General DiscussionStatistical Best Hero Combos

The difference is that the 64.57% cancer works at literally every level of pub play and everyone knows it. Same with the 66.86% cancer.

All the winrates with zues/ursa/necro are hold true for the main player base but not quite so much in higher skill games.

That 65% winrate with ursa and spec is kinda rediculous, i assure you that it drops to 50 or lower against good teams.

It would be nice to filter out combos into MMR brackets. im sure you would see something retarded like 20% of games above 5k have OD+omni.

Yeah that would be interesting. Unfortunately there wasn't an option to filter by MMR. I'm sure some combos do drop off at higher levels while others go up. But in general it shows how strong some combos can be.

What would be more interesting is explaining why this occurs...
Ill give it a try even if I may miss a few important points:
-Zeus + Omniknight : great damage output on all 5 enemyes + large heal/repel protection for the main core
-Necrophos + Spectre : don't realy get it, necro ist such a great healer during a fight
-Omniknight + Spectre : very similar to Zeus + Omniknight

-NP + Ember 37.85% : both need a free lane to splitpush lategame, witch leaves 3 people to defend the base
-AM + NP : same as first
-AM + QOP : I don't get it, qop is good vs pushes and team jumps

Really? Zeus spec? Omni ? http://www.dotabuff.com/matches/2277693502

Now you know how to counter zeus spec http://www.dotabuff.com/matches/2277621345

^yes, by playing on high skill where people can't put 2 and 2 together
SeemsGood

Necro + Spec combo have high winrate why?

you guys are stupid, all of those heroes have insanely high solo win rates so no shit they are going to have even higher combo win rates. Your top 3 combos are all top 5 win rate heroes, only one missing is abbadon and abbadon has a really low pick rate so he's probably only missing cus of that since you said it's top 1000 combos.

Same deal for other end, AM is like 45%, qop is 43%, NP is 45%, ember is 44%, there are some lower heroes but they have pick rates under 5%, while AM (and NP) still gets picked almost 15%. If all picks were evenly distributed every hero would get picked like 11% iirc so he's picked more than average.

So that combo stat is really pretty worthless, all it does is confirm in pubs you load up on high win rate heroes you are more likely to win. What would actually be interesting to see would be specific strat combos like visage + drow + venge + enigma, but it's not going to be picked enough or statistically different from their solo win rates probably.

Global combo ftw! Kreygasm

This isn't necessarily because certain heroes combo together, it's because high winrate heroes are coupled together, resulting in a even higher winrate.

The Exp Win% on YASP is incorrect. What they did is average the winrates of the paired heroes.
Individual winrates:
Spec 58.73%
Zeus 57.59%
Assume you pick Spec. On average, you have a 58.73% winrate. Now someone on your team picks Zeus. Your winrate does not drop to the midpoint at 58.16% but goes even higher, because the zeus 57.59% is better than the unspecified hero at 50%.
Let's calculate an expected winrate using Spec + Zeus as an example, with the assumption that the winrates of all unspecified heroes averages out to 50%

Correct me if the following calculation is wrong, but it seems to be very close to the actual YASP data

Divide the winrates by 50, so we can work with a scale of 0 = 0%, 1 = 50%, 2 = 100% winrate
We get Spec 1.1746 and Zeus 1.1518
However we can't just multiple the two (say hypothetically we had 75% winrates, we would get 1.5, if you multiply you end up over 2 which represents over 100% winrate)
Instead, we take the lose-rate ratio, by inverting it about 1 (calculate with 2 - ratio)
This makes our scale from 0 = 100%, 1 = 50%, 2 = 0% winrate, therefore we can multiple our terms and because they are fractions, they will account for diminishing returns with respect to attaining 100% winrate (asymptotically approach 0)
We get Spec 0.8254 and Zeus 0.8482, which multiply to get 0.7001 (rounded)
Now we convert back to a winrate by retracing our steps, i.e. 2 - 0.7 gives us 1.3
Multiply by 50 and we get a 65% expected winrate.
YASP data says 66.73% winrate for Spec + Zeus, so pretty close.

tl;dr
the high/low winrates of hero pairs is mainly due to compound winrates, not the synergy of the pairing itself

Yes that's exactly what I said just without the math part. It's kind of like flipping a coin. You flip it once and the chance of getting heads is 50/50, but in two flips it significantly increases. Combined probabilities.

And it's same deal for losses, it's not like np + am is a horrible combo, it's that both have horrid win rates to begin with so you can't expect them to perform well together.

The coin flip is a horrible reference. It's always 50/50, that never changes. The thing that changes the coin flip probability is the human flipping interaction. 4Head

Is there any 5 man lineup with 85%+ winrate?
My mobile dosnt load yasp

Sniper
Drow
Ursa
Bh
Juggernaut

I don't agree that this a thing of randomness, I don't thing its just and association of 2 high win rate heroes. Since the best winrate duo is zeus + Omni, zues is the 9th individual picked hero and Omni is the is the 31st individual picked hero, so you must be retarded to think that zeus+Omni having the highest win rate is simply people always separately picking those 2 heroes. No, these duos are a combo because they sync well together, I don't understand exactly why myself, but I'm sure that's the case...

@Kobby, pick rate doesn't matter because the win rate is calculated from the games that they are picked in, not from the total number of games, i.e. pick rate is 100%

after some more thought, you actually need 3 functions to model for the different combinations of >50% or <50% heroes, or my knowledge is just not good enough to do it in 1

Maybe we could find more reliable stats out of professional dota datapool?

@Destinii, I agree but that's irrelevant, because considering 2 heroes that are rarely picked individually have a very high win rate together is definitely not a thing of luck or statistical circumstances, at least that seems logical...

Because if winrate would be a thing of luck on pairs of heroes, its only logical that a higher winrate would apear on a high pickrate pair, but since that's not true, than winrate must be the result of the ingame output that those heroes do.

Just had a game with Np and ember on same team. Fuking 3v5 game, actually 2v5 with bh afk invi ..Worst heroes in the team..