ARAM hands you random champions and matchmaking drags every win rate back toward 50% — so "rank" gets ground flat: everyone copies the same build, and you don't pick your champ. Is there still a skill gap? Yes — it doesn't live in what you pick, but in how much you squeeze out of the same champion.
I split every game by average player efficiency into a high-skill half (top 25%) and a low-skill half (bottom 25%), then measured each champion's win-rate gap between them. That gap is the skill-scaling coefficient:
Skill-scaling = WR(high-skill lobbies) − WR(low-skill lobbies)
Positive = rewards skill (stronger in good hands); negative = stomps low elo (farms weak games, exposed against good players). In every row below, the left number is the low-skill-lobby win rate, the right is the high-skill-lobby win rate.
Rewards skill the most
+5.0pp
+4.4pp
+4.2pp
+4.2pp
+4.1pp
+4.2ppFarms low elo the most
−4.5pp
−4.3pp
−4.2pp
−3.8pp
−3.7pp
−3.5ppWhat it says
The ranking tracks common sense: the rewards-skill end is mechanically demanding, high-ceiling carries (Vayne, Smolder, Bel'Veth) plus the carry-dependent Lulu — they only pay off in good hands. The low-elo end is simple frontline bruisers (Sion, Nasus, Wukong): point-and-click value that stomps messy games, but gets exposed once enemies flash and focus-fire.
Sion is the most dramatic case: 61.3% win rate in low-skill lobbies (one of the highest of any champion), falling to 57.0% in high-skill ones — still strong, just with the free-win cushion stripped away.
One honest caveat
The signal is real (patch-stable, statistically significant) but small — at most ±5pp. It tells you whether a champion rewards skill; it cannot be flipped around to read a player's rank, because ARAM outcomes are too noisy and individual scores regress to the middle. Use it to pick a champ or decide what to practice — not as a ladder rating.
Data: matches collected by arammeta · queue 2400 (Mayhem) · patches 16.11–16.13 · lobby skill = players' champ-controlled damage/gold efficiency, top vs bottom 25%, ≥800 games per champion · skill-scaling cross-patch correlation r≈0.5.