Bonus heroes is a controversial feature, especially in ranked games. It is decent as a system that replaced true random heroes in competitive matchmaking, but no one really knows why it was in the game in the first place.
Two free mangoes is one extra regeneration and 250 extra mana on demand at the start of the game. It means two extra Shadow Word uses, almost three extra Arcane Bolts and many more other things. In a meta where laning stage is extremely important, it is easy to see how it can be seen as a very unfair advantage.
Arguing against the fact that it is unfair is futile: extra spells, extra regeneration, potential “free money” that can be spent on other items—all of these things are not accounted for in game’s balance. However unfair does not necessarily mean it is without merits.
We’ve gathered winrate statistics for heroes on their “bonus days” and compared it to their regular performance and some of our findings were quite interesting and we are excited to share them with you.
One of the interesting things we’ve concluded from gathering all these stats is that heroes rarely benefit from their bonus days and even when they do, it is by a very small amount.
Most heroes lose winrate on their bonus day, as a result of many players randoming a hero they are unfamiliar with or because it simply doesn’t fit their team’s lineup. It’s hard to win a game of Dota when the person you’ve saved the last pick for, expecting him to pick a mid core, decides to random and gets Crystal Maiden.
Only about 10% of heroes actually benefit from starting the game with two extra mangoes and most of them don’t gain more than 0.2% winrate. Two mangoes is an advantage, it is unfair, but in a game of Dota, where players make countless choices, the outcome of the match is much less dependant on a hero having 250 extra mana, compared to hero fitting the lineup and being played adequately.
For the most part this winrate increase is both statistically insignificant and probably comes from when the bonus hero is a good fit for the lineup and is being played by someone familiar with hero’s spells and general playstyle.
One of the possible conclusions drawn from this probably goes against community’s preconceptions: bonus mangos don’t really have much impact on the outcome of the game and one would rarely lose because the enemy had this objectively unfair, but nonetheless miniscule advantage. Perhaps potential for higher hero variety in our pubs is worth it?
There is a second side to the coin, however. Players might overestimate how much of an advantage the enemy gets from a couple of extra mangos, but they are certainly on point with their frustration when it comes to their teammates pressing the most unnecessary, tilt-inducing button in the game—”random hero”, or picking something they are completely unfamiliar with just because they are a bonus hero.
As mentioned previously, most heroes lose winrate on their bonus day, but the scale of it was larger than we expected. Seven heroes in the game lose more than 2% winrate, despite starting the game with two extra mangoes. Of these, three are actually at 50%+ winrate across all skill brackets and are game-winning picks when played adequately.
Lycan, Lone Druid and Arc Warden are all rather complicated heroes to play and involve heavy micro to be effective. Their skill floor is already very high and their skill ceiling is probably not reached consistently even in professional matches. These heroes lose ~2.8%, ~2.5% and ~3.0% winrate on their bonus days. They also generally lose more games than they win, even on normal days, sometimes to the extreme.
Three other heroes: Meepo, Broodmother and Visage are all game-winning heroes on normal days. When played by players dedicated to mastering micro, they can be decently effective. They also lose much harder when played by novices or players starting to learn them.
Meepo and Broodmother both lose around 4.8% winrate on their bonus day, but Visage takes the cake as the most experience-dependant hero in the game with a whopping ~5.8% winrate loss.
What we are trying to say is that maybe players really shouldn’t experiment and try out new things in ranked games. There are some very complicated heroes in the game that require months of practice before they can be played consistently on an adequate level and in reality, they are rarely worth the effort if your goal is to gain MMR and climb the ranks. Only learn these heroes in unranked and only if you actually enjoy them. Visage is the exception, with a very high winrate when played properly—something to consider when choosing a difficult “signature” hero to master.
This is also important for newer players. Getting bombarded with information and losing every single game is a very frustrating experience so if you are introducing someone to Dota, definitely don’t suggest the most experience-dependant heroes in the game: Visage, Broodmother, Meepo, Earth Spirit, Arc Warden, Lycan, Lone Druid, Oracle, Batrider, Elder Titan, Enchantress, Dark Willow, Tinker, Treant Protector and Ember Spirit are the 15 most complicated heroes in the game, based on their relative performance. Or they would be, if not for a couple of interesting outliers.
One interesting thing we found is that two of some of the most complicated heroes in the game, at least according to community, are surprisingly absent from this list. There is no denying that both Invoker and Io both require a lot of practice to be played effectively, they have a very high skill floor and skill ceiling, yet, they aren’t on the list.
In case of Invoker the explanation is actually quite simple: the hero is already played so much that his bonus day doesn’t lead to a significant increase in pick rate. The hero is in quarter of all of our games every single day, frequently played by players who aren’t dedicated to mastering all his spells and their timely usage and because of it, it really doesn’t show when there is a sudden influx of unskilled players.
The interesting thing about it is that despite all that, hero actually wins more than 50% of his games in 5k+ bracket. Believe us when we tell you that being a 5k+ player doesn’t automatically mean you can play every hero in the game at an adequate level and Invoker is extremely specific and experience-dependent. With that in mind, 50%+ winrate for an objectively complicated hero that experiences “bonus hero syndrome” every single day is actually very impressive. Hopefully we will see more of him at the International, than throughout the year.
Io is a little different. Perhaps the problem is his already extremely low winrate in pubs—in certain games impact of an individual player truly doesn’t matter and there is a baseline winrate for all heroes, for when they are simply being carried by four other members of their team.
Unlike Lone Druid, who is also an extremely unsuccessful hero in pubs, Io can’t also actively be a detriment to his team by feeding Spirit Bear over and over. Players rarely do things to intentionally ruin their game, so feeding a Spirit Bear unintentionally is very different to bringing your teammate to the enemy fountain with Relocate.
Looking at the whole “bonus hero” system with an open mind allowed us to see some of the benefits it provides. It gives Valve interesting stats to work with and perhaps incorporate in their balance decision making. It slightly diversifies the pub meta and the bonuses it provides to the players are insignificant enough to create a noticeable unfair advantage.
The problem mainly lies in how the community uses this system and we’ve tried our best to explain why randoming heroes is a very bad idea and how certain heroes do require a lot of practice, that won’t necessarily pay off in terms of MMR. We also inadvertently created a list of Dota’s most complicated heroes, potentially helping newer players refine their hero choices during their first several months. Hopefully, all of this was as interesting for you to read, as it was for us to write and maybe we will see less randoms in our pubs if people will understand how much good drafting and proper piloting affects their chances of winning.