League of Legends · Ranked Solo Queue
For every match we measured the LP gap between the two players in each lane, then asked how strongly that gap predicts the win. A steep response means the role is high-impact — a small Elo edge already swings the result. A flat response means the role barely moves the needle. Across 38,543 ranked games spanning Iron to Challenger in 9 regions, two patterns dominate: a lane Elo lead is worth far more in low elo than at the top — its impact roughly halves by apex — and role ranking shifts with rank. Mid moves the needle most below apex, Jungle converts best at apex, and Bot's reputation for "low impact" is really an apex effect — in Gold–Platinum it's the strongest lane.
Win-probability response to a lane Elo lead, in log-odds per +100 LP. Higher = the role's Elo advantage matters more. Univariate reads each lane on its own (the direct "does an edge here win games?"). Multivariate holds the other four lanes equal, isolating each lane's own contribution. These pool all four elo bands together — the next section shows how the ranking shifts with rank.
Blue-side win rate as the lane's LP delta moves from "behind" to "ahead". This is the definition made visual: a steep climb = high impact; a flat line = low impact. The dashed line is 50%. Bot lane climbs the most gently; the other four rise more steeply.
The same per-lane impact (log-odds per +100 LP), split into four elo bands — now 6,000–12,000 games each, so every band is decisive (z 7–25). Brighter = higher impact. The dominant pattern: a lead is worth far more in low elo and roughly halves by apex. Role order shifts too — Mid is strong across the lower bands, Jungle tops the chart at apex, and Bot is weak only at high elo (it's the strongest lane in Gold–Platinum). Hover any cell for z + win rate.
No log-odds, no curve — just a straight line divided out. Each cell is how many percentage points of win rate your team gains for every +100 LP your laner is ahead in that role (for a single LP, divide by 100). A blunt, intuitive companion to the log-odds slopes above — same four bands as the legend.
| Role | Uni. slope | z | Multi. slope | z | Low | Mid | High | Apex |
|---|
The same win-rate-per-+100 LP metric, computed separately per region — a replication check. If a role's impact holds across independent ladders, it's a real effect, not a regional quirk. Cells are percentage points of win rate per +100 LP; "Games" shows the Low/Mid/High/Apex split. Switch the band to test replication — e.g. is Bot the weakest at Apex across independent ladders?
The logistic-regression slope of winning on the lane's LP delta. +0.05 per 100 LP means a one-division lead (~400 LP) shifts win odds meaningfully; the "WR@+300 LP" column turns that into a plain win rate.
Bot is a 2v2 lane, so the ADC's individual rank shares influence with the support. Bot is the weakest apex role in most large regions (EUW, NA, EUNE, JP, BR) — but not universally — and in Gold–Platinum the ADC's lead actually converts best. Its "low impact" is an apex/high-elo effect, not a global one.
Four large bands: Low 6,274, Mid 10,529, High 12,069, Apex 9,671 games, pooled across 9 regions. |z| > 2 ≈ significant; here most cells exceed 10.
Riot Match-V5 ranked-solo games; each player's tier/division/LP mapped to a continuous ladder value. Lane delta = blue − red. Baseline blue win rate was 47.5% — the intercept absorbs side bias, so slopes are unaffected.
Rank = current displayed LP (a close proxy for MMR, especially in apex). Built with a stdlib-only Python collector and a hand-rolled logistic regression — no third-party libraries.