6.76 Retrospective: What We Can Learn from Hero Win Rates

6.77 is coming up soon, which sadly means we won’t get to do any 6.76 testing while it is still relevant.  In lieu of this, I thought I’d do a series on what we can learn about these last two patches using some public data from www.dotabuff.com and www.dota-academy.com.  Also the charts for the next phase of item analysis are being unruly, so procrastination ahoy.

These last two patches at a matchmaking level have been pretty much dominated by talk about the win rates of the new addition Centaur Warchief, the partially remade Drow Ranger, and to a lesser extent Jakiro.   So let’s first look at the history of their patch changes and corresponding win rates in 6.75, 6.76, and 6.76c.

6.76 saw the Dota 2 release of Centaur Warrunner, who immediately surged to a 67% win rate on the back of his completely reworked ultimate.  Unsurprisingly, this lead to a new patch revision just four days later which toned various aspects of our Centaur friend and brought him down to a more reasonable 60% win rate.

6.76 also buffed Drow Ranger quite substantially.  With her new global aura she managed to withstand the Centaur stampede and reach a 60% win rate in 6.76 and then maintain that win rate in 6.76c despite receiving some minor nerfs.  For reference, her 6.75 win rate had been 51.21% in light of some sizable buffs she received that past.  My snapshot of 3-day period of 6.74 from DotaBuff has her at a 48% win rate for the previous patch, but it also shows that in my Very High sample her 6.74 win rate dropped to 43%.  This was the 6th lowest win rate in the Very High sample and the third largest drop in comparing the Very High performance to the DotaBuff sample.

For the brief period of 6.75 Jakiro was actually the winningest hero in matchmaking, sporting a 60% win rate with the next closest heroes being the bugged Undying and perennial secret pubstomper Lich both at 55%.  Jakiro had actually been a secretly successful hero himself in public matchmaking — his 6.74 win rate was close to 53% — but the 6.75 rework to Ice Path ended up being a huge buff.  In 6.76 and 6.76c his win rate dropped to 57%, partially due to some light Ice Path nerfs but also thanks in a large part to the sheer gravitational pull of Drow and Centaur.

Now that we’ve established the patch history, let’s move on to the basic principles of what these win rates actually tell us.

First of all, there’s this oft repeated sentiment that a high public win rate is fine because Dota should only be balanced with tournament play in mind.  Even if we assume this to be true, there is a reason why new and reworked heroes are put into the public pool for a patch before they are allowed in Captain’s Mode.  Public matchmaking is the final testing ground for whether a hero should even be allowed into tournament play, and quite frankly, adding any hero with a 67% win is tantamount to removing one ban from the pool.

The follow-up argument is that players should be allowed some time to find counters and that Centaur’s win rate would drop as players adjusted to him.  My first response is that it is much more reasonable to have the players finding counters to a Centaur with a 57% win rate than a 67%.  My second response is why on earth should we expect his win rate to drop as the players become more experienced?  It’s just as likely that the people playing Centaur those first four days were inexperienced and had plenty to improve on.  This is a character with a global ultimate.  Global ultimates require team coordination to use to their full effectiveness.  If anything there was (and still is) plenty of room for players to evolve more creative line-ups for utilizing Stampede to its fullest extent!

Finally there’s the idea that Centaur was fine if you just picked a counter to him.  This argument was ridiculous.  People would put forth Huskar and Necrolyte as counters because they have tools to counter high HP pools, which completely missed the reason why Centaur was (and still is) an insanely potent hero.  Some mildly more astute players suggested that Dark Seer’s Wall could be used to counter Stampede.  I find the basic premise questionable, but the bigger objection is that you’re offering a localized ult with a 100 second cooldown as a counter for a global ult that had a 65 second cooldown.  Even if Wall were a potential counter, the global aspect and ridiculously short cooldown of Stampede would still make it a complete mismatch.

The second argument that comes up is that Dota as a game is defined by insanely powerful heroes and that nerfing heroes just because they are powerful ruins this aspect of the game.  This perspective misses the fact that what generally makes heroes overpowered in Dota isn’t that they have an environment where they are insanely strong; they become overpowered when they can reach this environment too easily.  When heroes like Centaur and Drow can easily reach their ideal phase of the game and stay there indefinitely, they become problematic for the game because they make it extremely less likely to see a variety of other heroes displaying their phases of insane strength.  Looking back to the International 2, the problem with heroes like Anti-mage and Morphling wasn’t the mere potential that they would get to a phase of the game where they would farm out of control.  The actual balance issue was the perception that it was too likely that they would hit this phase and dominate the game.

So really what we want from Dota isn’t actually a balanced game.  If we ever had any certainty that Dota were balanced, we’d honestly have a pretty boring game on our hands.  Instead, what we want is a game where we are constantly uncertain about what strategy is the strongest, and therefore the biggest balance violation is when a character becomes so strong that we no longer feel this uncertainty.

So what levels of public win rate amount to a violation of this uncertainty?  Well, we can safely say that 52-55% isn’t a big deal by itself.  Lich for example is pretty consistently around 54% because he is an easy to play support with a lot of lane control and decent teamfight potential.  These are all qualities that will help a hero win a bit more than their fair share of games in a public matchmaking environment, and there’s absolutely no reason to believe that this advantage persists at tournament level play.  Tidehunter is a similar case, though one that sees significantly more tournament action.  In short, 52%-55% is absolutely acceptable for 1. Easy-to-play Heroes, 2. Strong Laners, 3. Strong Initiation or Teamfight presence.

Carries are a bit sketchier.  I’m a little apprehensive if a farm dependent hero is pushing 55%.  Besides Drow, the two most worrisome cases are Luna and Phantom Lancer, but at the tournament level Luna sees significant but not overwhelming play and Phantom Lancer is virtually non-existent.  Maybe their tournament fortunes have the potential to surge in the near future, but in the meanwhile both are within acceptable levels.  Sven is more problematic based on tournament play and his matchmaking winrate is a reasonable 51%.  Morphling absolutely terrorized the International 2 but his 6.54 win rate was 48%.  One thing to keep in mind is that my (admittedly sample size limited) data suggests that Morphling saw a pretty big win rate increase in the Very High bracket.  It’s possible that carry win rates exclusive to High and Very High might be a better predictor of problem carries than aggregate win rates.

What this leaves us with are two ways to use public matchmaking data to zero in on problematic heroes.  The first is any hero pushing ~58% or higher.  Lycan, Ursa, and Treant were all guilty of this in 6.74, and all received crippling nerfs.  I don’t expect Centaur, Drow, and Jakiro to be nerfed to quite the same extent, but I do expect all three to be toned down further in 6.77.

For what is, quite frankly, the bigger issue for public matchmaking take a look at this chart provided by CyborgMatt.  A high win rate alone is problematic but not urgent, but a high win rate combined with an incredibly high usage rate cannot be allowed to continue.  A rough estimate from DotaBuff’s usage data would put Drow in 6.76c at being in over 33% of all matchmaking games.  This is despite the fact that she isn’t available in the majority of Single and Random Draft games, and completely unavailable in Captain’s Mode games.  That kind of ecological warping is a bad thing for the health of public matchmaking regardless of the tournament scene.

And to look at Drow more in-depth, is her current aura really something that the game benefits from?  Passives are fine, and globals are fine, but the combination provides too much to her team on the mere condition of her continued existence.  On top of that, the effects aren’t even that interesting besides the potential interaction between her aura and Visage/Enigma/Invoker minions.  I think it would be better if the creep portion of Drow’s aura were converted to a slightly stronger but temporary area buff somewhere along the line of Sven’s Warcry.  This would make her a pusher along the lines of Leshrac’s Edict, just with a significantly different flavor.  More importantly, it would also demand that she contribute both her presence and her mana to these pushes and give the ability both a higher skill cap and a deeper level of counter-play.  If one wanted to get really fancy it could give Drow temporary control over any nearby ranged creep and catapult targeting.

In any case, let’s move on to what kind of changes to expect in 6.77

Drow: I’ve already mentioned what I’d like to see from her, but besides the likelihood of her seeing some nerfs I don’t know what to expect.  I hope the aura sees a slight rework, because I feel the current incredibly low skill cap is more problematic than anything else about her or her kit

Centaur: His ult got hit hard in the last patch, but it quite frankly needed it.  I’m inclined to expect future nerfs will be targetted at his base stats and stat growth or at his mana costs, but further Stampede tweaking is possible.

Jakiro: Nowhere near as problematic as the other two, so I expect his changes to be fairly small.  The most likely change is reverting his Ice Path cooldown to something closer to the 16/15/14/13 (from 12/11/10/9) it was before 6.73.  Back then the change was warranted, but the skill is now much improved.  Further mana cost adjustments are also possible, but I’d be surprised to see any other substantial nerfs.

And based off 6.76c tournament showings

Undying: He has another bug fix in the nerf.  I don’t have a feel for how much this played into his recent popularity surge.  He may or may not see some nerfs depending on how much he benefited from the bug.

Sven: At the very least he’ll see his cleave range clipped a bit.  I think the complaints against him are a tiny bit exaggerated, but not enough to leave him intact through 6.77

Magnus: As much as I love him, he’s going to take a hit.  Hopefully it’s not a huge one, but I wouldn’t be surprised if his nerfs were the third most substantial in the patch.

Bounty Hunter: Maybe I overrate him, but I’d expect to see some small adjustments downward, possibly targeted at his track gold rates.

Templar Assassin/Bat Rider: I won’t be surprised if either of these characters receive nerfs, but I’d personally be inclined to leave them alone for now.  I don’t feel that they hurt the tournament ecosystem nearly as much as some of the other common first round bans.  They’re really strong at a certain stretch of the game, but they have the tendency to fall off in a way that makes them overall acceptable.

Dark Seer: I feel he’s at an acceptable point for the time being.  No, no one likes to go up against vacuum, but his tournament win rate this patch has been a modest 46%.

Buffs are much harder to predict, so I’m not even going to try.  I will say though that I expect Treant Protector to see further changes.  While I have made the argument that he’s not as bad as his 41% win rate makes him out to be, I feel that he’s extremely underwhelming as a hero right now.  Living armor on a 15 second cooldown takes way too much map awareness for what feels like too little reward (both strategically and viscerally), and none of his abilities are especially satisfying in any way.  If he doesn’t see anything in 6.77 then I expect to see a broader rework somewhere down the line.

5 Responses to 6.76 Retrospective: What We Can Learn from Hero Win Rates

  1. TC says:

    It’s curious that you don’t see a clear relationship between win-rates and games played. You’d think given how available win-rates are that there would be a strong correlation. I would guess most people don’t choose heroes to win, but rather because they like to play a certain hero.

    • phantasmal says:

      Part of it is as you say. Pudge, for instance, has never had a very high public win rate. People just like him because good hooks are exciting, and those exciting moments outweigh the slight majority of games that just don’t go your way. Part of this is that some heroes have a significantly higher skill cap than others, so they’re definitely capable of outperforming their overall win rate. It’s just that most of the people who think they’re a better than average Pudge actually aren’t.

      Another aspect is that many people don’t pick to win so much as they pick to not personally lose. I feel a huge portion of Windrunner’s extremely high usage rate in my sample is that she’s seen as a very safe offlane hero. People aren’t picking her with much of a thought for whether she complements their lineup; they just pick her so they don’t feed when they get stuck as the suicide solo.

      The final, possibly most important aspect is that the vast majority of players simply aren’t aware of hero win rates. If you told people that Zeus and Necrolyte were some of the most reliably winning heroes in public matchmaking over multiple patches a significant number of them would simply not believe you. Treant Protector had the same win rate of Ursa and Lycan in 6.74, but he did it quietly so his presence went completely ignored by nearly everyone.

      • TC says:

        It would be interesting if we could split the data by skill level and see whether you see a stronger relationship between win rates and hero selection. I would hypothesize that higher skilled players are more willing to select properly to win. Perhaps another way to look at the data is teams and hero selection. Perhaps if nothing else, higher level players (in my mind) would be more willing to seek information to improve their own play.

        Your second reason suggests that there is a coordination problem. And hearing some friends endlessly whine about the game, I would agree with that. They spend too much time complaining about the lack of teamwork without putting in effort to actually coordinate their team. Another interesting way to look at data is win rates by how many lines of chat there is by the team as an instrument to measure coordination.

        I wonder if there is a fascination with farming. Necrolyte and Zeus and some of the of the other intel heroes have the capability to dominate games but not quite as much in their ability to farm. It reminds me that when I played DotA on wc3, we used to have intel heroes take solo mid so that they can hit 6 quickly and apply pressure to side lanes, and now, that’s a rare occurance for me.

      • phantasmal says:

        Using https://docs.google.com/spreadsheet/ccc?key=0AoNi7mtSTYNzdFdyUUJ3X1ppQ3VxVFhVVXc3WEp2Z1E we can take a preliminary look at how usage rates shift in higher level play. There are still (predominantly) sample size issues, but it’s a decent first step and can hopefully be expanded in the future. The best way to use it is to download a personal copy so you can do your own sorting.

        Sorting by Very High Use %, there’s definitely fewer low win % heroes at the top end of usage than there are in the aggregate pool. This largely ties into the fact that Very High players pick carries less often.

        But beyond this, the other trends I pointed out still exist in Very High, or are even magnified there. Three of the highest usage rate characters — Windrunner, Mirana, and Bounty Hunter– are very popular suicide lane solos. Nature’s Prophet, Queen of Pain, Tidehunter, Enigma, and definitely Dark Seer also might see surges of various sizes thanks to their potential viability in the suicide lane. I think though that this can reflect badly on a character like Windrunner because she becomes the default option for a player who doesn’t know what to do to influence the game from the suicide lane. Most people get to Very High by utilizing strategies that rely on winning a lane and out-farming their opponents, and many of these players will struggle to do much when presented with a situation where they have to effect the outcome of the game without an easy source of farm. You could label this as a coordination problem, but I think the percentage of players capable of coming up with a cohesive drafting strategy in Dota is actually really small. As a result, even relatively organized premades will just make a half strategy and then tell one of their weaker players to just go to the suicide lane and don’t die, and as a result Windrunner becomes the path of least resistance.

        Pudge’s use rate doesn’t drop in Very High as his win rate drops, and Invoker’s use rate sky rockets. Very High, as a whole, probably trends to be even more narcissistic on average, so there’s a bias towards characters with the potential for ‘WoDota’ moments. See also: Rubick (though he also was possibly the best support in 6.74).

        The one curious thing is that Very High seems dramatically less likely to pick the most successful heroes even when their success does not fall off within Very High play. Ursa sees one of the 10 largest drops in Very High usage despite maintaining a 57.5% win rate in the bracket. Lycan sees a small drop despite actually being the most successful hero in the bracket with a 61% win rate. I’ve heard it suggested that there’s a weak honor code in Very High against picking Drow and Centaur in the latest patch, and it might just be possible that this is actually a real thing.

  2. xdv says:

    Maybe not really the “honor code” per se, but players who’re in Very High often get there because of their focus on getting better at the game and increasing their mastery, and playing a clearly overpowered and soon to be nerfed hero is counterproductive – if it gets nerfed in some dramatic fashion, all your time playing it will be for nought.

Leave a Reply

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

WordPress.com Logo

You are commenting using your WordPress.com 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 )

Google+ photo

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

Connecting to %s

%d bloggers like this: