Bayesian Thinking & Esports Predictions

Do Esports Players Have a Natural Bayesian Thought Process?

E-sports has come up in the world as an alternative medium for entertainment especially in the interim post-COVID world. Where initially live-sporting events screeched to a halt and the world was left in an endless loop playing board games, Esports proved to be an immediate relief for some people looking for something, anything to scratch that itch to watch a competitive, live sport of any kind. From virtual NASCAR and F-1 races conducted on makeshift gaming rigs to online worldwide events where teams — separated by hundreds, even thousands of miles — created the edge of your seat, nail-biting experiences for those of us who love to watch something competitive.

The growth of Esports has also taken a turn into a realm of live sports which is still looking to gain traction in all 50 states (legally), live betting. The attraction that Esports betting has over traditional sports comes in the form of mini-sets, the pace and speed of games, and the complexity of matches in which one could either place bets on the entire outcome of a tournament or individual sets and player/team matchups. It creates an endless opportunity for throwing away your money!

In this vein, there are some platforms that take the anxiety and real-world consequences of Esports betting out of the equation, leaving only the unknown and non-determined factor of the outcome into the hands of solely the players battling each other on the screen. In this piece we will take this prospect, and apply it to the Bayesian thought process to decide if the analytical mind of esports players and their predictions based on experience and practice could translate into a Naive Bayesian outlook, and whether or not you could use that knowledge to make informed wagers on Esports and other forms of probability-based wagers.

Bayesian thinking could be summed up as making predictive analysis based on not only what the data has to say, but also what your expertise, experience, and previously held beliefs tell you as well. Whereas the frequentist definition sees probability as the long-run expected frequency of occurrence (P(A) = n/N), where n is the number of times event A occurs in N opportunities, the Bayesian view of probability is related to the degree of belief. It is a measure of the plausibility of an event given incomplete knowledge.*

The Bayesian Theorem

Forget about the schematics of the formula for a second, and let’s examine how we could make this work for possibly predicting Esports games; given we know limited amounts of information about the players and even the game itself!

Twitch.tv, the Amazon-owned, billion-dollar API that came up from the humble grassroots beginnings of Justin.tv and now hosts everything from live video game streams, ASMR, foodie topics, and live DJ sets (what?) recently implemented the channel points system to spice up live streams and increase streamer-chat interaction in lots of fun and imaginative ways. The short breakdown is:

  • Viewers can earn channel points by watching streams, chatting, making donations, and other interactive methods
  • Channel Points can be wagered on simple polls and mini bets on the outcome of anything game or stream related
  • Viewers can cash out accumulated channel points for goodies provided by the streamer (channel) without actually spending any money.
Yes, you can bet channel points on almost anything!

This system creates a place where viewers have more incentive to engage in the streams and events of their favorite channels directly, and the host (streamer) has more avenues for engagement with their audience, potentially boosting viewership and revenue.

Let’s now take a look at one of my favorite streamers and professional commentators, Sajam, and his now-infamous “Will it Kill?” series and see how we can apply a Bayesian perspective into netting goodies from his bag.

“Will it Kill?” is a popular Twitch/Youtube series from host Sajam

Suppose we take a random clip from the video and try to substantiate whether the combo will kill based on the limited information given, context clues, and some inferred data. In that vein, let’s try and make a prediction based on the clip at 21:16 of the above embed.

We’ve already been given some information by the host and context clues from the game (if we know where to look). Let’s write it out to pose it as a problem.

Suppose Seth hits an opponent with 40% health remaining. The chances of dying at 40% life in this game is quite high, at around 80%. Seth can kill in this scenario at a rate of 30% if his meter is over 100 on regular hit. His opponent was hit by a high-counter. What would be the probability that the combo kills?

Using Bayes Theorem lets look to solve the probability that the combo kills.

Prior: P(A) = P(opp dying at 40% life) = .80
Likelihood: P(B|A) = P(normal-hit|kill) = .30
Posterior: P(A|B) = P(kill|high-counter) = ?
P(¬A) = P(opp not dying at 40% life) = .20
P(B|¬A) = P(normal-hit|no kill) = .70

Bayes formula using our values and given information

Breaking this down in numeric terms we have:

Solving for our Posterior, or P(A|B). In this case, P(kill|high-counter)

P(kill|highcounter) = .63157

This means that given what just occurred on the screen, Seth has a 63% chance to kill poor Nanaya given he just hit her with a high counter! But given that we now have the odds, can we make an informed bet based on this data?

Well, given that we had the time to complete a full Bayes problem before the 1 minute cutoff time of the Channel Points betting phase (which we don’t) it would be pretty unrealistic to do this every time by hand, or calculator, or even in our heads. However, competitive, pro gamers and viewers make these types of calculations all the time right in the heat of battle and under immense pressure. We take in various context clues and build off of past experience and research to make informed, calculated decisions in a split second! Very much like how you might learn to instinctively drive a car and make split-second adjustments (without ever really thinking about it).

Now that we have our prediction value of 63%, let’s go see if our combo is actually going to end Nanaya or not!

One more time, for extra impact

While useful in theory across many different real-world applications (from sports betting, to gambling, etc.), Bayes Theorem is also incredibly fun to play around with and throw against the wall to see if some theory or belief actually sticks. If interested in other analytical papers on how gamers think and make split second decisions, feel free to view my website: mynus.gg where I cover Esports topics and journalism, and other great reads listed below!

Recommended Readings

Bayesian Thinking: https://www.statisticalengineering.com/bayes_thinking.htm

Chaz Frazer Website: https://www.mynus.gg/

Sajam’s YouTube: https://www.youtube.com/channel/UCVsmYrE8-v3VS7XWg3cXp9g

Sajam’s Twitch Channel: https://www.twitch.tv/sajam

Playing to Win, David Sirlin: http://www.sirlin.net/ptw

The Will to Keep Winning, Daigo Umehara: https://www.amazon.com/Will-Keep-Winning-DAIGO-UMEHARA-ebook/dp/B01JOEKKWU

Data Scientist/Esports Analyst/Linguist/Japanese Interpreter: I do a lot in the intersection of gaming and data, helping esports grow.