Win rate: simple to calculate, easy to misread
The most common number in esports is the win rate: wins divided by games played. A team that has won 6 of 10 maps and one that has won 60 of 100 both show 60%, yet the two figures deserve very different levels of trust.
A rough way to see this is to estimate a range around each figure. Using the standard normal approximation for a proportion (which is crude for small counts, so treat these as illustrations only), a 60% record over 10 games gives a range of roughly 30% to 90%. Over 100 games the range narrows to roughly 50% to 70%. The ranges are what statisticians call confidence intervals, and they show that the first record is compatible with a team that is quite weak or quite strong, while the second narrows the possibilities considerably.
Sample size and what the sample contains
Sample size is only half the issue; what the sample contains matters just as much. Several things can make a large pile of results less useful than it looks.
- Patches and balance changes. Competitive games are updated regularly. Results from before a major change may describe a game that no longer exists in the same form.
- Roster changes. A team’s results with a previous line-up may say little about the current five or six players.
- Strength of opposition. A high win rate against weak opponents is not the same as a high win rate against strong ones.
- Format and setting. Online qualifiers, group stages and live finals can differ in conditions and in how seriously teams approach them.
When you see a statistic, it helps to ask how many games it covers, which version of the game, which players, and against whom.
Individual statistics: describing versus predicting
Many titles publish per-player numbers such as kills, deaths, assists, damage dealt or objective participation. These can describe what happened, but they are not always good at predicting what comes next. A role that sets up teammates, for example, may have unimpressive personal numbers while contributing a great deal to winning.
Two statistical cautions apply here. First, a player who has an exceptional tournament often looks less remarkable at the next one; that drift back towards an ordinary level is regression to mean, and it is a reason not to over-read a single hot event. Second, a streak of strong performances can feed the hot hand belief, the feeling that someone is “in form” in a way that overrides everything else. Form may be real, but it is hard to separate from chance in short stretches.
How series formats change the picture
Many tournaments are decided over a series of maps or games, so the chance of winning a series is not the same as the chance of winning one map. As a simplified illustration, suppose a team has a 60% chance on each map and the maps are independent.
- In a best-of-three, the team wins the series 2-0 with probability 0.6 x 0.6 = 0.36. It wins 2-1 by losing exactly one of the first two maps and then winning the third: 2 x 0.6 x 0.4 x 0.6 = 0.288. Adding the two pathways gives 0.36 + 0.288 = 0.648, about 64.8%.
- In a best-of-five, the same per-map figure gives about 68.3%.
Longer series favour the stronger side more because there is more opportunity for the underlying edge to show, a small-scale echo of the law of large numbers. The calculation is a form of the binomial distribution. In reality maps are not independent and are rarely a flat 60%: map pools, side selection and picks and bans all change the per-map probabilities, so real series are messier than this tidy example.
Reading claims about esports numbers
Statistics in commentary and social media are often accurate and still unhelpful. Some habits help.
- Check the denominator. “Undefeated on this map” might mean three games. Cherry-picked statistics tend to pick the stretch or the subset that tells the neatest story.
- Compare like with like. Win rates across different patches, regions or levels of competition are not directly comparable.
- Separate a rate from a trend. A rising or falling run is often within normal fluctuation.
- Remember uncertainty. Even a well-supported estimate of a team’s strength leaves plenty of room for upsets, which are an ordinary feature of competition rather than evidence that the numbers failed.
Prices offered on esports matches express these same estimates as betting odds, and they include a margin like any other market. Statistics can help someone understand a match more clearly, but they cannot make outcomes certain, and money wagered on any result can be lost.