.
Snowberg, E., & Wolfers, J. (2010). Explaining the favorite–long shot bias: Is it risk-love or misperceptions? Journal of Political Economy, 118(4), 723–746.
Thaler, R. H., & Johnson, E. J. (1990). Gambling with the house money and trying to break even: The effects of prior outcomes on risky choice. Management Science, 36(6), 643–660.
Tversky, A., & Kahneman, D. (1992). Advances in prospect theory: Cumulative representation of uncertainty. Journal of Risk and Uncertainty, 5(4), 297–323.
NZ Gambling Helpline: 0800 654 655 (free, 24/7). In-game coins have no real value.