Overwatch Mei Analysis

To Freeze or Not to Freeze? (TLDR at the Bottom)

The May 20, 2021 Overwatch 2 Developer Update put a lot of folks on edge, especially us Mei players. NO MORE FREEZE? Freezing is quintessential Mei, and for a lot of us it's what meikes her so fun to play. With it, what is she-- \(1/4\) Orisa with wall, \(1/4\) McCree with her rightclick? That still leaves half a character behind.

But even if Mei doesn't feel the same, it might not be all that bad. Today I'm going to be looking at the relationship between freezing the red team and Mei's performance.

The Data

Today we're going to be taking the top 100 Mei players according to OverBuff.com's Hero Score. While it's not necissarily the most representative sample, we're really after the "best performing" Meis to match Blizzard's top-down balancing strategy.

Basic Qualities

Before we dive into our analysis let's take a look at the data. Below we can see the different variables we have available to us for each of the top 100 players according to Mei: "Hero Score", SR, Games Played, Win Rate, % On Fire, Weapon Accuracy, Critical Hit %, Damage Blocked (in thousands), Final Blows, Enemies Frozen, Blizzard Kills, Eliminations, Objective Kills, Objective Time, Damage (in thousands), and Deaths.

We can also take a look at the distribution of each of these variables independently. Our data has a decent range in the top half of the SR distribution--from gold to GM+, and as we might expect from being the "top 100" Meis (according to Overbuff), we see higher values than the total average Mei win rate, which is around 51%. The majority of the other data looks well rounded and Normally distributed. But what about Games Played? We have some hardscore mei players in the mix, but in general why is it so low? Well, these data are snapshots from a certain time period in the season, and remember the season didn't start too long ago!

The best part about the Overbuff rankings data is each stat is attached to a single player. The figure below helps us visualize the relationships between the data, something we don't often have access to with summary stats. Here, the stronger the relationship, the larger the cirle, and the more similar the relationship, the more blue, while opposite relationships trend red. We see a lot of sensible stuff here. Firstly, every stat has perfect similarity with itself. Secondly, things like Final Blows and Elims are strongly positively related, while Winrate and Deaths are strongly negatively related.

How Might a Lack of Freeze Effect Mei?

Alright, so let's dive in- how is Mei's freeze related to her success? To check this out, we're going to estimate a couple of linear regressions. The first one is going to model Win Rate as a function of our other stats, and pay particular attention to Enemies Frozen.

To compensate for some Mei players logging a lot more hours into OW than others, we'll go ahead and weight our data proportionally to the number of games played. This will give Meis with more games making up their stats a higher influence in our regression, since we'd expect their stats to be a bit less variable.

Relationship with Win Rate

For this part, we'll be using

\[\text{Win Rate} = \text{Deaths} + \text{Elims} + \text{SR} + \cdots + \mathbf{\text{Enemies Frozen}} + \cdots + \text{Objective Time} + \varepsilon\]

After we run this model, we can check out the regression coefficients, which show us the relationship between each of the Mei stats and Mei's winrate. Remember, there Mei be different effects depending on Mei's SR! But we'll dive into that another time.

Here's a visual of our results. If the vertical line for the stat crosses the horizontal line at zero, eh, we're not super sure what the effect size is. But you can get a sense.

So what does this model tell us? Well firstly, oof, we should all really try to die less. Just one extra death per game is correlated with a 3% decrease in winrate on average!

Second, yep... it looks like Mei's freeze does help a bit. While there's some uncertainty (since our sample size is 100), it looks like an extra enemy frozen means an extra .8% increase in win rate, on average! Wow, that sounds pretty strong, but it does makes sense. When Mei freezes you... there's no getting out alive.

And for you stats nerds, here's a residual plot. This is just showing us how well a linear model is fitting our data (so does our \(R^2 = .36\), which isn't too bad). The errors our model is making look pretty random, symmetric, and distributed around 0, which is a great sign.

Relationship with SR

Okay, Mei's freeze is strong. But how much does it matter for climbing? In some ranks it's relatively easier to freeze the red team than in others. And we've all heard that GM Mei's are mostly right clicks.

Let's do the same analysis, but this time looking at the relationship with SR. So our model will be

\[\text{SR} = \text{Deaths} + \text{Elims} + \text{Win Rate} + \cdots + \mathbf{\text{Enemies Frozen}} + \cdots + \text{Objective Time} + \varepsilon\]

Again, we'll weight our model by games place and we're off to the races.

Wow, looks like a lot of things are related to SR. For example, an extra elim is related to a +70 change in SR, an extra percent crit is related to +60 SR. But there is some other interesting stuff going on. Firstly, an extra 1,000 damage blocked is related to a whopping +173 change in SR! Geez, I gotta work on my defensive walls. Second, being on fire and doing damage is somehow negatively related to SR? Well, it's not that those things are bad, it's just that at the highest ranks, Mei's don't really get the chance to do them. The same is true with freezing. Yes, with an extra enemy frozen being related to an 80 point drop in SR, GM Meis must rarely get the chance to freeze. But heck, she's still being played, and they're still GM, so Mei must be pretty multi-faceted!

Again, for you stats nerds, the residual plot. WARNING: Stats Lingo Ahead, Feel Free to Skip!!! This residual plot does show additional heteroskedasticity and potential nonlinear relationships in the data, but I went ahead and ran both polynomial regressions, used a Box Cox transformation on SR with \(\lambda = -.3\), and removed some potential outliers, and none of these actions substantially changed the findings.

In Conclusion (TL;DR)...

Mei's freeze is strong, there's no doubt about it. In general, an extra enemy frozen during a game is related to a .8% increase in winrate on average. But Mei's freeze is at its strongest at the lowest ranks where she gets the chance to do so. At the highest ranks, we see Meis are far more valuable for their Ice Wall-- 1k extra damage blocked during a game is associated with a whopping +173 SR!

So is it the end of Mei? Probably not. Mei is far more valuable for clutch Ice Walls. In fact, as long as they increase the damage enough, Mei might see even more play as her left click becomes a whole new monster to deal with. Hey, mei-be we should just give her a Reaper crossover skin now...