Internal data used for balance patches

Dear developers & testers,

In your balance patches you often write notes that refer to internal data gathered by you when explaining a balance change or talents change.

I would appreciate it if you provided an example of how the data are organized of a talent or a hero and how different categories of data are given weight to push into a change decision.

For example if we chose illidan’s “Friend or foe” talent , and I am going to assume that data is collected from all game modes, do you have a system of point allocation depending on the rating of the player ? , Is team’s composition are taken into consideration ? Or is it just simple statistics of pick & win rates over all game modes & ranks ?

Your answer and example will help us to better understand how you evaluate a talent or a hero before changing something & perhaps lower the ranting rates against every balance change and lower the differences in opinions between players a bit about heroes and talents balance wise.

Thank you for your time & effort.

7 Likes

If only they also released internal statistics so we don’t have to rely on 3rd party data collection and replay parsers for 5+ years.

I know they said Hotslogs and heroesprofiles are very accurate, but nothing beats the official source.

8 Likes

Their internal stats looks like this:

11 Likes

They will never show us the real winrates of every hero we got.

1 Like

I’m sure this is the shortest post I’ve ever seen of you.

1 Like

I don’t have a link on Hand, but when they’ve mentioned talent winrates before, they often have included “Diamond+ players, with at least 10 levels on the hero” to paraphrase.

So basically the top ~10-15% of the playerbase who are playing on heroes that they have over 10 levels on seems to be what they publicly share when they seldom do. So probably it is what they use.

There are good reasons not to do this, related to balancing talents.

The same reason why sharing how QM matchmaker works is a very bad idea.

3 Likes

Not bad , I wish for an example with official and more detailed description thought , thank you for the info

I’m not going to give away everything, but I can answer this question.

We get raw data on basically everything that’s happening in the game across all modes based on win/pick rate, as well as things like damage dealt/taken etc.

We are able to parse through this data to look at it through many lenses. This is where raw statistics become less useful and the skill and experience of a designer come into play more so that we can draw meaningful conclusions from the data, and decide what changes we want to make based on the results.

To get a full picture of what’s going on in the game, for example, I can and do look at things like win/loss data that’s filtered by different leagues and game modes. I can also see hero pick and ban popularity with this kind of granularity, which lets me see, for example, what the meta is looking like in the Bronze - Silver range as well as the Diamond - GM range.

While incredibly useful, I want to emphasize that it’s not just the raw data that’s important, but also having the game knowledge and experience to make correct conclusions based on the knowledge. This is something that’s a lot more esoteric, and there are often multiple right and wrong answers, which can and often does lead to a lot of debate within the design team and with the playerbase at large.

As an example, let’s say that you’re the designer and your job is to evaluate Samuro in the next balance patch with these facts (I made these up in my head right now because i’m at home, they are not what’s actually happening though there are truths in these statements):

He’s currently at a 55% win rate

He has a very low pick rate, let’s say he’s the 5th lowest pick hero in the game.

He’s considered a highly skilled but highly rewarding hero by the bulk of the community who argues that he should deserve a relatively high win rate as a reward for being hard to play. They argue that if you nerf or remove what’s unique about him, then why would anyone play a hard hero when they can pick someone like Raynor or Lili every game and just win more often?

Others feel he’s frustrating to play against and a different section of the community thinks he’s obviously way too strong and needs nerfs. They can and do show you videos of crazy things that he does that look and feel unfair.

The people that play him love the unique things he can do, like swapping to clones and tricking his opponents, which is what his entire design is based on. If you decide to change these things, the likelihood of having an extremely vocal and negative and reaction from his community is very high.

Do you make changes? If so, what do you do?

I can guarantee you not everyone will agree with your proposed changes, which is fine and normal. A large part of being a designer is having the courage to make these kinds of calls and dealing with the fallout, even if your ideas are not always perfect. Even so, there are many cases like this and there are many times where changes need to be made. That’s essentially what we spent the bulk of our time figuring out and doing.

Also as a side note, finding out community sentiment doesn’t come from data, and community sentiment is often a large part of the picture as well, so if you only rely on data to make changes, you’ll end up making a lot of weird changes that no one will agree with.

As an example, Probius has always had an extremely high win rate and statistically deserves heavy nerfs. I don’t think that’d go over too well with the community :stuck_out_tongue:

56 Likes

Ofcourse! that little probe is scary as f

9 Likes

A japanese cookbook?

I think the main point to take away is that game design is about more than just decision making, but a willingness to declare what your gaming principals are and then follow through with that choice using a design philosophy.

For example, a low winrate hero is potentially acceptable using the metric you mentioned which is difficulty vs. effectiveness. An easy hero to play would drop in winrate at higher tiers as greater difficulty heroes replace them.

That said, obviously you can have multiple principals, so I’m wondering if you could give us a hierarchy on which principals you believe are most important, which principals you focus less on?

So is there any data you guys might release in the future?

Even just pick rates of heroes in draft or talent selection would be extremely helpful.

Win rate and pick rate is only one aspect of a hero. Things like kerry and shrines about a year ago is also something to look at. So are how many one tricks there are. I am alsongping to assumr that you have the abilty to track damage over the course of a game. Do you have the abilty to track how fast the damage happens? It it takes say 15 aa from ray to kill x hero, it that is happening over 25- 30 seconds thats no big deal. If its happening over 7 or 8 consistently that is.

With just knowing what sammys wim rate and pick rate is, its pretty hard to make changes with out testing them. Which i am sure you guys can do on a test dummy/ against each other internally. Over all the balance is not bad. When you make kits that are over loaded, they tend to be very hard to balance. Which is not so much the balance team but design team.

Lets just say i had all the data you have on sammy, i would look very hard at some higher level players and play the " what if game" if we took 3 damage, he took x longer to kill y camp/ hero. If we messes with cd, he does make it out so often. With out the data we have no real way to evaluate heroes other than what we personally see and what it feels like to play with/ against.

I know me personally i hate playing grossly over powered heroes. Its not fun. When ming still had ess of johan, i knew it was broken, so i avoided playing her. It ment you ate every missle( plus seeker) and you could not get away from the laser.

But on the other side if someone is at 45% winrate we can help to balance the hero. Maybe if you give a example of a hero who needs improvement we can analyse and come with ideas (not tass or dva because rework)

Maybe 5 of our ideas are bad but one is very useful and could improve the game in general

So Blizzard proved stat sites are accurate. Because Probius is constantly on the #1 spot by win rate.

However he is picked very rarely, only by really good Probius players.

AZJackson:
As an example, Probius has always had an extremely high win rate and statistically deserves heavy nerfs. I don’t think that’d go over too well with the community

He is endorsing us to play probius

3 Likes

It doesn’t work that way…
The sites might be accurate, but… the logic in your post is terrible.

7 Likes

The stat sites are roughly accurate, but they’re not 100%. I don’t think anyone’s ever said they’re completely false and should be ignored, at least not anyone talking sense. It’s just that they’re not the word of God.

Outliers like Probius will show obvious trends, but the small differences between, say, heroes like Anub’arak and Mal’ganis might not be perfect.

The only reason why Probius is not the best mage is because he’s easily blown up, but man, he’s stronk :smiley:

6 Likes

How is the logic terrible? Probius is “extremely high winrate” (per AZJackson words") in internal Blizzard data, and he is extremely high win rate in heroesprofile data. He is also very low pick rate in both data sets.

Seems you just want to dispute things you are clueless about. You provide no real counterargument.

2 Likes

I wonder if any other hero had a talent removed so fast as Ming at release. Or removed at all. Kael Ignite comes to mind, but that was reinstated later on in a different form.