Dota 2: Who to Pick?

By Christopher Benson

No matter what game one plays, there will always be accepted and rejected strategies. A baseball team with no outfielders is simply ridiculous. Similarly, all eSports have guides and accepted notions on how to play with a certain hero, item, lane, or position. This is commonly called the metagame or simply the meta. Whenever there is an update (or patch), the meta shifts towards different heroes and strategies. On September 25, Dota 2 received the patch Rekindling Soul, which strengthened (buffed) some heroes and weakened (nerfed) others, as well made many mechanics changes and an alteration to the map.

One hero, Omniknight, was buffed in this patch. This was seen as ridiculous by pros, as he already had an outrageous win rate, but was, and still is, ignored by professional teams. This led me to wonder if there were any other heroes who won but were not picked. To conduct this analysis I used data for patch 6.82b from DotaBuff, which collects data from players who choose to make their matches public. Data was collected on the evening of October 11, 2014. After normalizing, I created this graph of Win Rate v. Pick Rate:

Screen Shot 2014-10-28 at 2.16.52 PM

This analysis assumes Normality for both win rates and picks, which is mostly correct. While win rate does follow a roughly Normal distribution (mean = 48.37%, standard deviation = 4.68%), picks are right-skewed; in other words, there are a few very popular heroes and then everyone else. The outlier in win rate is Omniknight, the cause of all this trouble.

Screen Shot 2014-10-28 at 2.17.07 PM

Screen Shot 2014-10-28 at 2.17.19 PM


Ideally, all heroes would be competitive and thus have a normalized win rate of zero (making a horizontal line on the first graph), but that is clearly not the case. Oddly, nine of the ten most popular heroes have a win rate above average. The oddball is Invoker, widely considered the most difficult hero to master. Most heroes have four abilities, of which 2 or 3 may be spells that must be cast by the player. Invoker has 3 reagents, 1 “Invoke,” and 10 possible spells, of which only 2 are available at one time. This means that instead of pressing one button to cast a spell, players have to press up to five, as well as remember the proper combination of reagents. No wonder his win rate is only 43.43%!

To answer my original question, there are 27 heroes who are picked below average and win above average (the top-left of the first graph). There are also 17 heroes who are picked above average and win below average (bottom-right). Hopefully someone can benefit from picking these underutilized heroes and bring them to the attention of the pros, or maybe just have a greater chance of stomping pubs.

The list of underutilized heroes with win rates above 50% is as follows, from highest to lowest win rate: Omniknight, Abaddon, Centaur Warrunner, Lich, Warlock, Undying, Medusa, Brewmaster, Venomancer, Crystal Maiden, Slardar, Elder Titan, Jakiro, Treant Protector, Dazzle, Ancient Apparition, Disruptor, Luna, Troll Warlord, and Enigma.

Seven heroes are underutilized and win above average but below 50%: Shadow Shaman, Clockwerk, Phoenix, Chaos Knight, Night Stalker, Terrorblade, and Keeper of the Light.

The list of overused heroes is, from lowest to highest win rates: Tinker, Nature’s Prophet, Invoker, Rubick, Legion Commander, Bounty Hunter, Ember Spirit, Windranger, Lina, Huskar, Doom, Skywrath Mage, Kunkka, Weaver, Shadow Fiend, Bristleback, and Lion.

Leave a Reply