Item Slam
Patch 7.41f
Battle Fury in patch 7.41f: WPA of the purchase on 5 heroes, by role and build slot, against the other items bought in the same slot. Highest WPA, still inconclusive: Juggernaut core, 1st item (+0.14 pp).
WPA of the item against the other items bought in the same slot, in matches of every rank. Slots with fewer than 30 purchases are left out.
| Hero | Role | Slot | Purchases | WPA | 95% CI | Verdict |
|---|---|---|---|---|---|---|
 Juggernaut | Core | 1st item | 14,051 (93.0%) | +0.14 | −0.02 to +0.28 | Inconclusive |
 Anti-Mage | Core | 1st item | 13,806 (96.3%) | −0.01 | −0.08 to +0.07 | Inconclusive |
| Anti-Mage | Core | 2nd item | 213 (1.6%) | −0.19 | −1.61 to +1.25 | Inconclusive |
| Anti-Mage | Support | 1st item | 36 (44.4%) | −0.80 | −3.04 to +1.63 | Inconclusive |
| Juggernaut | Core | 2nd item | 100 (0.7%) | −2.36 | −4.14 to −0.56 | Inconclusive |
 Slark | Core | 1st item | 560 (3.2%) | −5.36 | −8.95 to −2.15 | ✕ Worse |
 Lifestealer | Core | 1st item | 62 (0.4%) | −8.87 | −16.82 to −1.53 | ✕ Worse |
 Ember Spirit | Core | 1st item | 30 (0.2%) | −16.43 | −23.10 to −8.45 | ✕ Worse |
Win probability added: how much choosing an item changed the chance of winning, in percentage points, compared with what the hero usually buys in that same slot and situation. +1.0 means the choice won 1 percentage point more than the typical purchase. Unlike raw win rate, it accounts for the game state when the item was bought, so an item bought from ahead doesn't look good just because the game was already won.
1st, 2nd, 3rd and 4th+ count completed items. Components that become another item in most matches (Perseverance, Dragon Lance, Kaya) do not take a slot: the item they build into does. Boots get a card of their own.
The ✓ Better or ✕ Worse badge only appears when the effect passes the false discovery control (10%) in both ways of measuring: at the start of the decision and at the moment of purchase. Without a badge, the sign of the WPA is a trend.
Ranked (All Pick) matches parsed by OpenDota in the current patch. One model estimates the chance of winning from the game state; another, the effect of each item at the time of the decision. It is association, not causation: the model sees the game, not who is playing.