Rating deviation (RD) - chess term
Rating deviation
Definition
Rating deviation (often abbreviated RD) is a numerical measure of uncertainty attached to a player’s rating in systems such as Glicko and Glicko-2. It functions like a statistical standard deviation in “rating points.” A smaller RD means the system is more confident that your listed rating reflects your true playing strength; a larger RD means the system has less information and expects bigger swings as new results arrive.
How it’s used in chess
- Update size: In Glicko-family systems, the size of your rating change after each game depends on both players’ RDs. High RD leads to larger changes; low RD leads to smaller, steadier changes.
- Inactivity: RD increases when you don’t play for a while (the system becomes less certain) and decreases as you play more games with consistent results.
- Provisional status: Many platforms mark accounts with high RD as “provisional.” The rating is considered less reliable until sufficient games are played and RD drops below a threshold.
- Opponent weighting: Results against opponents with very high RD are discounted somewhat, because their own strength is uncertain. Conversely, results against well-established players (low RD) carry clearer information.
- Displayed or hidden: Sites like Lichess (Glicko-2) show a “deviation” number on profiles, while others compute RD behind the scenes to drive the rating algorithm without displaying it prominently.
Strategic and practical significance
- Planning your grind: If you want your displayed rating to stabilize (e.g., before entering a rated event), play a cluster of games to reduce RD first. This limits large swings from a single bad session.
- Returning from a break: After inactivity, expect bigger rating moves for a while. A strong comeback streak after a hiatus can yield rapid rating gains because RD is elevated.
- Interpreting opponents: Two opponents may both be 1500, but 1500±40 (low RD) is more “certain” than 1500±150 (high RD). Upsets versus the latter won’t teach the system as much about your true level.
Examples
- Same rating, different certainty:
- Player A: 1500 rating with RD 45 (established).
- Player B: 1500 rating with RD 180 (provisional/new or recently inactive).
- If both beat 1500-rated, low-RD opponents, Player B’s rating will jump more than Player A’s because the system is learning quickly about B.
- Inactivity effect:
- Player C: 1700, RD 55. After months without games, their RD drifts upward (e.g., to 120+), signaling uncertainty. On returning, a few results will move their rating more sharply until RD falls again.
- Tournament streak with high RD:
- A new account starts around 1500 with RD near the system’s maximum (often ~350). After a 10–15 game streak with consistent results, RD can drop under ~80, and the rating becomes more stable.
History and background
Mark Glickman introduced the Glicko system (1995) and Glicko-2 (1999) to address a limitation in classic Elo ratings: Elo uses a single development coefficient (the K-factor) for all updates, while Glicko attaches uncertainty (RD) to each player. This lets the system adapt—new or inactive players change quickly; stable veterans change slowly. Many online servers (e.g., Lichess with Glicko-2) use RD to manage rating reliability. Over-the-board federations like FIDE still base ratings on Elo, but they tune K-factors and activity rules to mimic some of RD’s benefits.
Common misconceptions
- “RD is a bonus to rating.” Not exactly. RD doesn’t add points; it scales how fast your rating moves given new results.
- “High RD means you’re stronger or weaker.” No—high RD means the system is unsure. Your point estimate (e.g., 1800) may be right, but the confidence interval is wide.
- “Swiss pairings use RD.” Traditional over-the-board pairings use your rating, not RD. RD primarily affects rating calculations and reliability labels online.
Interesting facts
- Typical ranges: In Glicko-style systems, new players may start with RD near ~350; well-established players often sit between ~30 and ~80, depending on activity.
- Dynamic certainty: RD rises automatically with time off, a feature Glickman designed so ratings “forget” outdated information.
- Volatility vs. RD: Glicko-2 adds another parameter (volatility) to track how swingy a player’s results are. RD is certainty; volatility is consistency.
Tips for players
- To stabilize your rating (lower RD): play regularly across a reasonably large sample of games; avoid very long breaks.
- If you want faster rating movement (e.g., you’ve improved a lot): a period of inactivity followed by strong results naturally increases the update size because RD is higher—though this is risky if results are mixed.
- Don’t over-interpret small changes when RD is high; wait until your RD falls before drawing conclusions about your “true” level.
Related terms
Optional visualization
This placeholder can render a personal rating trend if supported by your viewer: