Why Average MMR Is a Bad Team Fairness Metric in Dota

Updated 2026-07-13

Why is average MMR a bad fairness metric for Dota 2 teams?

Average MMR is a bad fairness metric because it collapses five very different players into one number, and two teams can post an identical average MMR while looking nothing alike underneath it. A tight team of five players within a few hundred MMR of each other and a team pairing a smurf-plus-Herald combination with three mid-pack players can average out to the exact same number, but they are not the same team to play against or as. Average MMR bad fairness metric complaints usually trace back to this exact blind spot: the mean hides variance, and variance is often what decides how a game actually feels.

Take two five-player rosters that both average 4000 MMR. Team A is 4200, 4100, 4000, 3900, and 3800 — tight, low-spread, every player within 400 points of the mean. Team B is 6500, 5000, 4200, 3600, and 700 — an Immortal-tier player carrying alongside a Herald-tier player, with an identical average MMR to Team A. On paper both teams are 'the same.' In practice the two rosters play nothing alike: Team B's high-MMR player can single-handedly carry lanes its low-MMR player is losing, while the skill and game-sense gap inside Team B's own five makes coordinating with each other harder than anything Team A deals with.

How does variance hide in the average?

Variance hides in the average because a mean only describes the center of a distribution, not its shape — averaging is a lossy operation, and the information lost is exactly the spread that decides whether a roster is playable as a coordinated five. Two datasets with the same mean can have wildly different spread, and Team A and Team B above are a clean illustration: same center, completely different shape. Anyone judging fairness only by comparing two averages is implicitly assuming both teams have similar internal spread, which is often false in a casual 10-stack pulled from a Discord server with a genuine Herald-to-Immortal range in the group.

This is also why an identical average MMR between two teams tells you almost nothing about role fit. A team that got its 4000 average from five closely matched all-rounders and a team that got the same 4000 average from one dominant core carrying four weaker teammates have different practical strengths even before role history enters the picture.

What should you check instead of the average?

Check total team strength as a sum, not a mean, alongside the internal spread inside each team and role coverage across both sides. Summing rather than averaging doesn't fix the variance problem by itself — two teams can also share an identical total — but pairing that total with role checks and a cap on how wide the internal spread inside any one team gets closes the gap the mean leaves open. A balancer that spreads a smurf-plus-Herald pairing's extremes across both sides, rather than letting a single team absorb an Immortal and a Herald together, avoids the exact scenario in the worked example above.

How does Dota 2 Groups avoid the average-MMR trap?

Dota 2 Groups avoids the average-MMR trap by working with total team strength and a strength-sorted snake draft rather than eyeballing a mean. The snake draft deals ranked players A-B-B-A-A-B-B-A across the two sides, which structurally prevents the kind of clustering that would otherwise pair an Immortal and a Herald together on one team while the other gets five mid-pack players — the draft order itself spreads the extremes before any swap iteration even runs. From there, up to 1000 swap iterations refine the split further, and the resulting balance score reflects actual team-total parity, not a single misleading mean.

Frequently asked questions