Rating Deviation (RD): Definition & Use
Rating Deviation (RD)
Definition
Rating Deviation (RD) is a number used in the Glicko and Glicko-2 rating systems that measures the uncertainty in a player’s rating. A low RD means the system is confident that the player's published rating is close to their true strength; a high RD means the rating is uncertain. Conceptually, RD acts like the “error bar” on a rating.
Informally, you can think of “rating ± 2×RD” as a rough 95% confidence interval for a player’s true strength in that pool (e.g., Blitz, Rapid). For example, a 2000-rated player with RD 40 is likely between about 1920 and 2080; with RD 200, they might be anywhere from 1600 to 2400.
How RD Works
- Starts high, shrinks with games: New or inactive accounts begin with high RD. Each rated game provides information and reduces RD.
- Grows with inactivity: If you stop playing in a pool, your RD slowly increases over time, reflecting growing uncertainty.
- Controls rating volatility: In Glicko-style systems, your own RD acts like a variable K-factor. High RD allows big rating swings; low RD restricts movement to smaller, steadier changes.
- Down-weights uncertain opponents: Results against opponents with high RD are treated as less informative, so they move your rating less than results against equally rated but stable (low-RD) opponents.
Usage in Chess
- Online rating pools: RD is maintained separately per time control (e.g., Bullet, Blitz, Rapid). You might be very stable (low RD) in Blitz but highly uncertain (high RD) in Classical if you rarely play it.
- Pairing and matchmaking: Some platforms prefer pairing players with moderately low RD to improve rating reliability and game quality, especially in early placements.
- Provisional status replacement: Older systems marked “provisional” ratings for new players. RD is the mathematical version of that idea—no special label needed.
- Site differences: Many chess servers use Glicko or Glicko-2 (e.g., Lichess uses Glicko-2; Chess.com uses a Glicko-based system). FIDE Elo does not track RD; it uses fixed K-factors and activity rules instead.
Examples
- New player spike: A new Blitz account starts at 1500 with RD ≈ 300. After winning 5 games in a row against ~1500 opposition, the rating may jump by several hundred points because high RD permits large adjustments.
- Established player stability: A 2000-rated Rapid player with RD ≈ 40 goes 4/5 against ~2000 opponents. Their rating will rise, but modestly—often by only a few to a couple dozen points—because the system already trusts their number.
- Effect of opponent RD: Beating a 2000-rated opponent with RD 200 moves your rating less than beating a 2000-rated opponent with RD 40, since the former’s strength is uncertain and carries less informational weight.
- Inactivity comeback: A 2100-rated Classical player with RD 45 stops playing for months; RD drifts upward (e.g., to 120+). Upon returning, a few games can move their rating dramatically until RD shrinks again.
Strategic and Practical Implications
- Stabilize your number: If you care about a reliable rating, play a cluster of games in that pool to reduce RD. Once RD is low, your rating will move more predictably.
- Expect swings after breaks: Returning from a layoff? Your first results will “count more” because your RD rose during inactivity.
- Separate pools matter: Your Bullet RD does not affect your Rapid RD. If you’re new to a time control, expect provisional-like behavior even if you’re established elsewhere.
- Interpreting opponents: A high-RD opponent’s rating is a rough estimate. Don’t overreact to an upset—systems already account for that uncertainty.
Historical Notes
RD was introduced by statistician Mark Glickman as part of the Glicko (1995) and Glicko-2 (2001) frameworks, designed to fix Elo’s limitations with new or inactive players. Glicko-2 adds a “volatility” parameter (often denoted σ) alongside RD; RD measures uncertainty at a given moment, while volatility measures how erratic a player’s performance tends to be over time. Modern online platforms widely adopted these ideas because they handle churn and varied activity far better than classic Elo.
Common Misconceptions
- “RD is the same as volatility.” Not quite. In Glicko-2, RD (sometimes written φ) is uncertainty at this instant; volatility (σ) is how much your rating tends to wander due to performance variability.
- “High RD means you’re overrated or underrated.” High RD means “we’re not sure yet.” It doesn’t say in which direction the rating will move.
- “RD is global.” RD is per rating pool; your Blitz RD doesn’t influence your Rapid RD.
Related Terms
Interesting Facts
- Many implementations start new accounts with very high RD (often around 300–350), enabling rapid placement near the correct strength within a handful of games.
- Because RD increases with inactivity, long breaks can make early results after returning feel “swingy”—by design.
- RD underpins why early upsets don’t overly distort the pool: results against high-RD opponents are automatically down-weighted.
At-a-Glance Summary
- What it is: A measure of uncertainty in a player’s rating.
- Low RD: Stable rating, small per-game changes.
- High RD: Uncertain rating, large per-game changes.
- Changes over time: Decreases with games played; increases with inactivity.
- Affects updates: Your RD scales how much your rating can move; opponent RD affects how much that result should count.