The amount of movement at a poker table is extensive. Every single second, multiple players are blinking, smiling, raising, and more. 

The whole purpose of Beyond Tells is to teach players how to navigate this information. In order to do this effectively, we first need to create systems for measuring behavior at the poker table. 

Below is an overview of our coding process and how we deal with measuring movement at the poker table. 

Step 1 - Code Every Contextual Action at the Table

Before we even take a look at behavior we start with coding all the actions that occur at the table, including every single bet, check, raise, flop, turn, river, and the exact second it occurred. We use an overhead camera angle to mark the time and details (bet size, cards, etc) of each event using a tool called Frame.io.

All of this data exists in our Poker CAT (Contextual Analysis Tool) so we can perform raw data analysis at a later point in time. We also take that data and create a table animator using a javascript compiler to generate the animation you see in our videos and below. The animation is synced to video for viewing purposes, and allows us to create our own poker HUD, like a behavioral version of Hold’em Manager. This step is really important because we are ultimately looking at connections between hand strength, actions, and behavioral tells. ‍

Step 2: Code Specific Behaviors Based on Frequency ‍

Frequency-based coding involves marking when specific behavioral events happen without taking into account the properties of that behavior. This type of coding would not be useful if we are interested in the qualitative variations of a behavior. So for example, there are so many different ways you can smile that just looking at the frequency at which someone smiles isn’t really the most effective way to analyze that type of behavior.  However, a blink is fairly straightforward in terms of coding. We can simply define a blink as occurring when the eyelids fully close and open again in under 0.5 seconds, and we are just looking to identify when the blink actually occurred. In Beyond Tells, a team of 40+ people manually marked every single time a player blinked. This was a very labor-intensive process, and we spent hundreds of hours counting over 500,000 blinks.

Someone watches the behavior and marks every single time someone blinks by placing a comment of “B” when the lids of the eyes close in Frame.io.  We can have more senior analysts physically spot check work to see if blinks are being coded correctly by just clicking each of these moments and making sure the eyes are fully closed. This improves our coding standards and helps establish a high level of inter-coder reliability. And yes, a team of 40 plus people typed B on a keyboard well over half a million times. Now some of you might be asking, “Blake, why not use automated facial coding software?” The short answer is we tried and I wasn’t 100% confident with the results we were getting, so we did the work manually instead. It’s very important to note that this type of quantitative measurement at the table is fairly straightforward, but it’s not the most useful method when it comes to the evolving game of poker. 

Step 3: Qualitative Coding 

When it comes to observing behavior at the poker table, qualitative coding methods are without a doubt the most powerful method to actually understand how a player’s behavior gives off information. However, establishing standards for qualitatively recording behavior is a challenging process. For example, take a look at this behavior below. ‍

‍
How would you describe this behavior? 

‍Do we say that he lifted his hand 2.5 inches off the table and bet? Was this bet tossed, flicked, placed, thrown?. There are so many ways to articulate it and that is the problem. If I classify this as a flick, and we classify a very similar behavior as a toss, the behaviors are different and we can’t draw the same conclusions. These types of behaviors need to be defined to a level that is useful. There is an unlimited number of different ways a player can bet. They can toss, throw, place, and do these all at different speeds and with different styles. Due to this diversity of behavior we need a practical method for classifying the behavior. When it comes to betting, we use a two label classification system that includes speed and style. Speed is used to describe the speed at which the action took place or how long it took for a player to touch a chip and execute the bet. Style refers to the way in which an action is executed. For example, I call the action in the above video a cavalier call. It’s when a player throws out a bet in a nonchalant, less intentional manner, and translates to a “sure why not” type of action.

When you watch the video this behavior is easier to understand but when you have to operationally define that behavior, it becomes a little bit more complex. Qualitative coding can be applied to facial expressions, vocal statements, concealment strategies, postural shifts, and a lot more. However, the real value of qualitative coding is at our next step.‍

Step 4: Player Relative Qualitative Coding ‍

Our goal is to find behavior that is actually useful at the poker table. The most powerful method for this is what I call Player Relative Qualitative Coding. I want to make sure every poker player understands that the usefulness of a style of bet (the way someone bets) is really only realized when compared relatively to that player. There are certain trends that you see, but ultimately if someone bets quickly it means nothing. However, when a player never bets quickly and now they are operating 4x times faster, and we see other behaviors indicating weakness along with that, we have something meaningful. The usefulness of behavior at the poker table is truly unlocked when you learn how to “code” in real time, meaning you describe movement at the table in a way where you can actually notice a difference. If I were to go back and redo the Beyond Tells study again, I would dedicate significantly more resources towards player relative coding. It’s the best approach from a practical perspective and provides us with a very deep understanding of how a player’s movement gives off information that poker players would deem useful.

Coding the behavior of poker players is something we are constantly improving and it’s definitely an evolving process. Make sure you check out How We Do a Full Behavioral Breakdown of a Poker Player to learn exactly what we do with that behavior. 

If you are interested in coding methods of nonverbal behavior, I strongly recommend New Handbook of Methods in Nonverbal Behavior Research (Series in Affective Science) 1st Edition. It’s a collection of some of the best journal articles on coding behavior and a lot of my ideas were built from the research in this great book.

Finally, if you have any questions about coding nonverbal behaviors or want to do a thesis or dissertation using our dataset, reach out and let us know. 

Note that this is just exposing you to the research side of behavior. If you are interested in actually transforming your game, please check out the Beyond Tells 2.0 training. You can get access to the first week for free.