How Is Elo Calculated

Calculate the new ELO ratings for two players after a match using the standard ELO system. Understand how skill levels change based on game outcomes.

Comprehensive Guide: How Is ELO Calculated?

The ELO rating system is a method for calculating the relative skill levels of players in competitor-versus-competitor games. Originally developed by Hungarian-American physics professor Arpad Elo in the 1960s for chess, it has since been adapted for numerous competitive games, sports, and even video game matchmaking systems.

Core Principles of the ELO System

The ELO system operates on several fundamental principles:

• Relative Skill Measurement: ELO doesn’t measure absolute skill but rather the relative skill difference between players.
• Zero-Sum Game: The total points in a match remain constant (excluding K-factor adjustments). What one player gains, the other loses.
• Probability-Based: The expected outcome is calculated using probability theory based on current ratings.
• Dynamic Adjustment: Ratings adjust after each game based on the actual result versus the expected result.

The ELO Formula Explained

The standard ELO formula for calculating new ratings after a game involves several steps:

1. Calculate Expected Scores:

   The expected score (E) for each player is calculated using their current ratings. For Player A against Player B:

   E_A = 1 / (1 + 10^{(R_B – R_A)/400})
   E_B = 1 / (1 + 10^{(R_A – R_B)/400})

   Where R_A and R_B are the current ratings of Player A and Player B respectively.

2. Determine Actual Scores:

   The actual score (S) is assigned based on the game outcome:

   • Win = 1 point
   • Loss = 0 points
   • Draw = 0.5 points
3. Calculate New Ratings:

   The new rating is calculated using the formula:

   R’_A = R_A + K × (S_A – E_A)
   R’_B = R_B + K × (S_B – E_B)

   Where K is the K-factor (development coefficient) that determines how much a player’s rating can change in a single game.

The K-Factor: Development Coefficient

The K-factor is a critical component that determines the maximum possible adjustment in a player’s rating from a single game. Different organizations use different K-factors:

Player Type	Typical K-Factor	Description
Beginners	40	Allows for rapid rating adjustment as the system learns the player’s true skill level
Intermediate Players	20-30	Moderate adjustment speed for players with some established history
Masters/Established Players	10	Minimal adjustment for high-level players with stable ratings
Chess Grandmasters (FIDE)	10 (or 0 for top players)	Very conservative adjustments for elite players

The K-factor can also vary based on:

• Game Importance: Higher K-factors for championship matches
• Player Activity: Inactive players may have higher K-factors when returning
• Rating Brackets: Different K-factors for different rating ranges

Practical Example Calculation

Let’s walk through a concrete example to illustrate how ELO calculations work in practice:

Scenario: Player A (rating 1600) plays against Player B (rating 1500). Player A wins the match. We’ll use a K-factor of 32 (common in many implementations).

1. Calculate Expected Scores:

   E_A = 1 / (1 + 10^{(1500-1600)/400}) = 1 / (1 + 10^-0.25) ≈ 0.65

   E_B = 1 / (1 + 10^{(1600-1500)/400}) = 1 / (1 + 10^0.25) ≈ 0.35

2. Determine Actual Scores:

   S_A = 1 (Player A won)

   S_B = 0 (Player B lost)

3. Calculate New Ratings:

   R’_A = 1600 + 32 × (1 – 0.65) = 1600 + 32 × 0.35 = 1600 + 11.2 ≈ 1611

   R’_B = 1500 + 32 × (0 – 0.35) = 1500 – 11.2 ≈ 1489

After this match, Player A’s rating increases from 1600 to 1611, while Player B’s rating decreases from 1500 to 1489.

ELO in Different Competitive Environments

The ELO system has been adapted for various competitive scenarios beyond chess:

Application	Modifications	Example K-Factors
Chess (FIDE)	Different K-factors for different rating levels, title norms	10-40
Video Games (League of Legends)	Team-based adjustments, uncertainty factor for new players	Variable (dynamic)
American Football (NFL)	Team ratings, margin of victory considerations	20-30
Esports (Dota 2)	Solo vs. party MMR, role-specific ratings	25-35
Academic (Peer Review)	Reviewer reliability scoring	5-15

Common Misconceptions About ELO

Despite its widespread use, several misconceptions about the ELO system persist:

1. “ELO measures absolute skill”:

   ELO only measures relative skill. A 2000-rated player is better than a 1500-rated player, but we can’t say exactly “how good” 2000 is in absolute terms.

2. “Higher K-factor is always better”:

   While higher K-factors lead to faster rating convergence, they also introduce more volatility. The optimal K-factor depends on the competitive environment.

3. “ELO accounts for all factors”:

   Standard ELO doesn’t consider:

   • Player fatigue or psychological factors
   • Home-field advantage in sports
   • Team chemistry in team games
   • Recent performance trends
4. “ELO ratings are permanent”:

   Ratings can decay over time if a player becomes inactive, as their true skill may change without competitive play.

Advanced ELO Variations

Several advanced systems have been developed to address limitations of the basic ELO system:

• Glicko Rating System:

  Introduces a ratings deviation (RD) that measures the reliability of a player’s rating. Players with higher RD (less certain ratings) can gain/lose more points.

• Trueskill (Microsoft):

  Uses Bayesian inference to model uncertainty. Particularly effective for team games where individual contributions are hard to measure.

• Elo-MMR Hybrids:

  Many modern games combine ELO with Matchmaking Rating (MMR) systems that consider additional factors like:

  • Win/loss streaks
  • Performance metrics (KDA in MOBAs, accuracy in FPS)
  • Role preference and effectiveness
• Dynamic K-Factor Systems:

  Some implementations adjust the K-factor based on:

  • Number of games played (decreasing K as players establish history)
  • Time since last game (higher K for returning players)
  • Rating difference between opponents

Mathematical Properties of ELO

The ELO system has several interesting mathematical properties that contribute to its effectiveness:

1. Zero-Sum Property:

   In a two-player game, the total rating points remain constant (excluding K-factor effects). What one player gains, the other loses.

2. Logistic Distribution:

   The expected score formula uses a logistic function, which naturally models the probability of winning based on rating differences.

3. Rating Difference Interpretation:

   The difference between two ratings can be interpreted probabilistically:

   • +100 points: ~64% chance of winning
   • +200 points: ~76% chance of winning
   • +400 points: ~92% chance of winning
   • +800 points: ~99% chance of winning
4. Convergence Property:

   With sufficient games, players’ ratings will converge to their “true” skill levels, assuming the K-factor is appropriately set.

Implementing ELO in Real-World Systems

When implementing an ELO-based rating system, consider these practical aspects:

1. Initial Rating Assignment:

   New players typically start with:

   • A default rating (e.g., 1200 in chess, 1000 in some video games)
   • A provisional status with higher K-factors until they’ve played enough games
2. Rating Inflation/Deflation:

   Monitor for:

   • Inflation: Average ratings increase over time (common if new players start too low)
   • Deflation: Average ratings decrease over time (common if new players start too high)

   Solutions include:

   • Periodic rating resets or adjustments
   • Dynamic new player starting ratings
   • Bonus points for new players
3. Cheating Prevention:

   Implement safeguards against:

   • Rating manipulation (throwing games)
   • Multi-accounting (smurfing)
   • Boosting (high-rated players carrying low-rated players)

   Common techniques:

   • Uncertainty measurements (like Glicko’s RD)
   • Performance-based bonuses/penalties
   • Manual review for suspicious activity
4. Visualization and Transparency:

   Help players understand their ratings with:

   • Rating history graphs
   • Expected vs. actual performance comparisons
   • Explanations of rating changes

ELO in Esports and Competitive Gaming

The ELO system has become fundamental to modern esports ecosystems. Here’s how it’s typically implemented in competitive gaming:

• Matchmaking:

  Players are matched based on similar ELO ratings to ensure balanced games. Most systems aim for:

  • ±100 rating difference for “fair” matches
  • ±200 rating difference as the maximum for reasonable matches
• Ranked Ladders:

  Many games use ELO to determine league placements:

  Game	Rating Ranges	League Names
  League of Legends	<1100: Iron 1100-1300: Bronze 1300-1500: Silver 1500-1700: Gold 1700-1900: Platinum 1900-2100: Diamond 2100+: Master/Challenger	Iron → Challenger
  Dota 2	<2000: Herald 2000-2700: Guardian 2700-3400: Crusader 3400-4100: Archon 4100-4800: Legend 4800-5500: Ancient 5500+: Divine/Immortal	Herald → Immortal
  Counter-Strike: GO	<1000: Silver 1000-1500: Nova 1500-2000: Master Guardian 2000-2500: Distinguished Master 2500+: Legendary Eagle/Global Elite	Silver → Global Elite

• Team ELO Calculations:

  For team games, systems often use:

  • Average Team ELO: Simple average of all team members’ ratings
  • Weighted ELO: More weight given to top performers
  • Role-Specific ELO: Separate ratings for different roles/positions
• Decay Systems:

  Many games implement rating decay for inactivity:

  • League of Legends: -35 LP per day after 28 days of inactivity at Diamond+
  • Dota 2: -30 MMR per week after 30 days of inactivity at Ancient+
  • Rocket League: Soft reset each season with partial decay

The Psychology of ELO Systems

ELO systems don’t just measure skill—they also influence player behavior and psychology:

• Motivation and Engagement:

  Visible rating progression provides:

  • Clear goals for improvement
  • Sense of achievement when climbing
  • Feedback on skill development
• Frustration and Tilt:

  Negative aspects can include:

  • “ELO Hell” perception (feeling stuck at a certain rating)
  • Anxiety about rating loss
  • Overemphasis on results over learning
• Social Dynamics:

  Ratings create social hierarchies that can:

  • Encourage mentorship (higher-rated players helping lower-rated ones)
  • Create elitism or toxicity in high-rated communities
  • Influence in-game communication patterns
• Self-Fulfilling Prophecies:

  Players may:

  • Perform better when expecting to win (high ELO advantage)
  • Underperform when expecting to lose (low ELO underdog)
  • Develop fixed mindsets about their “true” rating

Criticisms and Limitations of ELO

While widely used, the ELO system has several limitations that have led to alternative approaches:

1. Assumption of Normal Distribution:

   ELO assumes player skills follow a normal distribution, which isn’t always true in real populations.

2. No Margin of Victory:

   Standard ELO only considers win/loss, not how decisively a game was won.

3. Team Game Limitations:

   ELO struggles with:

   • Measuring individual performance in team contexts
   • Accounting for team composition and synergy
   • Handling dynamic team sizes
4. New Player Problem:

   Initial ratings for new players are arbitrary and can lead to:

   • Unfair early matches
   • Rapid rating inflation/deflation
   • Frustration for new players
5. Dynamic Skill Changes:

   ELO assumes relatively stable skill levels, but players:

   • Improve with practice
   • Decline with inactivity
   • Have good/bad days

Alternatives and Extensions to ELO

Several systems have been developed to address ELO’s limitations:

System	Key Improvements	Best For	Complexity
Glicko	Adds ratings deviation (uncertainty measurement)	Games with intermittent play, new player integration	Moderate
Trueskill	Bayesian approach with uncertainty, handles teams well	Team games, games with partial information	High
Elo-MMR Hybrid	Combines ELO with performance metrics	Games with detailed performance data	Moderate
Dynamic K-Factor	K-factor adjusts based on game context	Games with varying match importance	Low
Whole-History Rating	Considers all past games, not just current rating	Sports with long seasons, historical analysis	High

Implementing Your Own ELO System

If you’re developing a game or competitive system, here’s how to implement a basic ELO system:

1. Choose Initial Ratings:

   Common starting points:

   • 1200 (chess standard)
   • 1000 (some video games)
   • 1500 (higher starting point)
2. Set K-Factor Rules:

   Determine how K-factors will work:

   • Fixed K-factor for all players
   • Variable K-factors based on:
   • Player experience (games played)
   • Rating bracket
   • Time since last game
3. Implement the Core Formula:

   Use the standard ELO update formula for each game result.

4. Add Safeguards:

   Protect against:

   • Rating manipulation
   • Extreme rating swings
   • New player exploitation
5. Create Visualizations:

   Help players understand their progress with:

   • Rating history graphs
   • Win/loss streaks
   • Expected vs. actual performance
6. Monitor and Adjust:

   Regularly check for:

   • Rating inflation/deflation
   • Match quality (are games balanced?)
   • Player satisfaction with the system

For most implementations, starting with the basic ELO formula and then adding modifications as needed is the best approach. The simplicity of ELO makes it easy to implement and explain to users, while still providing effective skill matching.

ELO in Non-Gaming Applications

The ELO system has found applications beyond games and sports:

• Search Engines:

  Some search algorithms use ELO-like systems to:

  • Rank web pages based on “contests” between pages for search terms
  • Determine the most relevant results through iterative comparisons
• Recommendation Systems:

  Used to:

  • Rank products based on user preferences
  • Create personalized recommendations through pairwise comparisons
• Academic Peer Review:

  Some journals use ELO to:

  • Rate reviewer reliability based on agreement with final decisions
  • Match papers to appropriate reviewers
• Financial Markets:

  Applied to:

  • Rank financial analysts based on prediction accuracy
  • Evaluate trading strategies through head-to-head comparisons
• Cybersecurity:

  Used for:

  • Ranking threats based on detection rates
  • Evaluating security system effectiveness

The Future of Rating Systems

As competitive gaming and data science advance, rating systems are evolving:

• Machine Learning Augmentation:

  Future systems may incorporate:

  • Neural networks to detect performance patterns
  • Natural language processing for post-game analysis
  • Computer vision for physical sports analysis
• Real-Time Adjustments:

  Systems may:

  • Adjust ratings during matches based on in-game events
  • Incorporate biometric data (heart rate, reaction times)
• Cross-Game Ratings:

  Potential for:

  • Universal gaming skill ratings across multiple games
  • Transferable ratings between similar game genres
• Behavioral Factors:

  Future systems might consider:

  • Sportsmanship metrics
  • Teamwork and communication skills
  • Adaptability and learning speed
• Blockchain Applications:

  Potential for:

  • Decentralized rating systems
  • Tamper-proof rating histories
  • Tokenized achievement systems

While ELO remains foundational, these advancements may lead to more sophisticated, nuanced rating systems that capture a broader range of competitive aspects.

Authoritative Resources on ELO Calculation

United States Chess Federation – Official ELO Rating System Rules FIDE (World Chess Federation) – International Rating Regulations Cornell University – “A Bayesian Approach to Skill Rating” (Trueskill paper)