Get rid of Overbuff?

[Titanium-1686]

Since one of my posts has been referenced in this thread, I feel the need to clarify the intent of that post:

Fine. I'll Say It

My post was not written as an attack on the reliability and usefulness of statistics. Statistics are incredibly useful and can create a very accurate depiction of the game’s current balance state from a purely objective standpoint. Statistics are amazing, and they serve as a much-needed anchor to reality. I had to select one element of the developer balancing triangle (community feedback, developer opinions, statistics) to dominate the balance of Overwatch, I would select statistics without hesitation. I’ll go even further and wager that Overwatch would be in a much better state than it is today if the game was balanced purely around statistics from the start.

With all of that in mind, I said this in the aforementioned thread:

@Titanium-1686

Winrates are not accurate depictions of hero viability. They just aren’t. Stop pretending they are just because they align with your arguments.

And I explain why this is the case:

@Titanium-1686

stop pretending that winrates are effective metrics for hero viability in a game where mirror selections are possible.

@Titanium-1686

to use that winrate as evidence for an argument is nothing more than a bit of confirmation bias derived from a wildly erratic and unpredictable measure.

Symmetra 2.0 and Torbjorn 1.0 both featured 65% winrates when it was almost unanimous among the Overwatch community that these two heroes were flaming garbage, routinely labeled as “troll-picks” because of how underpowered they were.

Without looking, can you guess which hero has the highest winrate in GM this month? I couldn’t. I never would have guessed this character.

Everyone have their predictions?

It’s Pharah.

It’s not Reinhardt, who was hands-down the best Tank for the majority of that time.
It’s not Ana, who was hands-down the best Support for the majority of that time.
It’s not McCree, who was hands-down the best DPS for the majority of that time.

It’s Pharah.

McCree, the undisputed best DPS hero in the game for the past six months has the lowest GM winrate in the game this month with 49%.

Because while statistics are amazing, you also need to know which statistics to look for, and which statistics contribute no value based upon the game’s mechanics. In Overwatch, heroes can appear on both teams. Because of this, any hero that is picked enough to ever appear on both teams simultaneously will naturally have a winrate that gravitates towards 50%. When you combine this with random variation and other factors such as one-tricks and “niche” picks, it becomes incredibly difficult to get a steady measurement of how good a hero is through winrates. As mentioned in the post above, we have seen heroes known to be underperforming reach incredibly high winrates.

Winrates, in general, are very poor at illustrating the actual state of the game. When a winrate aligns with an argument, it is often coincidental rather than causational.

In that same post, I go from refuting winrates as an effective measure to using pickrates not just as evidence, but as the very backbone of my argument. Not just pickrates in general either, but specifically the Grandmaster pickrates. I explain my reasoning for this when someone mentioned it in the replies to that post:

@Titanium-1686

I’m basing my points off of GM statistics, which removes fun from the equation almost entirely.

If you play a character that’s fun in GM, but the character isn’t good in GM, then you’re not going to stay in GM.

@Titanium-1686

It’s just logic. There are three “filters”, if you will, regulating who stays in GM and who drops out.

• Players want to win, so they play the best heroes and climb to or stay in GM.
• Players want to play [hero that just so happens to be good], so they climb to or stay in GM.
• Players want to play [hero that just so happens to be bad], so they do not climb to or fall out of GM.

Thus, the population in GM will be reflective of what heroes are viable and optimal.

@Titanium-1686

I didn’t mean that as an absolute. Of course the “meta” characters are not going to have 16.67% GM pickrates while everyone else has 0%. You can certainly one-trick sub-optimal characters to GM if you put the time and effort in; I did so with Mercy back when she was hands-down the worst healer in the game prior to her rework. What matters is how difficult it is to reach and maintain GM with that given hero. The easier it is to reach GM as a certain hero, the higher that hero’s pickrate will be in GM. The harder it is to reach GM as a certain hero, the lower that hero’s pickrate will be in GM.

Of course, when working with a data set as large as this one, there are bound to be some outliers, but that doesn’t take away from the overall trends that can be observed from pickrates.

As an example, in season 4, Mercy had a 12% Quickplay pickrate (the highest in the game) and a 13% overall Competitive Play pickrate (also the highest in the game). Yet, when you sorted for GM, her pickrate was dead last of all healers at 3-4%. That’s because GM weeds out the influence raw popularity has over pickrates. Again, there were outliers; there were still Mercy players in GM, but every hero will see at least some playtime in every rank, regardless as to how bad they are in regards to viability.

…Except for Bastion. Fun fact: Bastion is the only character in the game to ever maintain a 0% pickrate in GM for a month.

The take-away from my post isn’t that statistics are bad or inherently misleading; they most certainly are not. Take-away is that you, as someone looking at the statistics, need to know what metric you are trying to measure and what statistics provide the most clear and unimpeded depiction of that metric. You need to identify what confounding variables there are, if any, when interpreting that statistic.

Statistics are not bad; the problem is that not enough people know how to use them.