Why statistics helps with games
Probability tells you what could happen on the next roll, deal or spin. Statistics works in the other direction: it starts from what did happen and asks what that can and cannot tell you. Both matter for anyone who plays games, because most of what we see is a short run of results, and short runs are noisy.
This overview covers the ideas that come up again and again when you look at game results. Each one has a longer treatment elsewhere on the site. The goal here is to show how they fit together and to give you a few questions worth asking before you trust a number.
Everything below is educational. Statistics can describe randomness and help you avoid misreading it, but it cannot turn a game with a built-in disadvantage into a winning proposition. Outcomes in games of chance remain uncertain.
Averages and what to expect
The most useful single number for a random event is its average outcome over many repetitions, often called the expected value. A fair six-sided die averages 3.5. No roll ever shows 3.5, yet that is where the long-run average of many rolls settles. That average is the basis for judging whether a bet or choice is favourable on average.
An average on its own can hide a lot. Two games can have the same average and feel completely different to play, which leads to the next idea.
Spread: variance and standard deviation
Spread measures how far individual results typically sit from the average. For one die, the typical distance from 3.5 is about 1.7. That figure is the standard deviation, and its square (about 2.92) is the variance.
Spread matters because it sets how bumpy the ride is. Imagine two made-up games where you win or lose a fixed amount on a fair coin flip. In one the amount is 1, in the other it is 10. Both have an average of zero, but the second swings ten times as widely. People often describe this difference as volatility and return, and in card games it shows up as poker variance, where sound decisions can still lose for long stretches.
Totals bunch up around the middle
Add up many small random events and the total tends to cluster around its average, with fewer results far out in the tails. Roll ten dice and the total is most often somewhere near 35, and almost never near 10 or 60. Plotted, the totals form the familiar bell shape of the normal distribution.
This is why a typical range can be estimated with a simple rule of thumb: roughly two-thirds of results land within one standard deviation of the average, and roughly 95 percent within two. For ten dice the standard deviation of the total is about 5.4, so most totals fall between about 30 and 40, and nearly all between about 24 and 46. The rule is an approximation and works best for sums of many small independent events.
More trials: what the law of large numbers says
As you repeat something many times, the average result gets closer to the true average. That is the law of large numbers. It is often misread as a promise that luck evens out. It does not: the share of heads in a long run of flips approaches one half, but the raw gap between heads and tails can keep growing. Old results are diluted by new ones; they are never cancelled.
This is also why sample size is the first thing to check when you see a win rate or a streak. A hundred results tell you far less than ten thousand.
Uncertainty around a win rate
Suppose you win 60 of 100 games of something with a roughly 50 percent chance. It is tempting to say your true win rate is 60 percent. Statistics offers a more honest statement: a range. Using a common approximation, the 95 percent interval for 60 wins in 100 games runs from about 50 percent to about 70 percent. With 1,000 games and 600 wins, the same observed rate gives a range of roughly 57 to 63 percent.
These ranges are called confidence intervals. They shrink slowly: to halve the width you need about four times as many games. They also assume the games were comparable and independent, which is often not true in real play where skill, opponents and conditions change.
Updating on new information
Probabilities change when you learn something. If one ace has already been dealt from a standard 52-card deck, the chance that the next card is an ace falls from 4 in 52 to 3 in 51. This is conditional probability: the chance of an event given something you know.
Bayes theorem extends the idea. It is a method for updating how likely a hypothesis is after seeing evidence, such as how suspicious a die should look after a surprising roll. Its main lesson is that you must weigh the evidence against how likely the hypothesis was to begin with. A related puzzle, the Monty Hall problem, shows how much a piece of information can shift odds that seem to be 50-50.
Extremes tend to fade
An unusually good or bad run is usually a mix of underlying ability and luck, and the luck part does not repeat on command. After an extreme result, the next result is typically closer to the average. This is regression to mean, and it explains many apparent hot streaks and slumps without any change in skill.
It does not mean the game owes you a correction. If you lose ten flips in a row, the next flip is still a fair flip. The distinction between luck versus skill is partly a question of how much of a result you can expect to repeat.
Reading claims with suspicion
Numbers in advertising, forums and social posts are often chosen because they look impressive. Common tricks include choosing a favourable time window, showing only the winners, quoting a percentage without saying how many cases it is based on, and testing many variations until one looks good. These are the habits behind cherry-picked statistics.
A few questions help: how many cases is this based on, who is missing from the data, what else was tried, and what would the result look like by random chance alone? A claim that cannot answer these is weak evidence, however confident it sounds.
Putting it together
A sensible routine for any set of game results is short:
- Work out the average outcome you would expect from the rules alone.
- Estimate how much spread is normal, so you know what an ordinary swing looks like.
- Note how many results you have, and treat small samples as tentative.
- Put a range around any observed rate rather than quoting one figure.
- Ask whether the sample was chosen in a way that favours a story.
If you want to explore these ideas without risking anything, a Monte Carlo simulation of dice or cards lets you watch averages settle and ranges form on your own computer.
A final reminder: games with a negative expectation lose money on average over time, and no statistical technique changes that. If gambling is causing you or someone close to you harm, a local support service can help, and it is never a good idea to borrow money to play or to keep playing to recover losses.