Elo vs Glicko-2

Glicko-2 is the rating system behind Lichess, Chess.com's ratings and a good deal of online matchmaking. It is usually described as "Elo, but smarter about uncertainty". This is what that actually means, and when the difference is worth anything to you.

What Elo does

Every player is one number. After a match it moves up or down, by an amount set by the rating gap and the K-factor. That is the entire state of the system: one number per player.

The simplicity is the feature. Anyone can check a rating change by hand, and there is nothing to tune beyond K. The longer explanation is here.

What Glicko-2 adds

Glicko-2, developed by Mark Glickman, tracks two more numbers alongside the rating:

  • Rating deviation (RD) — how confident the system is in that rating. A new player has a high RD; someone who plays weekly has a low one.
  • Volatility — how erratic that player's results have been. A player who beats grandmasters one week and beginners the next carries high volatility.

Both feed back into how far a rating moves. A high-RD player's rating swings quickly, because the system does not trust it yet. A low-RD player's barely moves, because it probably already knows. And crucially, RD grows back over time when you do not play — so a rating that has sat untouched for a year is treated as the stale guess it is.

That last part is the real difference. Elo has no concept of time at all. A rating from 2019 and a rating from last Tuesday are the same kind of fact to it.

Side by side

Elo Glicko-2
State per player One number Rating, deviation, volatility
Confidence in a rating Not tracked Tracked explicitly (RD)
Effect of not playing None Rating becomes less certain over time
New players Settle slowly, or need a separate high-K rule Settle fast by design, then stabilise
Updated After each match Classically over a rating period of several matches
Checkable by hand Yes Not realistically

Which matters for your group

Glicko-2's advantages are all advantages at scale. In a pool of a hundred thousand players with wildly different activity levels, where you are constantly matching strangers against each other, knowing how much to trust each number is worth a lot. That is the problem it was designed for, and it solves it well.

A group of eight friends is close to the opposite situation. Everyone plays everyone, the pool is tiny, and activity is roughly comparable. There, Glicko-2's extra machinery buys you very little: the ratings settle just as accurately, and you have given up the one thing casual players actually ask for — being able to see why a number changed.

There is also an honest point in Glicko-2's favour for uneven groups: if half your players turn up twice a year, Elo will happily present their stale ratings next to current ones with no warning. Glicko-2 would at least know it was unsure. Whether that is worth the complexity depends on how much you care about the ranking being defensible rather than fun.

We chose Elo for Custom Elo Games because the groups using it are small, the pool is closed, and being able to check a rating change yourself matters more here than statistical precision at the third decimal place. You can do exactly that on the Elo calculator.

Elo, for a group your size

A small closed pool is where Elo is at its best. Create a Game, add the players, and check any rating change yourself.

Create a Game Free