In this article I have compared the differences in Win Rate difference of a hero in the Normal skill bracket, with that in the Very High skill bracket. Data is from two fortnights from 6.81b in July and August 2014.
Win Rate Range
The first thing I noticed with my data, was that across the board the entire hero pool had a much smaller win rate range in Very High than in Normal. In the Normal bracket, Earth Spirit came last with a terrible win rate of 35%, while Necrophos topped with a 60% win rate. This meant that the win rates for heroes in Normal ranged from 35% to 60%. In Very High however, almost all the heroes that had low win rates in Normal gained a few percentage points, while all the heroes that had high win rates in Normal lost a few percentage points. The Win Rate Range for Very High was 42% to 57%.
Out of the the 30 heroes with the lowest Normal win rates, 28 of them improved upon their win rate in Very High. For the 30 highest win rate heroes, 23 decreased in win rate.
The Top 10 and Bottom 10 Heroes By Normal Win Rate Reveal An Interesting Trend
Biggest Winners
The heroes who gained the most seem to be an interesting collection of heroes, with a strong emphasis on low win rate heroes. Primarily heroes with a high Skill Floor such as Earth/Ember Spirit.
The Top 20 Gainers, With The Top 10 On The Left
There were some notable exceptions to the Win Rate Range trend however. Pugna, Enigma, Doom, Skywrath Mage and Faceless Void all improved in win rate, despite not being that low to begin with. Enigma for example went from a solid 51.11% win rate in Normal, to a even better 54.63% win rate in Very High. I doubt much better micro has much to do with Enigma, so perhaps a combination of superior jungle efficiency mixed with effective pushing and the constant pressure of a well placed Black Hole.
Could you argue that a low win rate hero improving is because they have a high skill floor, and if a high win rate hero is improving, they have a low skill floor and a high skill ceiling? Which might explain that while Void is strong in Normal due to the hero being easy to play good enough, as people get even better at placing his Chronosphere he shoots up in win rate. Resulting in Void being an entry level hero that has a lot of room to grow, perhaps akin to Blizzard’s Starcraft design philosophy of “easy to learn, hard to master”.
Biggest Losers
There is a plethora of fascinating data with the heroes who lose Win Rate between brackets.
The Top 20 Losers, With The Top 10 On The Left
There’s a heap of Supports that do much worse in Very High. Over half of the 20 heroes are predominately played as a Support, like Disruptor and Crystal Maiden. I’m honestly not sure what to make of this. A common criticism of low tier Dota is a lack of support heroes, but in Very High where they are more common, some just do worse.
We then get to micro heroes like Beastmaster, Enchantress and Meepo, all of whom lose out. It has been hypothesized in the past that Meepo’s drop is a result of Smurfs skewing the Normal Data. But I disagree with this due to the sheer amount of Normal games, putting forth the idea that it’s probably more likely that Very High players are simply much better at playing against a Meepo. Especially with focusing their attacks so one of the clones dies quickly.
Enchantress has a low 43.69% win rate to begin with, and then drops to 42.09% in Very High which makes her the lowest win rate hero in the bracket. Something is seriously wrong with Enchantresses in Very High for this to occur. Are they trying so hard to smoke gank that they get put behind due to ineffective rotations? Are they being too greedy and their Midas rush is costing their team dearly? Enchantress also happens to have an even lower win rate in ranked VH games, falling another raw percentage point. If you play in Very High, I suggest you leave Bambi on the shelf.
But then what about Beastmaster? Why do Micro heroes like Chen and Visage gain in win rate, while Beastmaster falls off? Do low tier players not think about the hawk vision at all, perhaps causing them to constantly be spotted by it and never kill it? That’s just one possible reason I’ll put forth. It makes sense to me that PA, Sniper and Axe fall off in win rate, but BM is not a hero I would have guessed has a lower VH win rate.
Is It Me or You?
To explain data, there are two main ideas: are noobs really bad with some heroes, or are pros are just really good at dealing with them? There are of course other reasons that can explain it, but these are two reasons people are instantly going to think about. Interpreting Data is a very complex beast, especially when we are using data from tens of millions of games played all over the world. It’s all open to interpretation, and is very susceptible to misinterpretation as a result of a preconceived notions.
On another note, Dotametrics did a similar article two years ago. There are some similarities with the results, like Axe and Sniper being worse in Very High. But there’s also a lot of differences, Beastmaster improved on win rate in his test back then. Whereas now it’s the complete opposite.
Overall, this is definitely a very thought provoking data set.
____
For non-noob Dota 2 tips and Mechanics, follow @ZotaTips on Twitter.
Zota 
Since you identified a clear trend towards 50% win rate, reasonably attributed to increased knowledge of hero strengths and weaknesses, would like to see the results adjusted for this. Ie say heroes on average are expected to move 25% of their difference towards 50-50, who are the biggest winners or losers relatively?
I think supports falling off as you go higher is easily explained: In a game of 5 carries versus 4 carries and a support – a matchup that is common in lower brackets and unheard off in higher – the team with a support will win more often than the team with none. That alone ought to bias win rates for support heroes a fair bit in the lower brackets.
Also, supports can usually do well even without farm, which helps at a skill level where nobody is good at last-hitting.