Dota 2 Hero Winrates by Game Duration

Today we’re going to look at the relationship between the success rates of different heroes based on the length of the game.  Some heroes perform distinctly better in games that end early, and others prefer games that drag on, and while most of the placements aren’t terribly surprising, it’s still something worth establishing.

My method here is pretty straightforward.  I took the +35k game Very High sample that I’ve been recently, and created a list of game durations.  As it turned out, 1/3 of games were shorter than 33:20, and 1/3 of games were longer than 41:46, so I used those as my break points to define short, medium, and long games (the overall average was 37:50).  I found each heroes win rate in all 3 types of games, and sorted them by how skewed their win rates were towards either end of the spectrum.  To keep things simple, I’ll just be expressing this skew in the tables by the difference in win rate between the short and long games or vice versa.  So let’s kick things off with the top 15 heroes that prefer short games and the top 15 that prefer long games.


The biggest surprises here for me were Nightstalker and Axe both doing significantly better in longer games.  You might also notice that the Long list is very carry oriented while the short list is more support oriented.  Part of this is likely that every support you have on a team is a position that isn’t filled by a carry and will automatically bias your team strategy towards quicker games.  What this means is that your core farmers likely exert the stronger pull on the idea team timing, so if you have both a Chen and Anti-Mage on a team, Anti-Mage’s duration preference probably trumps Chen’s.

So given this tendency, I thought it’d be helpful to point out the supports that skew the most towards late wins and the carries that skew the most towards early wins.


One word of warning I feel I should add here is that while it can make sense to pair a bunch of early oriented heroes together (say, Power Ranger’s lineup with Chaos Knight and Weaver as their carries with Treant as a support), the reverse doesn’t necessarily hold.  If you’re running a late-game oriented lineup you might need some strong early heroes just to make sure that you actually make it to the late game.  This is particularly true when pubbing, and to illustrate why that is, let’s look at the heroes with the highest observed win rate in Very High games.


You can see pretty clearly that this list pretty heavily biased towards early game heroes (though the big exception in Spectre is pretty noteworthy).  Being able to put the game away early is a huge advantage in a disorganized pub environment, so if you’re trying to bump up your win rate make it a point to pick heroes that you’re capable of making a sub-20m impact with.

For those of you who want to see the entire list here’s the table with all the data (Click to enlarge)


As well as the entire dataset for both Very High and Normal in Google Drive.

Finally, have the overall match duration for every hero in both Very High and Normal



15 Responses to Dota 2 Hero Winrates by Game Duration

  1. 342asd23 says:

    use safe lane and offlane lane instead of short/long. short lane = safe lane for many dota players

    • paradigm says:

      Why is there even a confusion at all? Isn’t the offlane always the long lane? I mean you’re furthest from the tower in that lane. And in the safelane your closest. Clarification on why I’m confusing this would be nice.

      • phantasmal says:

        The confusion he’s referring to is between short lane being the lane that is physically shorter (the left and right sides of the rectangle) or the short lane being the lane with the shortest distance between the neutral creep meet and the tower. Using Safe and Hard/Offlane avoids this issue entirely.

        My confusion is that I can’t find any mention of lanes in this article.

  2. Why do you think Axe is so solidly in the late-game group? I play Axe a lot, but I’ve always thought his usefulness burns out after the midgame.

    • phantasmal says:

      3 ideas

      1. It could just be sample noise. Axe isn’t intensely popular in VH.

      2. It could be Berserker’s Call. Heroes with Carry-stopping CC seem to skew late (Batrider being the most extreme example), and Call might fit the bill.

      3. It could be that people still spend too long jungling. These results still represent current popular opinion about the heroes rather than some completely objective evaluation of the hero. One example is Phantom Assassin. If I had enough games and created two sets of PA data, one that built or was building Battlefury and one that was not, it’s possible they’d have two significantly different skews. Axe could have the same effect based on whether he jungles or lanes. Unfortunately, this would be harder to test since you would need to replay parse to determine whether or not any particular Axe was jungling.

      • Boush says:

        In addition to your theories, believe it or not I think Axe excels at the late game due to his Counter Helix. While it falls off immensely against heroes whose armor rises as the game goes on, it becomes more powerful against the creep waves when more of them spawn per wave. Late game, Axe can charge right through huge creep waves clearing them very quickly with Counter Helix by doing nothing but running through them.

        I’ve had games with Axe where I had to do hard supporting early on. The games ended up dragging on for 55-70 minutes and my CS reached 250-350+ by the end without even trying to farm, just from clearing the huge creep waves that form late game.

      • erittainvarma says:

        One chance is also that Axe is mostly picked to keep enemy team busy so that own team late game carry can farm in peace, or at least get better farm than other teams main carry. And if it fails, game is pretty soon over.

  3. CND says:

    Do you have raw parsed data from those games? While it’s informative I’d like to see what it’d look like with more arbitrary time marks since the 1/3 rule assumes there are just as many “short” games as there are “long” games, same with medium length games.

  4. CND says:

    Oh, nvm, didn’t see the other pages in the google doc.

  5. Boush says:

    As expected, Tinker holds the longest average match duration. His ability to drag games on is greater than any other hero due to how hard it is to push through March of the Machines spam. This goes several ways: his MotM spam keeps the lanes pushed to the opposing side, and it is very hard to break high ground against Tinker. He is also not a “team fight” hero; his playstyle is very PvE, often avoiding contact with other players, and I think that further contributes to his games extending longer.

  6. Eggs says:

    I don’t think the spectre result in VH is as surprisingly as you think. Analysing is win-rates in more detail, we can that is fast-game win-rate isnt _that_ bad and he scales extremely well, having a respectable medium-game performance and moving into the highest win-rate in long games of anyone, more even than Anti-Mage and Void. Therefore hes a pretty safe pick in VH where teams understand how to nurse him through the tricky laning stage.

  7. Artoes says:

    Interesting to see lich is such a late skewed hero. Poor pusher, pairing with late game carries perhaps, armor skilled late and being significant buff in late game fights?

    Would be interesting to see if the game duration distribution differs in more controlled play (all pick vs CM / CD) and how the couple of latest patch have affected the general distribution.

    From quickly checking through my latest few dozen games it seems all pick games are longer duration by far compared to CD. Guessing it has something to do with higher quality games in CD, people seem to have quite a bit better knowledge of game which probably translates to pushing the advantage more once acchieved. Pushing and snowbally line-ups are popular in the latest patches also.

    In general would be very nice to have some analyzing of CD picks/bans if you ever find the means, time and interest :)

    Keep up the great work!

    • phantasmal says:

      >In general would be very nice to have some
      >analyzing of CD picks/bans if you ever find the
      >means, time and interest :)

      It’s unfortunately tough to do something like this for CD and CM under my system. I can only get < 10k games a day per bracket, and only a fraction of those are CM/CD. Someone doing a MatchSequence parse would have significantly more CM/CD matches, but the problem there is that they won't have skill bracket information. In my experience, CM/CD skew heavily towards the Very High bracket compared to other game mode so that it'd be impossible to say what is changing because of the mode and what is changing because of the player skill composition.

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s

%d bloggers like this: