Using clutch stats effectively

Clutches can turn a game on its head, but how do we best measure that impact?

1vsX situations are some of Counter-Strike's most captivating and it is no wonder that when you rattle off a list of the game's best ever highlights, many of them are clutches. They turn a team game into a solo affair, an opportunity for the scene's brightest stars to have the limelight all to themselves.

But it is not easy to quantify the impact of clutches, they are statistically rare and even the best closers average around 0.02 1vsX wins per round. There are barely a handful of these moments in each map, leaving glaring gaps in sample size unless we take data from a huge time span. When the best players in the world average 0.02 in a statistic and the worst 0.01, the problem is obvious — there simply is not enough data.

That is why clutches are listed in raw numbers on our site, such as in the leaderboards or a player's profile. They are simply too rare to account for opportunity over the varied sample sizes a HLTV profile covers. Statistics on a player's profile need to be useful over all time, one event, and even one map. Clutches per round fail to do that.

So do we stop there? Leaving clutching statistics in raw numbers is safer, but it makes comparing players next to impossible. Richard "⁠shox⁠" Papillon has the most clutches of all time, but he also has close to the most rounds played; that metric reveals as much about shox's playstyle and his longevity as it does his aptitude in the clutch.

Another common metric, though not on our site, is 1vs1 win percentage. This is useful because it makes it easier to compare players across sample sizes, as well as being incredibly easy to understand. If a player wins more 1vs1s than they lose, we understand that they are doing something right in these situations.

But sample size is still important. Ricardo "⁠boltz⁠" Prass's 100% win rate in 1vs1s during 2022 seems brilliant, but when you realise that comes across just eight situations it is clearly less impressive than Oleksandr "⁠s1mple⁠" Kostyliev winning 34 and losing just 13. As sample size increases, the bar for what is outstanding decreases.

The solution to this is by using a differential rather than a percentage or ratio. Just like a kill-death ratio inflates players with only a few deaths, a 1vs1 win percentage inflates those with small sample sizes. That's why you just see a 1vs1 win differential on the site, such as Andreas "⁠Xyp9x⁠" Højsleth's +211 1vs1 differential over his career.

A raw percentage is still useful as we can see in the above chart, but you can see the effect of sample size. The dots form a triangle-type shape as the gap between a good and bad percentage decreases with a larger sample.

Another way to circumvent the rarity of clutches is to give more value to more difficult clutches. The simplest way to do this is to literally give one point for a 1vs1 win, two for a 1vs2, three for a 1vs3, four for a 1vs4, and five for a 1vs5. Then players are given more reward for winning in more difficult situations. Because there are more points on offer now, dividing by round is not quite so catastrophic to illustrate the differences between players.

We can increase the difference further, though, by dividing not by one round played but by thirty. This is an alternative to dividing by map, with thirty rounds the maximum per map in regulation but retaining the accuracy of round-based division. Thus, players are not punished for having shorter or overtime-ridden maps. Once these clutch points are divided by a player's '30s', we get the following list:

Ricky "⁠floppy⁠" Kemery was top of the leaderboard for raw 1vsX wins per round, but in clutch points he drops to fourth thanks to Dmitry "⁠sh1ro⁠" Sokolov and Kaike "⁠KSCERATO⁠" Cerato's proficiency in harder clutches. Helvijs "⁠broky⁠" Saukants also climbs up the rankings, winning 30 1vs2s compared to 37 1vs1s in 2022.

Still, we are not talking about a perfect statistic. We are, for the most part, still measuring a player's opportunity to win clutches rather than their ability to win them; the players that top this leaderboard are all late-round specialists, and will naturally rack up clutches due to their role.

They are in this role for a reason, of course, but we can still do more to work out which players win clutches because of their role, and which are showing consistent skill in those situations.

Naturally, the answer is a scatterplot. We combine a player's clutch points per 30 rounds with their 1vs1 win percentage to show both the quantity and quality of a player's performance in the late round. Players that score highly in both metrics are therefore not only getting into clutches often, but also winning them more often than the average player. This chart only features players with a minimum of 20 1vs1 attempts in 2022 so as not to skew the data, leaving a pretty clear picture of a player's ability:

The median 1vs1 win percentage (and thus divider between the four corners) among this sample is 60% rather than the expected 50%, likely due to a '1on1' here being a 1on2 for their opposition.

sh1ro and KSCERATO, the top two performers by clutch points, are still clear of the pack when we include 1vs1%. Both win more than 65% of their 1vs1s over a hefty sample size, establishing themselves as the world's best closers. s1mple, Fredrik "⁠REZ⁠" Sterner, Olek "⁠hades⁠" Miskiewicz and Karim "⁠Krimbo⁠" Moussa also appear in the top right of the chart with an even higher 70%+ 1vs1 win percentage.

We can also see the players who are left behind when we look at both metrics at once. Perhaps surprisingly, Benjamin "⁠blameF⁠" Bremer, Ilya "⁠Perfecto⁠" Zalutskiy, and Keith "⁠NAF⁠" Markovic are around the average mark in terms of winning 1vs1s, meaning their clutch points might be inflated by the sheer number of attempts their role allows them to have. The same is true of Ilya "⁠m0NESY⁠" Osipov and Florian "⁠syrsoN⁠" Rische above them.

However, this is by no means conclusive evidence. We have still not solved the sample size issue — can we really say that blameF and Perfecto are weak in the clutch when their sample sizes are just 60 and 49 respectively? These are high sample sizes in terms of 1vs1s, but would be far too low to reach a judgement on in any other metric.

This is an irrefutable fact, one that no method can overcome. So, while we can still use clutching statistics as measures of a player's quality in those situations, we have to be exceedingly careful in doing so. It is a situation where an eye test of an expert will always have more value than hard statistics: seeing how calm they are, how much fear the opposition have, or how many clutches a player 'deserved' to win rather than just the outcome.

That does not mean that these statistics are useless, it just means we need to come at them from a different angle. And, when a statistic is hard to use to measure ability, it is nearly always still useful at measuring a player's style. We cannot say if blameF is good at winning clutches over a sample size of 60 1vs1s, but we can say that Astralis create situations where he is in the clutch more often than the other players on his team.

This leads us to our final metric, clutches won as a percentage of the team's total for 2022. Only players who played the majority of maps for their organisation in 2022 are considered, meaning some players are absent from this list.

Remember that this list does not show a player's ability directly; it just shows the faith the player's team have in them to win these situations. They are not in these clutches by accident, with teams devising plans to put certain players into clutches more than others. Because of that, a player's percentage of team's clutches can operate like opening kill attempts, showing their style in direct relation to their teammate.

A lot of this just confirms common knowledge. Dzhami "⁠Jame⁠" Ali wins 34.6% of Outsiders clutches, Mathieu "⁠ZywOo⁠" Herbaut 33.3% of Vitality's. But this is the same as opening kill attempts, where we do not need statistical assistance to know that Mareks "⁠YEKINDAR⁠" Gaļinskis or Andrei "⁠arT⁠" Piovezan are more aggressive than the usual player.

The interest is in the players that you might not know as dedicated closers, like Genc "⁠gxx-⁠" Kolgeci (32.7%) for Bad News Eagles, Igor "⁠Forester⁠" Bezotecheskiy (32.2%) for ex-Entropiq, or Pavel "⁠S1ren⁠" Ogloblin's 29.6% for Spirit. There is also interest in the players at the bottom of the scale. Vladislav "⁠nafany⁠" Gorshkov wins just 8.8% of Cloud9's clutches, Kristian "⁠k0nfig⁠" Wienecke won 11.1% of Astralis', and Håvard "⁠rain⁠" Nygaard 14.2% of FaZe's.

Like any statistic, clutching metrics can lead you down the wrong path, something which their rarity only makes worse. But we still have enough to make something of it as long as we proceed with caution; we can roughly see who wins the most clutches and show who each team prefers to have in these situations.

The reason it is still only 'rough' is that we cannot go deeper. We have not, so far, nailed down a way to reward clutches in more important situations, like gun rounds or late in the map: A clutch when both teams are low on money swings a game far more than an average 1vsX. Like flashbangs, clutches can still be recorded in a way that shows a player's style but it is harder to measure their skill. Because of that, our eyes remain the king — for now.

Read more

Using flashbang statistics effectively