The Value of Saving

A brief analysis of save rounds in Counter Strike: Global Offensive

Counter Strike: Saves

Counter strike is a First Person Shooter (FPS) game that has been around in some competitive form since the turn of the millennium. Over time it has grown alongside esports, into one of the biggest games. One thing that is mostly unique to Counter Strike (Until Valorant) in the world of competitive FPS games is the economy. Unlike Call of Duty or Overwatch, the arsenal that a player has at a given time is purchased with money earned from kills, or from round-end payouts. With economy playing a factor, teams have to make a decision on rounds where they find themselves shorthanded, either go for it, with less people alive, or cut their losses and hide away, saving their equipment for the next round, where they will begin back at five on five.

Advantages to Saving

These players aren't just slacking off and taking a break from work, saving has a wide array of benefits, some being more tangible than others. Here is a graph of how much equipment a team saved on a given round, and their win rate in the following round, broken down by side, grouped by every $500 in value saved.

To quantify the impact of saving, we can compare the win rates after saving to a baseline win rate after losing a round (Horizontal lines on the graph). More often than not, saving has a positive value on the next rounds win rate. We can also see a few neat observations: The CT side rarely saves at full strength, most saves happen when a site is lost and the rest of the team hides away, The T side saves much more often than the CT side, holding 76% of all save rounds (which is also why the T graph is much smoother).

Modeling Saves

All of this is great, but it doesn’t really tell the whole story. It shows that when a team saves, their future chances of winning go up, but it doesn’t take into account the fact that their current chances of winning go to zero. To account for this we can use predictive models, similar to the one used for Operation Broken Fang, to get the saving teams win chance at the time of saving.

By breaking it down into possible outcomes, we can then get the expected value of each decision. The outcomes being: save, go for it and win, go for it and lose. Using the models outputs, the average save occurs when a team has a 14% chance to win the round, we also know that a teams average win rate after saving is 40%, and the average win rate after losing (and not saving) is 35%.

Here are the outcomes of going for it, and their expected rounds:

Win Both Rounds = 0.14*0.65 = 0.091 * 2 = 0.182 Expected Rounds

Win First, Lose Second = 0.14*0.35 = 0.049 * 1 = 0.049 Expected Rounds

Lose First, Win Second = 0.86*0.35 = 0.301 * 1 = 0.301 Expected Rounds

Lose Both Rounds = 0.86*0.65 = 0.559 * 0 = 0 Expected Rounds

Adding them all up gives us 0.532 Expected Rounds from going for it.

The outcomes of saving are much simpler, since the first round is a wash. The expected rounds of saving, is just the win rate after saving, which averages 40%, or 0.4 expected rounds. This brings up the question, is saving is even worth it? Here is a way to visualize it:

The blue and yellow lines show the marginal gains of saving a given value, or how much saving affected their win rate in the next round. The black line shows their current chance of winning the given round.

The way I interpret the outcome is that teams are pulling the trigger too quickly, or over valuing saves. The minimal increase in win probability of 5% is much less valuable than both the potential economic impact, and the still existent chance to win the current round. Saving is a smart decision when the next round win probability gain is greater than the current win chance (Not taking into account the fact that saving is automatically a lost round). If only teams could know their current win chance, and how much actual value they get from saving, teams would be able to see if saving in a given situation would be the smart play.

A couple of points on the data:

• Data used is from January 1st 2021-December 31st 2021, and the model is trained on all 2 star rated games from the 5 years preceding.
• Win probability at time of saving is taken at the earliest tick that a team had the same amount of players alive as they did savers, if 3 people saved, it’s the earliest tick in that round the team had 3 people alive, which is a close approximation to when the save would be called. This means it doesn’t take into account times when a saver is unsuccessful, meaning a lower probability than actual is returned, which could drive the 53.2% chance even higher
• Saved values that occurred less than 40 times were removed from the graphs, but not the win rate calculation, they made the graphs look off.
• This is no where near a conclusive result, just something I was curious about after reading a paper on pulling your goalie in hockey.

Playstyles, and Future Ideas

The numbers presented earlier are an aggregate, and more digging would need to be done on a more context based analysis, looking at different teams, players, maps, weapons saved, opponent economy, round loss bonus etc. For example we can look into team and player playstyles.

Looking at a subset of the year, here is a graph outlining team’s save rates per round lost:

Unsurprising to most, Virtus Pro, home to Russian AWPer Jame, has the highest rate of saves. Teams that are more reliant on a single AWPer to carry them through rounds save much more often, as shown in the following graph, where the best AWPers save much more often than riflers.

There are also intangibles that come with saving, for example Team Titan. Now that tactical pauses are standard in leagues, it doesn’t apply as much, but VeryGames/Titan would take save rounds as a way to calm themselves, and workout strats for when they could full buy, known as Titan Timeouts.

Closing Thoughts

A lot more research would need to be done to fully conclude the value of saving, and to figure out when it is beneficial to do so, but in my eyes, not only is it more exciting when players attempt clutches, but it is so much more rewarding when it works out. There is also so much room for more analytics in counter strike, an area that feels underutilized. If you are interested, the dataset for save value can be found here.