What Is ELO Rating? How Competitive Rankings Work

If you've ever played a competitive online game, you've encountered a rating number — a single figure that supposedly captures how good you are. That number is almost certainly based on the ELO rating system, or something derived from it. Originally invented for chess in the 1960s, the ELO system has become the universal language of competitive ranking, used in everything from League of Legends to FIFA rankings to college football. Understanding how it works — the actual mathematics, not just the concept — gives you a deeper appreciation for what your rating means, why it changes the way it does, and how to climb effectively.

The History: Arpad Elo and the Chess Problem

Before the ELO system, chess ratings were a mess. The United States Chess Federation used the Harkness system, which assigned ratings based on average opponent strength and win percentage. The problem was that it was easily distorted: a player could inflate their rating by selectively playing weaker opponents or avoiding strong ones. The ratings didn't reliably predict match outcomes, which is the entire point of a rating system.

Arpad Elo, a physics professor at Marquette University in Milwaukee and an avid chess player, proposed a better approach in 1960. His key insight came from his physics background: treat each player's skill as a normally distributed random variable. A player rated 1600 isn't exactly 1600 — they perform at a level that fluctuates around 1600, sometimes playing like a 1650, sometimes like a 1550. When two players compete, the outcome is a function of both players' current performance levels, which are drawn from their respective distributions.

This statistical framing led to an elegant formula for predicting match outcomes and updating ratings. The USCF adopted it in 1960, and FIDE (the World Chess Federation) followed in 1970. By the 2000s, it had been adapted for nearly every competitive domain imaginable.

The Math: Expected Score and Rating Updates

The ELO system rests on two formulas. The first calculates your expected score against an opponent. The second uses that expected score to update your rating after the game.

Expected Score Formula

Your expected score against an opponent is:

E = 1 / (1 + 10^{(R_opponent - R_you) / 400})

This is a logistic function that outputs a value between 0 and 1. If you and your opponent have equal ratings, E = 0.5 — you're expected to score 0.5 points (in chess, a win = 1, a draw = 0.5, a loss = 0). If you're rated 200 points above your opponent, E ≈ 0.76 — you're expected to win about 76 percent of the time. If you're rated 400 points above, E ≈ 0.91. The function is symmetric: your expected score plus your opponent's expected score always equals 1.

The 400 in the denominator is a scaling constant that was chosen so that a 200-point rating difference corresponds to approximately a 75 percent win probability, which matched empirical data from thousands of chess games. Different implementations sometimes use different scaling constants, but 400 is the standard.

Rating Update Formula

After the game, your new rating is:

R_new = R_old + K × (S - E)

Where S is your actual score (1 for a win, 0.5 for a draw, 0 for a loss), E is your expected score from the formula above, and K is the K-factor. The term (S - E) is the "surprise" — how much the actual outcome differed from the prediction. If you win a game you were expected to win (S = 1, E = 0.8), the surprise is only 0.2, and you gain a modest number of points. If you win a game you were expected to lose (S = 1, E = 0.2), the surprise is 0.8, and you gain significantly more.

A Worked Example

Suppose you're rated 1500 and your opponent is rated 1700. Your expected score is:

E = 1 / (1 + 10^{(1700 - 1500) / 400}) = 1 / (1 + 10^0.5) = 1 / (1 + 3.162) = 1 / 4.162 ≈ 0.24

You're expected to score 0.24 — roughly a 24 percent chance of winning. Now suppose you win. With K = 20:

R_new = 1500 + 20 × (1 - 0.24) = 1500 + 20 × 0.76 = 1500 + 15.2 = 1515

You gain about 15 points. Your opponent, using the same formula but with S = 0 and E = 0.76, loses 15 points, going from 1700 to 1685. Note that the total points in the system are conserved — what one player gains, the other loses. This zero-sum property is a key feature of ELO.

Now suppose instead that the higher-rated player wins as expected. The 1700-rated player gains only 20 × (1 - 0.76) = 4.8 points. Expected victories produce small gains; upsets produce large ones. This is the mechanism by which ELO naturally sorts players toward their true skill level.

The K-Factor: Volatility Control

The K-factor is the most important tuning parameter in any ELO implementation. It controls how responsive your rating is to individual game results. A high K-factor (30-40) means your rating swings dramatically with each game — useful for new players whose rating needs to converge quickly to their true skill. A low K-factor (10-15) means your rating is stable and changes gradually — appropriate for established players whose skill is well-estimated.

FIDE chess uses a tiered K-factor system: K = 40 for players under 2300 who have played fewer than 30 rated games, K = 20 for players under 2400, and K = 10 for players rated 2400 and above. This means a grandmaster's rating barely moves after a single game (±5 points for a typical result), while a new player's rating can jump 30+ points in a single round. Most online gaming platforms use K-factors between 20 and 32, balancing responsiveness with stability.

ELO Beyond Chess: Esports, Sports, and Beyond

The ELO system's elegance — simple formulas, zero-sum updates, self-correcting convergence — has made it the default rating system across competitive domains:

• Online gaming: League of Legends, Dota 2, Overwatch, and most competitive multiplayer games use ELO-based systems (often modified) for matchmaking and ranking. League's "LP" (League Points) system is ultimately driven by a hidden ELO-like MMR (Matchmaking Rating).
• Football (soccer): FIFA uses an ELO-based system to rank national teams, adopted in 2018 after decades of using a different formula. The World Football Elo Ratings, maintained independently by fans, have been tracking this since the 1990s and are widely considered more accurate than FIFA's pre-2018 system.
• Tennis: While the ATP and WTA use points-based ranking systems, ELO-based models (like those maintained by FiveThirtyEight) are better predictors of match outcomes.
• American football and basketball: FiveThirtyEight's NFL and NBA forecasting models are built on ELO ratings that update after every game.
• Online Scrabble, Go, and other board games: Virtually all online competitive board game platforms use ELO or a close derivative.

Rating Inflation and Deflation

A persistent challenge with ELO systems is rating inflation (average ratings drifting upward over time) and deflation (drifting downward). In standard ELO, the total rating points in the system are conserved: every point gained by one player is lost by another. But in practice, points "leak" out of the system when high-rated players retire (taking their points with them) and "enter" when new players start with a base rating of, say, 1200 and then lose points to established players.

In chess, FIDE has battled inflation for decades. The average rating of the top 100 players has increased from about 2600 in the 1970s to over 2700 today, partly because of genuine skill improvements (better training, computer preparation) and partly because of inflationary pressures. Whether Magnus Carlsen's peak rating of 2882 represents genuinely stronger play than Bobby Fischer's peak of 2785 is an active debate — some of the difference is real, some is inflation.

Online gaming platforms handle this differently. Many reset ratings seasonally, which prevents long-term inflation but creates "grind" periods where strong players temporarily have low ratings. Others use placement matches that can set an initial rating far from the default, reducing the distortion caused by new player entry.

Glicko and TrueSkill: Modern Alternatives

While ELO remains dominant, two notable alternatives have gained adoption in specific contexts.

Glicko and Glicko-2

Developed by Mark Glickman at Harvard, the Glicko system adds a second parameter: rating deviation (RD), which measures how uncertain the system is about your true rating. A new player with few games has a high RD (the system isn't sure where they belong), while a veteran with hundreds of games has a low RD (the system is confident). When you haven't played for a while, your RD increases — the system becomes less certain that your old rating is still accurate.

Glicko-2 adds a third parameter, rating volatility, which captures how consistently you perform. A player who alternates between brilliant and terrible games has high volatility, and the system treats their results with more caution. Lichess, one of the largest chess platforms, uses Glicko-2 for all its ratings.

TrueSkill and TrueSkill 2

Developed by Microsoft Research, TrueSkill was designed for Xbox Live and handles scenarios ELO can't: multiplayer games with more than two players, team games where individual contributions are unclear, and games with partial ordering (first through eighth place, rather than win/loss). TrueSkill represents each player's skill as a Gaussian distribution (mean and variance), similar to Glicko, and uses Bayesian inference to update after each game. TrueSkill 2 adds further sophistication, modelling team chemistry and individual performance within teams. Halo, Gears of War, and many Xbox titles use TrueSkill for matchmaking.

What Your ELO Number Actually Means

Your ELO rating is not an absolute measure of skill — it is a relative measure within a specific population. A 1500 rating on one platform might correspond to very different skill levels on another platform, because the player pools are different. What your rating tells you is your expected win rate against other rated players in the same system:

• Against a player rated 100 points below you: ~64% win rate
• Against a player rated 200 points below you: ~76% win rate
• Against a player rated 400 points below you: ~91% win rate
• Against an equally rated player: ~50% win rate

These percentages are the same in every properly calibrated ELO system, regardless of the absolute numbers. A 200-point gap means the same thing whether it's 1000 vs 1200 or 2500 vs 2700.

How Player Benchmark Uses ELO for Ranked Matches

On Player Benchmark, the ELO system powers ranked matchmaking across all competitive games. When you enter a ranked queue, the system matches you with opponents of similar ELO to produce close, competitive games. After each match, both players' ratings update using the standard formula. New accounts start with a provisional rating and a high K-factor so the system quickly converges on your true skill level. After your initial calibration period (typically 10-15 games), the K-factor decreases and your rating stabilises.

The platform displays your ELO on your profile and on game-specific leaderboards, allowing you to track your progress over time and compare yourself to other players. Seasonal resets create fresh competitive cycles where everyone recalibrates, keeping the ladder active and giving players concrete climbing goals.

Strategies for Climbing Your ELO

Knowing how ELO works gives you practical insights for improving your rating:

1. Play consistently. ELO rewards consistent performance over many games, not a single brilliant result. A player who wins 55 percent of their games will steadily climb, even without any single dominant performance.
2. Avoid tilt. After a loss (or several losses), your rating drops and the system matches you against slightly weaker opponents, making it easier to recover. But if you keep playing while frustrated ("tilted"), you play worse than your true skill, and your rating drops below where it should be. Take breaks after losing streaks.
3. Learn from losses to higher-rated opponents. A loss to a much higher-rated player costs you very few points but offers the most learning opportunity. Seek out these games and study what the stronger player did differently.
4. Focus on improvement, not rating. Paradoxically, the fastest way to raise your rating is to stop thinking about your rating and focus on improving your actual skill. Rating follows skill with a slight lag — if you're genuinely improving, the number will catch up.
5. Play enough games. ELO takes time to converge. If you've played fewer than 30 games, your rating may not reflect your true skill yet. Play through the calibration period before drawing conclusions about your level.

Rating Anxiety and Mindset

Rating anxiety — the fear of losing points that prevents you from playing ranked games — is one of the most common psychological barriers in competitive gaming. It affects players at every level, from beginners to professionals. The root cause is treating your rating as a measure of your worth rather than a statistical tool for matchmaking.

Reframing helps. Your ELO is not a grade; it's a prediction. It predicts what percentage of games you'll win against players of various ratings. A rating drop doesn't mean you got worse — it might mean you had a bad day, or that your opponent had a good one, or simply that the natural variance in game outcomes went against you temporarily. Over a large enough sample, your rating will always converge back toward your true skill level. A 50-point drop after a bad session will correct itself in 10-20 games of normal play.

The most helpful mindset shift is to think of each ranked game as a data point in a long-running statistical process, not as a single high-stakes event. Each game matters very little individually; the trend over hundreds of games is what reflects your skill. Embracing this perspective reduces anxiety and, ironically, improves your play by removing the performance pressure that causes suboptimal decisions.

See your ELO in action — play ranked matches across dozens of games at Player Benchmark Ranked Mode and track your competitive progress.