Rating Deviation (RD) - Glicko uncertainty
Rating Deviation
Definition
Rating Deviation (RD) is a number that measures the uncertainty in a player’s rating. It is a core part of the Glicko and Glicko-2 rating systems, which many online chess platforms use instead of a plain Elo system. A lower RD means the system is more confident that your rating reflects your true playing strength; a higher RD means your rating is less certain (often due to few games played or a long period of inactivity).
In Glicko, RD is expressed in rating points (e.g., 60, 120, 200). In Glicko-2, an internal “phi” is used and converted to rating points (sites usually display the rating-points version so it reads like “RD = 65”).
How it is used in chess
RD influences how much your rating changes after each game:
- High RD (uncertain rating): your rating can swing more after a result, allowing it to “find” your level quickly.
- Low RD (stable rating): your rating changes more slowly, since the system is confident in your current estimate.
- Inactivity: if you don’t play for a while, RD rises to reflect growing uncertainty. After you resume, early results cause bigger rating moves until RD shrinks again.
Practical note: most sites pair players based on rating, not RD. RD mainly affects the size of rating updates, not who you’re paired with.
Strategic and practical significance
- Stabilizing your rating: play regularly to keep RD low; your rating will then move in smaller, steadier increments.
- Returning from a break: expect large swings at first because RD is high. If your strength improved during the break (study, coaching), a high RD helps your rating adjust upward quickly; if you’re rusty, it may drop faster.
- Assessing opponent ratings: an opponent with a high RD may be much stronger or weaker than their listed rating suggests—exercise caution in assumptions.
Historical background
RD was introduced by statistician Mark Glickman in the late 1990s with the Glicko system to address limitations of Elo, which uses a fixed or piecewise K-factor and does not explicitly track uncertainty. Glicko/Glicko-2 became popular for online chess because they adapt to player activity and volatility. Many over-the-board federations (e.g., FIDE) still use Elo with varying K-factors and do not display an RD, but most major chess servers use a Glicko-style approach under the hood.
Examples
- Equal ratings, low RD: Two 1800-rated players both have RD ≈ 50. If White wins (say after a solid Ruy Lopez with 1. e4 e5 2. Nf3 Nc6 3. Bb5), the winner might gain only around 8–12 rating points because both ratings are already considered reliable.
- Equal ratings, one player has high RD: White is 1800 with RD ≈ 170 (new or inactive), Black is 1800 with RD ≈ 50 (established). If White wins, White might gain roughly 25–35 points. If White loses, the rating might drop by a similarly large amount. The system is rapidly “learning” White’s true level.
- Inactivity effect: A 2000-rated player with RD ≈ 60 stops playing for a while; RD rises (platform-dependent). On returning, the first few games may produce swings of 20–40 points each. After a dozen or so games, RD shrinks and the rating stabilizes with smaller changes per result.
Note: The exact numbers depend on platform settings (Glicko vs. Glicko-2, time controls, volatility parameters). The examples above illustrate typical behavior, not exact calculations.
Common misconceptions
- RD is not the same as K-factor: K controls how fast ratings change in Elo; RD is an uncertainty measure that Glicko/Glicko-2 use to compute updates. Some systems may adjust K by experience, but RD is a separate concept.
- RD is not your rating: Two players with 1800 can have very different RDs. The one with higher RD is simply less “known.”
- RD doesn’t usually affect pairings: You’ll still be paired by rating; RD just changes how much your rating moves after the game.
Interesting facts and anecdotes
- Provisional accounts: New players often start with a high RD so the system can converge quickly to a true rating within their first dozen or so games.
- Long breaks: If a titled player disappears for months, RD would climb on Glicko-based sites, warning that their displayed rating may be out-of-date. In classic Elo lists (without RD), long inactive periods didn’t show this uncertainty explicitly.
- Multiple pools: Blitz, rapid, and bullet ratings each have their own RD. You might be very “stable” in blitz (low RD) but highly uncertain in bullet if you rarely play it.