Stop Abusing Overbuff

General Discussion
1

Introduction

Some - indeed very “intelligent” - guys in Reddit and here tried to translate Overbuff data into comprehensive pick rates. Thus, they concluded that the best way to do so, was to search in google or ask in Quora. After deep research, they came up with a formula to calculate those pick rates: Overbuff pick rate multiplied by 6.

Why do they multiply the result by 6?

The explanation is simple. Addition represents the argument “or” in probability theory. Therefore, that formula represents the following probability:

Player 1 chooses "X" hero

or
Player 2 chooses “X” hero
or
Player 3 chooses “X” hero
.
.
.
Player 6 chooses “X” hero

In short this formula represents the probability for Team A to have hero “X” in its roster.

The fallacy

However, this glorious and insightful conclusion assumes that those probabilities are independent and mutually exclusive. Before we proceed to this problem let me explain to you what this assumption means in Overwatch:
It states that every single player chooses independently of what the rest of the team needs or – even more precisely – these players will pick whatever they like no matter what!

However, the picks are dependent to each other. You are never going to pick a fourth support, or a bastion without a shield on attack. Most of the times the picks are dictated by insta-locks, which do not even have a statistically observable pattern.

In short the formula Prob(Hero)*(# players) does not make any sense.

Example

Let’s say, you just joined a competitive game; you are a DPS main, but you can also play Roadhog if needed. Two of your teammates insta lock DPS and then other two pick Roadhog and DVa. The team requests to play support, but you know that you are not good, so DVa changes to Zenyatta to help you out. However, you explain that you can only play Roadhog from the Tank category. Thus, he changes to Orisa and you play Roadhog.

  1. In this example the DPS picks are dependent to each other because we do not know if the second insta lock took under consideration the first insta lock.
  2. The tank picks are dependent to the DPS picks and each other.
  3. The two players remaining are dependent to the picks of the first 4.

Results if the events were independent

  1. DVa should have been picked (assuming that he would have picked DVa anyways)
  2. You should have picked a DPS
  3. Zen and Orisa should have not been picked
  4. You should have played with a maximum of 1 support in that game

Funny outcomes that derive from those results.

Heroes Overbuff (all/weekly) Multiply by 6 Multiply by 5
Mercy 8.95% 54% 45%
Brigitte 8.84% 53% 44%
Moira 7.76% 47% 40%
Zenyatta 3.66% 22% 18%
Ana 3.25% 20% 16%
Lucio 2.98% 18% 15%

Given those numbers the joint probability to have any couple of supports in your roster except Brigitte is 82%. Now let’s put Brigitte into the equation. The chance to have any couple of supports in a game is 154%. The jointed probability to have any triple support composition is 54%. However, a problem arises here; the only viable triple support composition includes Brigitte, where the probability is 36%. Thus, according to your conclusion there is a 18% chance to engage triple support composition without Brigitte.

So, in every 6 games, half of them include a triple support composition and in those 3, 1 of them does not include Brigitte.

Does it make sense? Of course not.
Do some of those numbers make sense? Yes, because it is the partial – and misleading – truth.

38 Likes
2

Night, here, is doing the Jeff’s work. Praise be.

2 Likes
3

Insightful. Thanks. Surprised this didn’t get enough attention earlier.

1 Like
4

There is also the huge issue of how and who overbuff gets it’s data from. I have never heard a definitive explanation about how overbuff generates player data. Yet people worship this data and absolute…

2 Likes
5

Basically via the Overwatch API. However to my knowledge you have to log on to Overbuff so they can use your data, which heavily skews the sample.

2 Likes
6

I never understood why people ever rely on Overbuff data for their arguments. I can win 3 games in a row, and Overbuff says I lost 2 of them. I can start a game as one hero on defense, swap to someone different on attack, and win the game, yet Overbuff decreases my winrate on the first hero I used to sucesfully defend.

2 Likes
7

The numbers we can get from Overbuff can be hard to interpret in a vacuum because their absolute meaning isn’t worth much. Very few statistics are worth anything without context. However, Overbuff does give us a way to evaluate those numbers relative to the same numbers for other heroes.

Even if saying that Hero A has 50% pick rate and Hero B has a 25% pick rate may be completely inaccurate in terms of raw numbers, we can still say that Hero A is picked twice as much as Hero B and be confident about that conclusion.

As for the dependency of hero picks on each other, there are ways we can analyze the Overbuff data for trends over time to determine those dependencies by showing correlation between the varying pick rates of specific pairs of heroes, such as Pharah and Mercy.

Lastly, while the number you get from multiplying Overbuff pick rates by 6 isn’t perfect due to the way Overbuff calculates pick rates (especially where more than one player plays the same hero at different times during the same match), it still does give you an idea of what the chance is per-game that you will see that specific hero on a team. We can’t use that particular information to say anything about how likely it is to get both Hero A AND Hero B on your team at the same time, but we definitely can compare the relative pick rates of Hero A and Hero B.

Edit: as a side note, if you want to calculate the probability of getting a team comp with triple support, you actually need to address the problem a different way than OP tried to demonstrate. First, add up all the pick rates for all the supports: 54 + 53 + 47 + 22 + 20 + 18 = 214. So, for every game in this collection of data, there is a 214% chance of having a support on your team. If there were always 2 supports, that number would be 200%. There also some games where there is only one support to make up for, so this math is also not quite accurate, but looking at that number 214, the claim that there is a 14% chance per game to see a 3-support team composition is still far more accurate than the 54% OP came up with.

I miss stats class.

8 Likes
8

Can we sticky this?

(20 char)

1 Like
9

Are we playing the same game? Most of us do have that person who picks independently of what the rest of the team needs all of the time. Need a second healer? Too bad, that guy wants to go doomfist, so he is playing doomfist.

Your argument is equally fallible if you assume this never happens.

2 Likes
10

People rely on Overbuff stats because it is the best resource on hand to gauge arguments about pickrates and viability by looking at what the overall trends are.

This could easily be remedied if Blizzard posted the statistics they use to make judgements so that we can see a more accurate picture of the situation at hand. Also, this would probably shift forum discussion away from the toxic sides of things such as “Devs dont know balance” etc etc. Because we can see the data teh devs make their decisions on and more accurately understand the thought process theyre using, rather than looking at meta data on Overbuff.

4 Likes
11

While I agree that pick rate comparisons could be useful to interpret hero dynamics qualitatively, I disagree on this statement. Correlation does not indicate causation. Dependencies are the result of an cause and effect relationship.

For example, Zenyatta picks may be positively correlated with main healer picks, because he cannot support a team by his own. However, a main healer doesn’t cause a Zenyatta pick and vice versa.

I would like to emphasize this part. It is the partial truth, therefore it can indeed provide some useful insights if used with caution. This is my primary objection with the majority of forum users. They treat these numbers as they are definitive.

This is not true. By adding those numbers you state the following:
The chance of
player 1 to 6 to pick Brigitte, or
player 1 to 6 to pick Mercy, or
…
…

However this statement is fundamentally problematic. You indirectly assume that a player can pick more than a hero at a time and there is no 1 hero per team restriction. In short, the correct proposition is the following:
what’s the chance of
player 1 to 6 to pick Brigitte
player 2 to 6 to pick Mercy if he cannot pick Brigitte
player 3 to 6 to pick Lucio if he cannot pick Brigitte and Mercy
and so on…

However, we are not provided the conditional (if statement) probabilities, so we cannot really calculate it. We don’t really need to have statistics knowledge to understand that those if statements are going to deflate the probability along with the reduction of player multipliers.

Also, if we want to calculate the probability of getting X supports at the same time we need to use joint probabilities. I do not like to mess with OW numbers right now, so let me explain my rationale with a couple of examples

Dice:

  • The probability to roll your desired number is 1/6.
  • The probability to roll your desired number twice in a row is (1/6)*(1/6) = 1/36.

However, overwatch picks are more like a card game:

  • The probability to draw an Ace of Spades from a deck is 1/52
  • The probability to draw an Ace of Spades and a King of Hearts in the row is (1/52)*(1/51) = 1/2,652.
  • If we did not care about the card order the probability is (2/52)*(1/51) = 1/1,326 i.e. the half from the previous one.

In the case of Overwatch, we care about the order, because we do not pick simultaneously and nobody is proficient with every hero.

Therefore, the best (not correct) way to calculate the probability of getting any set of 2 supports at a time is:
(6*Mercy) * (5*Brigitte) + (6*Mercy) * (5*Ana) + …

If you do that for every support combination (no duplicates), you will find the probability of having any combination of two supports in a team at the same time.

Note: It is not correct because normally you calculate the duplicates and then subtract the conditional probabilities (if statements).

1 Like
12

This.

Overbuff is simply the best source we have ,and it’s a good enough source for forum arguments.

13

I understand and I respect it. However, people tend to abuse it; I do not suggest to stop using it, but people should be more cautious. Two days ago, a guy wrote a reply arguing that Reinhardt is fine because of his pick and win rates.

This is clear abuse of overbuff, because he does not mention that Reinhardt is a mere outcome of Brigitte meta. His pick rate is highly inflated because if player X picks Brigitte, then the probability of picking Reinhardt skyrockets. The probability of picking Reinhardt by its own hasn’t changed from a day to another, since there isn’t any update on this hero.

14

Most arguments I see around here revolve around funz, feelz, anecdotes and hypotheticals. Stats carry little weight for most. Overbuff is just a good reference tool, everyone acknowledges to some degree its inaccuracies.

15

i dont personaly trust overbuffs stats and " cradintels " that every one uses it for. personaly i rather have hard data from the devs them self telling us what the pick rats and things are. jeff or one of the other devs did do this in another thread at one point and it was veary well accepted.

if the devs posted a bi-monthly data dump showing the pick rates and over all %s that overbuff users seem to pleage by then maybe we can finaly find a safer way to slow down some of the more narly threads that pop up on subjects like buffs and nerfs or " devs dont know anythign " threads that keep croping up.

by perviding the data derectly from there archives blizzard can not only help the community help them. but it can allow the player base to go " oh well maybe so and so hero is not being picked as much as over buff says it is "

i dont know. iv never been a fan of any third party tools or sites that " have the best intrist " of the player base in there intrist. sooner or latter someone that runs sites like theses can use them to stake the states in there own faver and use it inorder to make players lean a certen way when it comes to things like buffs and nerfs.

look at the arguments when brig first came to comp due to overbuff stats showing that she had a over 50% winrate and was siting at a huge 50%+ pick rate. it was barly a week in and the forum was flooded with " brigs needs nerfs " threads and " im quiting becuse of brig " threads. now that the stats have settled out theses threads are fewer and fewer. but this is the type thing i dont trust about sites like theses .

if someone that modarates the site hates a hero. the numbes can be fuged inroder to insite the player base to hate the hero as well thus forceing blizzard to buff or nerf a hero that didnt need one (( mercy ? junkrat ? >.> )) but gets smacked with one anyway ?

1 Like
16

By definition the concept of probabilities involves assumptions. Some argue that assumption is the mother of BS. In my opinion, setting wrong assumptions is dangerous (this does not go only for OW). This is not a statistics course to speak about probability theory. The concept is to start a discussion with people who may or may not have the same backgrounds.

17

also some picks are map dependent, or if you are defending or attacking .

an analisis with the most played heroes (sets of 6) in different maps in attack or defence would be the best way to analyse the data

1 Like
18

Which Overwatch API? I’ve been looking at the Blizzard Dev site, there is no such thing for Overwatch.

Blizzard does not provide an official API for Overwatch, contrary to other games like Diablo 3/World of Warcraft. All data is being generated using web scraping technics from the official profiles on Battle dot net.

19

Blizzard will never provide us a full picture of what’s going on with the game. The reason is simple: Marketing.

Every now and then the game should refresh to keep the feeling of flow. In short, sometimes we get changes just for the sake to reshuffle the deck and rekindle the interest of the fans.

If they give us numbers, it would be impossible to justify such changes.

20

They get the data from our profiles (that’s why we need to close overwatch). I don’t know which procedure they use, but the simplest way to implement it is to scrap our profiles. Everything in our profiles (from the link to the stats) are standardized, so it merely needs 1 day to implement an algorithm that picks our stats from there.