Skip to content
The Chat Archive
llama-4-maverick-17b-128e-instruct· Meta LMArena Human Preference 140K

[code] I have this data across two players pools about online chess rated games. Can you…

  1. User
    Chess.com  Lichess Blitz
    Blitz	   (+/- 75)
               N=2620
    500	       945
    600	       1030
    700	       1115
    800	       1205
    900	       1305
    1000	   1395
    1100	   1485
    1150	   1520
    1200	   1560
    1250	   1600
    1300	   1630
    1350	   1685
    1400	   1720
    1450	   1760
    1500	   1795
    1550	   1825
    1600	   1855
    1650	   1895
    1700	   1925
    1750	   1960
    1800	   2000
    1850	   2025
    1900	   2060
    1950	   2100
    2000	   2120
    2100	   2205
    2200	   2270
    2300	   2330
    2400	   2405
    2500	   2495
    2600	   2585
    2700	   2630
    2800	   2715
    2900	   2830
    3000	   2925
    

    I have this data across two players pools about online chess rated games.

    Can you explain to me (in detail!) why the two converge despite having large gaps for low ratings? I mean the starting rating of the two systems is different, wouldn't that make the points available for each player in the system different and hence have different rating values for the players?

    My point would be, if chess.com starts at 500 and lichess starts at 1500, there would be less points (per player) in chess.com to reach 3000 than in lichess, so it should be more difficult. And so on and so forth.

  2. Assistant

    The data you've provided represents a comparison between the rating systems of Chess.com and Lichess, two popular online chess platforms. The data suggests that despite starting from different initial ratings (500 for Chess.com and presumably around 1500 for Lichess, though the exact starting point isn't directly relevant), the ratings across the two platforms tend to converge as the rating values increase.

    Let's dive into the details of why this convergence occurs, despite the initial differences and the intuitive expectation that the distribution of rating points might differ significantly between the two systems.

    Understanding Rating Systems

    Most online chess platforms use variations of the Elo rating system, a method for calculating the relative skill levels of players in competitive games. The Elo system is based on the principle that the rating of a player should be adjusted based on their performance against other players. A win against a stronger opponent results in a larger rating gain than a win against a weaker opponent, and conversely, a loss to a weaker opponent results in a larger rating loss than a loss to a stronger opponent.

    Key Factors Influencing Convergence

    1. Relative Nature of Elo Ratings: The Elo rating system is relative. It doesn't measure absolute skill but rather compares players against each other. As such, the distribution of ratings is more important than the absolute values. Over time, the distribution of ratings on both platforms adjusts to reflect the relative strengths of the players within their respective pools.

    2. Inflation and Deflation: Rating systems can suffer from inflation (where ratings tend to increase over time due to weaker players being overrated) or deflation (where ratings tend to decrease as stronger players are underrated). The design of the rating system, including factors like the K-factor (which determines how much a player's rating changes after a game), can influence whether a system tends towards inflation or deflation. Both Chess.com and Lichess likely have mechanisms to mitigate extreme inflation or deflation, helping to stabilize the rating distributions.

    3. Convergence through Rating Distribution: As you move towards the higher end of the rating spectrum, the player pool becomes more selective. Top players on both platforms are likely to have played a significant number of games against a wide range of opponents, and their ratings are more reflective of their relative strength compared to others within their pool. The rating differences between the two systems become less pronounced at higher ratings because the top players are more likely to be matched against similarly skilled opponents, regardless of the platform. This matching process tends to normalize the ratings across platforms for the strongest players.

    4. Player Migration and Comparison: Many top players play on both platforms. When a player competes on both Chess.com and Lichess, their performance relative to others on each platform helps to anchor the ratings across the two sites. If a player is rated 2000 on Chess.com and performs similarly on Lichess, their Lichess rating will likely be close to 2000 as well, assuming they've played enough games against a comparable pool of opponents.

    5. Normalization and Anchoring: While the initial rating points are different, the systems tend to normalize around certain anchor points, such as the average rating of a "beginner" or the rating distribution of a large subset of players. For instance, if both platforms have a large number of casual players, the ratings around the median or average player strength will tend to be comparable, even if the absolute numbers differ.

    Addressing Your Point

    Your intuition that the different starting points should result in different rating distributions is valid in theory. However, the key lies in how the rating systems are calibrated and how players are matched against each other. The convergence at higher ratings is largely due to the relative nature of the Elo system and the fact that top players on both platforms are competing against similarly skilled opponents.

    The perceived "difficulty" of reaching a certain rating is more related to the player's performance relative to their peers rather than the absolute rating value. Thus, while the "distance" to travel from the starting rating to 3000 might be different in terms of rating points, the challenge in terms of improving one's chess skill is more consistent across platforms.

    In summary, the convergence of ratings between Chess.com and Lichess at higher rating levels is a result of the relative nature of the Elo rating system, the distribution of player skills, and the mechanisms in place to stabilize the rating systems. Despite different starting points, the systems adjust to reflect the relative strengths of players within their respective pools, leading to a convergence in rating values for top players.

Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.