How Good Are The Stats We Use?

In collaboration with @smartbackwards

Jul 05, 2025

Recently we’ve put the finishing touches on RWAA, which stands for Round Wins Above Average, as well as our own Win Probability stat. If you don’t have the greater context, there’s an entire series of articles that go over how the stat has evolved, as well as how it compares to other stats alongside highlighting the shortcomings of these other stats. But it’s very easy to get caught up in measuring something or critiquing something theoretically that we fail to consider its practicality.

At the same time, we all have our personal preferences when it comes to stats and what we use to evaluate players. Whether you are a recent adoptee of RWAA, into HLTV rating, or maybe you're an old school KPR or even K/D ratio type of guy, we all have one number we especially consider when doing player comparisons.

But is there a king of the hill, one stat to truly rule them all? If not, is there at least one that is the “best”?

While it’s hard to qualify what “best” actually means, we can come up with methods to evaluate them. One such method is looking at how these stats correlate with team success. The easiest way to define team success is to use their round win rate, a team that wins most rounds out of the ones they play is usually going to be the best.

The best teams this year by Round Win Rate at Big Events are as follows:

While obviously not perfect, it's a quick and dirty way to evaluate which team is the best, especially since almost every stat is based on per round measurements.

For this exercise, we’re using 37 different stats. Some of these are individual stats tracked across an entire team’s roster (KPR, Rating, WPA), while some of these are purely team based (5v4%, Flash Assists). The full list will be posted at the end so if you just want to see all the information without reading, scroll to the bottom.

We won’t be talking about *every* stat because some of these are pretty similar and some are included just for fun and aren’t meant for greater analysis. The full list will be available at the end, or in the graphic you’ll see if you saw this on twitter or something.

So here’s our process. We’ll look at every team that played at a Big Event this year, and we’ll mark down all of the stats we want to test aggregated across the entire team. That just means that instead of looking at just one individual, all five players' stats on the team are accounted for.

Then we will use a weighted correlation formula based on the number of rounds each team played to look at the correlations. What this means is that the teams that play more rounds will have more “weight” to their stats, while teams with less rounds played will have their stats matter less.

As for data sources, all of the basics that are covered on team or player pages are from HLTV. Meanwhile, all economy related stats, headshot related stats, and snake/flashed stats are from @smartbackwards on Twitter (or

btrams

on Substack). If you’re somehow reading this and don’t already follow him, then I sincerely wonder what planet sized boulder you live under. Secondly though, go follow him, he puts out high quality stuff.

As for WPA stats, it’s the same version used from my previous article and is sourced from me, so if you haven’t read that then I recommend it.

Without further ado, let’s look at some stats:

Headshot Stats

Rifler Headshot Damage%

R: 0.1232
R^2: 0.0152
Rank: 25th

Rifler Headshot Kill%

R: 0.1030
R^2: 0.0106
Rank: 27th

As stated, these stats were provided by btrams who subsequently told me not to do it and that a correlation was unlikely. And what a surprise, after a few hours of me adding everything up, he was right.

Hey but now we know zont1x doesn’t tank Spirit by going for body shots, and that headshot stats are just fun! On a serious note though, the spread between the teams wasn’t very dramatic at all, for headshot kill% most teams were between 50~60%, and for headshot damage% between 37.5% and 45%.

Funnily enough, the highest team by HSK% was Imperial Valkyries and the second was TheMongolZ (One of those teams was definitely not expected, but in some ways it makes sense that they relied more on getting headshots to get kills because their time to damage would be slower, so if they weren’t getting the headshot then they would probably just die).

Team Stats

4v5 Conversion Rate

R: 0.9396
R^2: 0.8829
Rank: 8th

5v4 Conversion Rate

R: 0.8813
R^2: 0.7768
Rank: 13th

Traded Death%

R: 0.6659
R^2: 0.4355
Rank: 17th

Pistol Round Win%

R: 0.4829
R^2: 0.2332
Rank: 18th

Utility Damage per Round

R: 0.3735
R^2: 0.1395
Rank: 19th

Flash Assists per Round

R: 0.2518
R^2: 0.0634
Rank: 22nd

These stats are courtesy of HLTV, and are featured when you go to the team stats page, FTU page, and pistol rounds page.

This is just another point to the idea that opening kills are kind of overrated, as 4v5 conversion rates actually had a higher correlation to team success than 5v4 conversion rates. Granted, both are pretty good and feature in the top half of the stats we’re looking at.

Pistol rounds are another stat we should highlight, because while the correlation isn’t very strong, it’s still higher than you would expect considering it only occupies two rounds of any given match. But correlation also does not mean causation, and it’s unclear whether this is a result of better teams winning pistols more often, or because winning pistols wins you a disproportionate number of rounds. As with everything nuanced, it’s probably a bit of both.

Also, utility is just overrated, I guess, as there’s basically no correlation between utility damage and flash assists to round win percentage.

“Who needs utility when you can just use your teammates as flashes”

-Albert Einstein, probably

Economy Stats

Full Buy Kills per Round

R: 0.8959
R^2: 0.8026
Rank: 12th

Full Buy Kill Differential per Round

R: 0.8721
R^2: 0.7606
Rank: 14th

Full Buy Deaths per Round

R: -0.7235
R^2: 0.5235
Rank: 16th

Anti Eco Kill Differential per Round

R: 0.3199
R^2: 0.1023
Rank: 20th

Anti Eco Deaths per Round

R: -0.2895
R^2: 0.0838
Rank: 21st

Eco Kills per Round

R: 0.2354
R^2: 0.0554
Rank: 23rd

Anti Eco Kills per Round

R: 0.1993
R^2: 0.0397
Rank: 24th

Eco Deaths per Round

R: 0.1423
R^2: 0.0202
Rank: 25th

Eco Kill Differential per Round

R: 0.1072
R^2: 0.0115
Rank: 27th

As stated previously, these stats are also courtesy of btrams. Just for clarification, a buy round is defined as a round where both teams have an average equipment value of 3800 or greater, and an anti-eco/eco is when one team’s average equipment value is greater than 2800 and the others is less than 800. This data does not include force buys.

As expected, full buys are a large majority of rounds played, so the stats in those rounds are most impactful. What is unexpected is that KPR in Buy Rounds has the highest correlation among these stats, because, as you’ll see later, Kill Differential per Round has a higher correlation than KPR when looking at all rounds played.

Other than that, Anti Eco stats matter more than Eco stats *slightly*, but the correlations aren’t different enough to say that definitively, and aren’t strong enough to make greater statements about what those rounds mean in the broader context of all rounds played.

Snake/Flashed Stats

Snake or Flashed Kill Difference per Round

R: 0.8840
R^2: 0.7814
Rank: 13th

Snake Kill Difference per Round

R: 0.8097
R^2: 0.6556
Rank: 17th

Flashed Kill Difference per Round

R: 0.8049
R^2: 0.6479
Rank: 18th

Snake or Flashed Deaths per Round

R: -0.7136
R^2: 0.5093
Rank: 20th

Snake Kills per Round

R: 0.6516
R^2: 0.4246
Rank: 22nd

Snake Deaths per Round

R: -0.6490
R^2: 0.4212
Rank: 23rd

Snake or Flashed Kills per Round

R: 0.6295
R^2: 0.3962
Rank: 24th

Flashed Deaths per Round

R: -0.5809
R^2: 0.3374
Rank: 25th

Flashed Kills per Round

R: 0.3681
R^2: 0.1355
Rank: 28th

Again, stats courtesy of btrams. A snake kill (or a turned kill for short) is a kill where the victim has the attacker out of his approximate field of view. To be more specific, it's all instances where a victim is looking at an angle 45 degree or greater away from pointing straight at the attacker. Snake deaths basically measure the number of times you're looking away and get shot in the back (or side) of the head. Also the threshold for being flashed here and having a flash assist by HLTV are different, so keep that in mind.

Overall these stats are some of the most interesting because they reveal the tactical intricacy of the game in multiple ways. Firstly, let’s think about what the context in-game would be for these kills to happen. The most common way for a flashed kill to happen is either in a bomb site execute or in a set piece play; as an example, flashing ramp on Mirage to clear it.

On the flip side, when are you most likely to kill somebody turned, that’s going to be in lurking scenarios or info pushes that stem from defaults. Think creating mid pressure so your lurker can creep out of palace to kill a guy ticket that’s looking connector. In essence, when you combine these two stats you’re kind of looking at ‘how good are this team's executes, and how good are this team's defaults’.

So if we read into these stats a little more, what they tell us is that it’s not just enough to be good at one of these things, or to be able to not be punished by one of these things, but the teams that win the most are both executing and defaulting well, as well as not being punished by those same things. It’s not enough to do just one or the other.

One such example is M80. Out of all of the teams in the sample, they have the fourth highest Snake or Flashed KPR, yet they have only a round win rate of 45%, so why is that? Well, when you look at their Snake or Flashed DPR, they are also second highest in that metric as well, which leads to them having a negative differential overall, leading to their losing record. It’s not just enough to have good team play on your own accord, you also need to be able to counter other teams and not lay victim to their traps as well.

The Basics

Kill Death Ratio

R: 0.9674
R^2: 0.9358
Rank: 4th

Kills per Round

R: 0.9655
R^2: 0.9322
Rank: 5th

Average Damage per Round

R: 0.9504
R^2: 0.9032
Rank: 6th

Deaths per Round

R: -0.9286
R^2: 0.8623
Rank: 10th

Turns out, the basics are pretty good. There’s another basic stat to calculate that’s in the top three but nobody really uses it so we’ll cover it later. KDR slightly edges out KPR, but you’re not going wrong using either of them. ADR ranking this high might be a little surprising, but doing damage often means getting kills so it’s not too surprising. And overall, the bare bones basics occupy half of the top 10, which goes to show that some stats are timeless.

Advanced Stats

Multikills per Round

R: 0.9474
R^2: 0.8975
Rank: 7th

Rounds with a Kill

R: 0.9337
R^2: 0.8718
Rank: 9th

KAST%

R: 0.8987
R^2: 0.8077
Rank: 11th

Opening Kills per Round

R: 0.8717
R^2: 0.7598
Rank: 15th

The first thing worth noticing when it comes to these stats, is that they all had a lower correlation than KDR, KPR, and ADR, which is really really interesting. Now, granted, these numbers might be slightly off because I’m taking these figures from the HLTV website and they only mark these down to 3 decimal places, so there is some degree of imprecision, so if two stats are close together it’s *possible* that one leapfrogs the other with more precise data, but three points is usually precise enough.

Multikills placement shouldn’t be surprising, if you read How Much is a Kill Worth, you’d know multikills all have a win rate roughly north of 75%. It’s pretty hard to lose rounds where you get multikills, so the more the merrier.

On the flip side, you might be confused by what “Rounds with a kill” means. Basically, it’s KAST without the A, the S, or the T. So it’s looking at if a player got a kill in that round or not and it’s averaging that among the team members. The fact that KAST is *worse* than the same stat with three less factors calls into question why we even use that metric in the first place.

Opening kills being the lowest from this group hearkens back to what was already mentioned between 5v4 and 4v5 conversion rates.

The Top 3

HLTV Rating 2.1

R: 0.9726
R^2: 0.9459
Rank: 3rd

You probably didn’t expect this. Now you would expect 2.1 to finish pretty highly, which it has, 3rd is a pretty good placement and it performs better than the most common basic stats as well as everything else.

But 3rd for what is also the most widely used stat in the community is also perplexing, which other stat ranks higher? Well it’s always worth reiterating that this correlation is still very very strong, and it’s not a bad stat to use. We’ll dive into more of the nuances of the differences in these stats, but it’s really up to you to come to your own conclusion on what to use. In other words, use whatever stat proves you right and you’ll never go wrong.

Kill Difference per Round

R: 0.9739
R^2: 0.9485
Rank: 2nd

Now if there’s one thing to take from this article, it’s this because (K-D)/R is a stat that isn’t listed anywhere that is almost never used in any context, and it turns out to be one of the best stats in terms of correlation to team success. Like……………. what the fuck?

It does make sense why this stat is good, it intuitively makes sense. The flaw of using K/D is that it lacks the context of round efficiency. For example, while Jame might have a high K/D, he’s not impacting in as many rounds as other players so his (K-D)/R might be worse than a player with a lower K/D. So if you’re somebody who enjoys and uses basic stats, and want something that on paper is just a little better, consider using this stat more.

Win Probability Added per Round

R: 0.9802
R^2: 0.9608
Rank: 1st

Just to clarify, this is my version of WPA and not HLTV’s Round Swing or Leetify Rating. I would have also checked those but Round Swing isn’t public yet and Leetify Rating is kind of a bitch to track in an exercise like this. I won’t lie though, a part of the motivation for this exercise was seeing just how effective this stat would be compared to others, and so far it passes the test.

Now what’s really funny is that WPA is actually just another version of Kill Difference per Round. All it does is instead of weighing every kill and death the same, it weighs them based on the change in probability, and it has some extra caveats for round ending events like saves, defuses, etc. But in that context, it’s not that surprising that both of these stats rank so similarly as they are essentially the same thing, just slightly different.

The Greater Point

It’s easy to get caught up in the finer details, but there’s a broader point that needs to be made. The most commonly used stats for player evaluation all have correlation coefficient of .95 or more. Even if some stats have better correlations, this isn’t some result that is permanent for any of these; there are and will be fluctuations over time just because that’s how data works.

What that’s all to say is, all of these stats are pretty damn good, and if you have a certain preference for any of those top 6, you’re gonna end up in a pretty good place.

The biggest differences between these stats really is how value is allocated. That might not make sense, so let’s use HLTV Rating vs. WPA as an example. In a team context, these stats perform pretty similarly when compared to round win rate, but the big difference between the two is which players in that team are rated higher.

When examining the top players for instance, WPA rates AWPers very highly, passive rifles decently, and usually rates aggressive riflers the worst. Meanwhile, HLTV rates aggressive riflers much greater, AWPers slightly worse, and passive riflers are usually about the same.

The same sort of breakdown can be spread with KPR vs ADR. AWPers usually have pretty low ADRs but higher KPRs, while that’s reversed when looking at aggressive players who can do a lot more damage but sometimes end up with less kills.

When you break it down on a team level though, the total value spread across a roster is roughly the same between all of these ratings, that’s ultimately what this experiment tells us. So the real tricky part and the biggest question we have yet to solve as a community is:

Which stat allocates “value” most accurately?

If you’ve read my other articles, you already know what I think, but at the same time I can’t “prove” anything. When I said earlier to use the stat that backs up your biases the most, I was only half joking. Ultimately it’s up to you to answer that previous question and come to your own conclusions.

The entire reason I even started writing these articles was because I had different ideas about this stuff compared to the greater community, and so it would be pretty hypocritical of me to proclaim my unorthodox answers as most correct. I’m hardly infallible, hell last year I said donk might’ve been the third best player in the year. Then I did the role adjustments for my last article and it was like a third eye opened because holy shit AWPers are insanely high rated in a way I never realize before.

I only mention this anecdote to say that as somebody who is projecting my beliefs to the world that I’m still asking myself these questions constantly. And all I ask of you is to do the same. Never take what I say for granted.

Thank you for reading, I hope you have a good day.

If you enjoyed the article, please consider following me on Twitter or Substack, I would greatly appreciate it. If you ever have any questions, never hesitate to ask.