Setting Up the Largest Poker Behavior Study Ever Created
Whenever I speak about Beyond Tells it’s important to understand that I am essentially discussing two things. First, it’s a study we used to gather data on behaviors at the poker table. Second, it’s a system for categorizing, analyzing, and understanding nonverbal behavior at the poker table and a product that we call the Beyond Tells Training. In this video I am going to discuss the Beyond Tells study, which took over 2 years to come to fruition and is still ongoing.
In order to conduct this study we used a similar framework to something called Grounding Theory, which is a method of research used in the social sciences. The key thing to understand is that going into Beyond Tells we weren’t collecting data to prove or disprove a set of hypotheses. The goal was to create a massive database that we could analyze and that would help us answer practical questions to advance the game of poker.
This entire process can be broken down into 4 stages: Data Collection, Coding, Analysis and Product Development. This video will only be concerned with the first three stages.
The first step was to find a good sample of players. We wanted men and women of a variety of ages and experience levels. We used our email list and sent out a list and asked for volunteers. The response was overwhelming. We ended up with 53 players, ranging in age from 22 to 61 years old. There were 10 women and 43 men containing people with a wide range of experience levels, from complete beginners to seasoned professional poker players.
Once the players were selected, we assigned them to games based on their play preferences. We split up the sessions between cash games and tournaments. The cash games ranged in stakes from $0.50/$1, $1/$2 and $2/$5. The tournaments were a $100 buy-in, winner take all unless a deal was made between the players. All players used their own money to play and cashed out as they would have at a regular home game. We did not take a rake. Every session was 6 handed and lasted approximately 4hrs.
Players were recorded from 9 different camera angles. Despite the cameras, we wanted to keep the game as natural as possible. We hired a professional dealer and decided against using hole cams or RFID technology to record the cards. Hole cards and RFID require players to bring the cards to a specific point at the table and it changes their normal behavior, which would go directly against our purposes.
Instead, we allowed players to play as normal and when they folded, the dealer handed the cards to someone at the side of the table who recorded the cards manually.
The CodingThe next part was to code: come up with ways of recording and articulating behaviors at the poker table and then inputting them into a central database. We basically built our own version of Poker Tracker but for behaviors. Our database can connect actions at the table to observable behavior. So, if a player blinked 45 times on hand 56, we can record that and identify that the player had Aces and the hand lasted 4 minutes, and the board was a AJ4, etc. The purpose is to merge contextual information with behaviors.
How did we do this? Manually. We took all of the recorded hand information in these binders and input them into our database. Then teams performed coding on specific parts of the body using both qualitative and quantitative methods. For example, the most quantitative form of analysis we can do is to count blink rate. We counted every single time someone blinked in order to look for correlations between blink rate and card strength.
This coding stage is an ongoing process and if you are an intern at The Nonverbal Group or Beyond Tells, This is what you will probably start off doing for the bulk of your day.
Once we start to standardize behavior and code it, then we have the fun part: the analysis. This is where we look for patterns in behavior and see if they mean something. For example, I can see how many times someone checks their cards and how that compares to an actual hand strength. We can do this at an individual level, game level, or the entire sample. I can watch clips of every single time someone has a hand at the top of their range and compare that to the bottom. Or, I can watch every single hand a player plays looking for different things each time.
This was fun because this is where all the insights of Beyond Tells come from and as a poker player this has had a profound impact on my game. I mean there were nights where I would be in bed thinking to myself, “I wonder how upper facial movement changes when people have a strong hand?” Then I would comb through my database of videos and see if I could find anything there.
Over the past 2 years we have gotten considerably quicker and more effective at this process and we have plans to dramatically increase the effectiveness of what we are doing this year. This is also what I ultimately spend all my time doing. Still, this is just half of it, compared to the product development this was the easy part of Beyond Tells.