There are many facets to a hero that influence the ways it can be played effectively and the impact it can have on a game. Perhaps the most intuitive aspect of each hero is how its role and impact evolves over the course of a match. Moreso than for champions in League of Legends, heroes in Dota rise and fall in efficacy in predictable ways. The most natural progression is for the carry. Weak at the start of the match, time to farm turns a feeble early-game hero into a fearsome late-game monster. Spell based heroes experience the reverse transition; in the early phase of the game, their spells hit hard, regardless of what items they have. Later on, as other heroes pick up items that give them magic immunity and more health, the damage output of spell-based heroes fades. This fundamental transition over the course of the game gives Dota much of its richness. Certain team compositions have phases of the game where they feel useless and phases where they feel dominant. This visualization seeks to explore this effect.
In this visualization, I aim to measure and display the extent to which heroes played in professional matches exhibit these characteristics. It's difficult to judge performance at different stages, so instead we can sort matches into their length and look at short matches (ended in less than 30 minutes), medium matches (30-40 minutes) and long matches (longer than 40 minutes). For each hero, I calculate the win rate for matches in that time period. The assumption here is that in games that end early, teams with effective early-game heroes will tend to win more often. Alternatively, we all know the feeling of a game getting out of control when an enemy carry is given time to get strong.
The bars in each time period represent the absolute number of games won/lost. This is useful to get a sense of how frequently the hero is picked as well as to see if teams that pick carries manage to push games longer, to the point where their carries are more likely to be successful.
It's quite difficult to isolate the impact of one hero on a game that has ten heroes involved. The team compositions in play usually both have early and late-game oriented heroes, and the way players use the heroes has more of an impact than simply the selection of the hero itself. Nevertheless, we can see some trends that make intuitive sense. Batrider, Templar Assassin, and Wisp dominate the early game. Sven, Shadow Fiend, and Faceless Void do well in long games. But there are lots of surprises too. Why is Phantom Lancer so much stronger in mid-length games and has a poor win rate in long games? Why is Shadow Demon so bad in short games?
This data is pulled from professional games only, from the datdota.com database, and filters out heroes without enough games played in each match duration. A future version of this visualization might draw data from non-professional games. If you're interested in helping to pull that together - let me know!
Got comments, questions, or suggestions? Please feel free to email me. I'll update this page as appropriate with clarifications and improvements.