Win Rate Doesn't Mean Everything

ƒoes·Sep 20, 2015·4 votes·1,064 views

Seriously, I see people picking champions who have high win rates saying, "Oh, this champion has a 56% win rate, they must be good!"

loses

People pick Vayne (or some meta high-winrate pick) and say, "They must be good!" They check champion.gg and take a quick glance at the graphs, and say, "Yeah! Let's play this champion and win!"

bronze 5

Come on, people. Look at the numbers. If a champion has, say, a 50% win rate, and only 2 people played the champion (highly unlikely, but it's an example), chances are, you're not gonna win those next few games. However, if a champion has a 48% win rate over the course of 20,000 players and games analyzed, that champion technically has a higher win rate than a champion with a 54% win rate with 8,000 players and games analyzed.

So, I'm looking at Yasuo's page on champion.gg, and he has a 46.92% win rate on the most frequent skill order, and a 53.91% win rate on the highest win rate skill order. This does not mean that you will do better if you max E first. If you check below the Most Frequent Skill Order graph, it will say "46.92% Win Rate | 8087 Games". If you check below the Highest Win % Skill Order graph, it will say "53.91% Win Rate | 1803 Games". The reason the Highest Win % Skill Order graph has, well, a higher win rate, is because less games were recorded, therefore the % is not as precise. You would honestly do better if you choose to choose the Most Frequent Skill Order graph, because you have a higher chance of getting a 46.92% win rate than a 53.91% win rate because the 46.92% win rate is more narrowed down.

TL;DR don't pick the Highest Win Rate over the Most Frequent, because the Highest Win Rate is less precise.

EDIT: Bonus information:

{quoted}

champion.gg samples games in a very strange way; they sample a bunch of games and say, "Oh, this is more frequent.", but then they sample a small amount of games (say, there's 100 games where people max W on Yasuo first, and for some odd reason that has a 60% win rate) and say it has a higher win rate. 60 out of 100 of those games were won where people maxed W on Yasuo first, but that 60 may not be proportional. If you, perhaps, ramp it up to 200 games measured, there's a chance that there will still be a 60% win rate (120 out of 200 games won), but it's not very likely. Typically the rate would lower, but can sometimes go higher.

If there are 10 games recorded where players max W on Yasuo first, and it has a 60% win rate, 6 out of 10 games were won. However, if you go ahead and decide 60% is an incredible number without checking the number of games recorded, you can lose, and then the win rate drops significantly; 6 out of 11 games are won, and now the win rate is 54%.

Basically, my point is, you should usually go with the most frequent because the results are way more narrowed down. Think of the number of games recorded as a "ticking" system. Each game recorded adds an update to the winrate graph. If 2 games are played, and one is lost, and the other is won, you have a 50% winrate, and now the very beginning of the graph is very spiked. However, over the course of 18,000 ticks, the graph begins to even out and becomes an almost perfectly straight line. Nerfs and buffs to the champion can change this number, so you'll get yet another spiked graph until it can even out. This is why, in most cases, when a champion is nerfed or buffed, their winrate can change drastically.

{quoted}

To add further to this point and really bring it all home, whenever you look at a Statistic, the more people are accounted for or the bigger the sample size is, the more reliable the statistic is.

15 Comments

5000000000000000Sep 20, 20153 votes

Also it could be that Yasuo players max E first in a favorable matchups so they can fight more in lane, this giving him higher win rate.

I'm no Yasuo main so I don't know if this is true, but there's gonna be a lot of confounding variables.

ƒoesSep 20, 20151 votes

Some people max E on Yasuo first because he

gets a ton of mobility
stacks the damage from 150 to 50% bonus damage
builds a Trinity Force for the Sheen passive so you can E>Q (Q is treated as a basic attack, therefore procs Sheen passive)

5000000000000000Sep 20, 20151 votes

Ah, maybe then newer players generally max Q to play more safely and experienced players generally max E to play aggressive.

ƒoesSep 20, 20151 votes

I'm an experienced Yasuo player (80.5k points) and know him extremely well, but I still typically max Q because I tend to go with Most Frequent instead of Highest Win % simply because Most Frequent is way more accurate. Also, my runes are more based around using Yasuo's Q and maxing it because I start with 20% crit chance.

5000000000000000Sep 20, 20151 votes

Err, I'm gonna stop trying to guess how maxing E has higher win rate.

DorkunedAurasSep 20, 20151 votes

From what I understand, Every point in Yas's Q gives the Q 20 more damage. I believe that's it, the CD is from AS.

Every point in Yas's E reduces the CD on re-dashing opponents and increases dmg by 20, but it still has the 25-50% dmg ramp up for consecutive dashes, meaning a good yasuo would get 30 dmg per lvl instead of 20. If you play him well, those E points will do more for you than the Q ones. That's just my understanding, I don't play him either.

TheShadyDealerSep 20, 20151 votes

Picking a high winrate champion and actually learning how to play well before getting into ranked isn't so bad. I get tired of seeing people pick Vayne and have no idea how to tumble though lol

ƒoesSep 20, 20152 votes

Well, sure, if you know how to play Vayne or some champion that has a high winrate (I think right now Malpite has the highest win rate), you got lucky.

NoPaxtSep 20, 20151 votes

You're speaking math at people... Numeracy is a lost art, I hope they understand!

ƒoesSep 20, 20151 votes

Heh, it's not even math, it's just showing people numbers and explaining that higher popular/frequent numbers are more accurate.

AwesomeChadSep 21, 20151 votes

To add further to this point and really bring it all home, whenever you look at a Statistic , the more people are accounted for or the bigger the sample size is, the more reliable the statistic is.

Silver FluffenbuSep 21, 20151 votes

So, tell us what increase in accuracy we can expect with an increase in sample size? is it a linear correlation, or is there a drop off at a specific sampling?

ƒoesSep 24, 20151 votes

english please.

champion.gg samples games in a very strange way; they sample a bunch of games and say, "Oh, this is more frequent.", but then they sample a small amount of games (say, there's 100 games where people max W on Yasuo first, and for some odd reason that has a 60% win rate) and say it has a higher win rate. 60 out of 100 of those games were won where people maxed W on Yasuo first, but that 60 may not be proportional. If you, perhaps, ramp it up to 200 games measured, there's a chance that there will still be a 60% win rate (120 out of 200 games won), but it's not very likely. Typically the rate would lower, but can sometimes go higher.

If there are 10 games recorded where players max W on Yasuo first, and it has a 60% win rate, 6 out of 10 games were won. However, if you go ahead and decide 60% is an incredible number without checking the number of games recorded, you can lose, and then the win rate drops significantly; 6 out of 11 games are won, and now the win rate is 54%.

Basically, my point is, you should usually go with the most frequent because the results are way more narrowed down. Think of the number of games recorded as a "ticking" system. Each game recorded adds an update to the winrate graph. If 2 games are played, and one is lost, and the other is won, you have a 50% winrate, and now the very beginning of the graph is very spiked. However, over the course of 18,000 ticks, the graph begins to even out and becomes an almost perfectly straight line. Nerfs and buffs to the champion can change this number, so you'll get yet another spiked graph until it can even out. This is why, in most cases, when a champion is nerfed or buffed, their winrate can change drastically.

I don't know if I answered your question (because quite frankly I didn't really understand it because I'm not a math kind of person), but I hope it helped.

MongooseSep 24, 20151 votes

as a yasuo main im telling you to max e first. after the q cd buffs, there is no reason to want to max q first except for a 20 base dmg gain each level. with e max you get a 20 base damage gain, a cd buff of .1 per level, and a cd buff of the on champion cooldown of 1 sec per level. with e max you can move around much more fluenty and have an easier time farming under turret, and you can dash harass more.

also diamond no backtalk.