General DiscussionThe advantage numbers in the Best versus / Worst versus section are t...

Appendix:

I just realized, that the match-up specific win rates of Abbadon and Anti-Mage don't even add up to 100% and they definitively should:

35.82 + 62.76 = 98.58

That is too much to be a rounding error.

uhm, i think its not 100 because of uncounted games.

So you want spirit breaker warlock and abaddon in every heroes' "worst versus" list?
and AM with brood and morphling in best versus?
That would make more sense than this? think about it.

You completely misunderstood me. Of course it wouldn't make any sense to have the heroes who perform the best on average on every heroe's "worst versus" list, but that is not what I am suggesting. Let me try to explain again by another example.

Lets say someone picked Abaddon in the opponent's team. Abaddon has go an average win rate of 57.33% (pretty high). However, he doesn't have 57.33% winning chance in all match-ups. In some he performs worse, in some he performs better. Now you want to counter pick Abaddon by forcing him into a match-up where he performs below average. According to the "worst versus" list, the optimal choice would be to pick Anti-Mage, but if you look at the winning chances, you see that you actually would do Abaddon a favor with this pick, because his winning chances would increase from 57.33 (average) to 62.79%. So YOUR winning chances decrease. This is clearly not the intention of a counter pick.

Am I making any sense now?

You make complete sense and I'd like to see what other people have to say about this topic

Maybe that "advantage" thingy is not about winrate but about KDA/game impact instead?

dota has more than 2 heroes in a game. remove topic.

@Menni:
I was wondering quite the same, in fact looking at the "counter pick" list atm doesnt make much sense, as you suggested you only risk to make it worse

but i am also not sure if there is a specific way to adress this

no it's not
you just have to know what the numbers mean
am's win rate is 41%
that means that his win rate against other heroes, which is on average slightly >50%, is 41%
now when you compare him to a hero like abaddon who has a 57% win rate you expect am to win a lot less than 41%
what the advantage % thing means is how much better am performs than expected just based on both win rates

There are lot of different ways you might calculate the advantage based of different expectations of what a 57% winrate hero vs a 41% winrate hero "should" be.

You might say well .57x.59=33.6% of the games Abaddon is expected to win and AM is expected to lose. You might say .43x.41=17.6% of the games Abaddon is expected to lose and AM is expected to win. But really these numbers make as much sense as saying .57x.41=23.4% of games both win and .43x.59=31.3% of games both lose.

Mathematical models of expectations are always an opinion represented in numbers. You have to decide if this model generally makes sense or does not for yourself...just like you do with statements made with words.

Whatever calculation model you use its going to be an opinion represented in numbers. Abaddon won at least 62.79% of games (probably a bit more because some were abandons). Is that as many as expected? Dotabuff says no. They thought he should have won 2.99% more. Antimage won at least 35.79% of games. Is that what he was expected to win? Dotabuff though he would win 1.41% fewer.

The advantage measures how far off their model of expectations for the matchup was from what really happened (minus the confusion of no stats recorded games). They did not attempt to model any special hero vs hero advantages so they conclude that the difference between their prediction and reality shows that advantage.