The problem with a single percentage
Suppose a player wins 12 of 20 games, or a loot-drop setting appears to pay out 3 times in 50 tries. The observed rate is easy to compute, but the real question is different: what is the underlying rate that produced these results? A limited sample only gives a noisy estimate of it.
A confidence interval is a way of stating that noise honestly, as a range rather than a single number.
The basic formula
For a win rate measured over n independent trials, with an observed proportion of wins p, a common approximation for a 95% interval is:
p plus or minus 1.96 x the square root of (p x (1 - p) / n)
The square root term is the standard error, the typical size of sampling noise in the estimate. It comes from the same ideas as the standard deviation of a binomial count, divided by n. The number 1.96 is the usual multiplier for 95% when results are close to a bell curve.
This approximation works best when n is reasonably large and p is not near 0 or 1. For very small samples, other methods are used, so treat the first example below as a rough guide only.
Worked examples at 50%
Imagine a fair-looking process that produced exactly half wins each time, at different sample sizes. These are illustrations, not data from any real game.
- 10 wins in 20: the standard error is the square root of (0.25 / 20), which is about 0.112. Multiplying by 1.96 gives about 0.22. The interval runs from roughly 0.28 to 0.72.
- 100 wins in 200: the standard error is the square root of (0.25 / 200), about 0.035. Multiplying by 1.96 gives about 0.07, so the interval is roughly 0.43 to 0.57.
- 1,000 wins in 2,000: the standard error is about 0.011, the margin is about 0.022, and the interval is roughly 0.48 to 0.52.
Notice the pattern. To halve the width of the interval you need about four times as many trials, because the width shrinks with the square root of n.
What 60 wins in 100 really says
Take 60 wins in 100 games. The observed rate is 0.6. The standard error is the square root of (0.6 x 0.4 / 100) = the square root of 0.0024, about 0.049. The margin is 1.96 x 0.049, about 0.096. The interval is roughly 0.50 to 0.70.
The lower end touches 50%. So 60 wins in 100 is a result that a true 50% process could plausibly produce, and it does not cleanly show a better-than-even rate. It also does not rule out a true rate around 70%. The sample is simply too small to say.
This is why small sample sizes produce misleading records. A short winning run feels like evidence of skill or a favourable rate, but the interval is wide enough to include quite different explanations.
What a confidence interval does not mean
Confidence intervals are often misread.
- A 95% interval does not mean there is a 95% chance the true rate sits inside this particular range. The 95% describes the method: if the same procedure were repeated on many fresh samples, about 95% of the intervals built would contain the true rate.
- It does not account for bias. If the trials are not independent, if the game changes between plays, or if you selected which results to report, the interval is not trustworthy. Cherry-picked statistics break the assumptions behind the formula.
- It does not describe the next result. Even if you knew the true rate exactly, each trial would still be uncertain.
Applying it to games of chance
For a casino game with known rules, the true win probability can usually be worked out directly, so a confidence interval is less useful than the calculation. Intervals matter where the true rate is unknown, such as estimating whether a coin is fair, checking an advertised drop rates figure from player reports, or judging a player’s results in an esports win probability discussion.
They also help with a common trap in gambling. A player who has been ahead for a few hundred plays may feel they have found something that works. The interval around their observed rate is often wide enough to include a rate that is still unfavourable. Any game with a house edge is expected to cost the player over time, and recent results do not alter the odds of the next play.
A practical habit
When you see a percentage, ask how many trials stand behind it. If the answer is small, expect a wide range. Estimating sensibly means accepting uncertainty rather than treating a few results as the truth.
If gambling is becoming a problem, a local support service can offer help.