What “edge” means in an analytical sense
In esports, an “edge” is usually discussed as a difference between two estimates: how strong a team looks on paper and how strong it really is. Analysts try to narrow that gap by building better estimates of win probability. Nothing here is a way to make money from matches. The aim is to understand what a number like “62% to win” means and how much weight it deserves.
From ratings to a win probability
The most common starting point is a rating system. Each team or player has a number, and the gap between two numbers is turned into an expected result. Elo ratings use a simple formula: the expected score of the higher-rated side is 1 divided by (1 + 10 raised to the power of minus the rating gap divided by 400).
Using that standard scale as an illustration:
| Rating gap | Expected score of the higher-rated side |
|---|---|
| 0 points | 50% |
| 100 points | about 64% |
| 200 points | about 76% |
A 100-point gap gives 1 / (1 + 10^-0.25), which is 1 / 1.562, or about 0.64. A 200-point gap gives 1 / (1 + 10^-0.5), which is 1 / 1.316, or about 0.76. Real systems used for matchmaking or rankings may use different scales, so these figures are illustrations of the method, not values for any particular game.
“Expected score” is also not quite “chance of winning” in games with draws, but for a game with a clear winner and loser they are treated as the same thing.
From one map to a whole series
Many esports events are played as a series: best-of-three or best-of-five. If a model gives a team a per-map chance, the series chance follows from the binomial distribution, assuming maps are independent. With a 60% chance per map, a best-of-three is won with probability 0.6 x 0.6 + 2 x 0.6 x 0.6 x 0.4 = 0.36 + 0.288 = 0.648. A 60% map favourite is therefore about a 65% series favourite. The edge grows slowly, and the underdog still takes the series more than one time in three.
The independence assumption is itself shaky. Momentum, adaptation between maps and the order in which maps are picked can all make results within a series connected.
Why the estimates are uncertain
The key point is that the rating is not a fact. It is an estimate built from a limited number of results, and it has its own error.
Small sample sizes. Teams in many esports play fewer officially recorded matches against each other than traditional sports clubs do, and opponents differ from region to region. A rating built from a few dozen results can move a long way with a few more. Statisticians describe this with confidence intervals: a team rated at a particular value might plausibly be somewhat stronger or weaker, and that range feeds straight into the win probability.
Changing conditions. Game patches can alter which characters, items or tactics are strong. A roster change, a new coach or a stand-in player can make last month’s rating out of date. A rating describes a team as it was, and the team entering the next match may be different.
Format and context. Map pools, side selection, travel, online versus in-person play and tournament pressure all affect results, and ratings usually ignore them.
Unmeasured factors. Preparation, draft strategy and individual form are hard to put in a spreadsheet. Treating a number as complete leaves out the things that often decide a close match.
How analysts try to improve estimates
Better models do not remove uncertainty, but they can reflect it more honestly.
- Adding more information. Some models combine team ratings with individual player ratings, recent form, map-specific results or head-to-head history. More inputs help only if they are measured reliably and the sample behind each one is large enough.
- Updating with new evidence. After each match the estimate shifts. This is the same logic as Bayes theorem: a prior view about team strength is combined with new results to produce a revised view.
- Pulling extreme results back. Unusually good or bad runs tend to partly reverse, so careful models shrink extreme ratings towards the average, reflecting regression to the mean.
- Checking calibration. If a model says 70% many times, about seven in ten of those events should happen. Comparing forecasts with outcomes over time is the honest test of whether a model is any good.
Reading a percentage sensibly
A few habits help when you see a win probability quoted on a stream or in an article:
- Treat it as a range. A figure like 64% may be well supported or may reflect very little data. The headline number rarely shows which.
- Expect the less likely side to win regularly. A 36% outcome occurs roughly one time in three.
- Be careful about causes. A pattern between a statistic and winning does not show that the statistic produced the win, a point covered under correlation versus causation.
- Do not stretch one match into a verdict. A single upset does not prove a rating was wrong.
Risk note
Because estimates in esports are rough, they are a poor basis for staking money. Any odds offered on matches also build in a margin for the provider, and results can be volatile. This article offers no betting advice. If gambling is causing harm to you or someone close to you, a local support service can help.