Nonverbal Research at the The Poker Table: Behavioral Coding
I frequently mention how Beyond Tells is the largest and most extensive study of poker players ever conducted. In this piece we’ll discuss how we went about it and explain one of the key aspects of the study: behavioral coding. There is a lot of movement at the poker table, and the whole purpose of Beyond Tells is to determine if this movement actually means something. In order to really understand if movement means something you need to come up with methods for the consistent and accurate measurement of human behavior.
We use a mixture of quantitative and qualitative methods that can be described simply as: Frequency, Duration, and Style. I am going to go over each and then explain how they work together to create very accurate descriptions of behavior at the table.
Frequency refers to how many times something happens. Things like how often a player checks his cards, blinks, words spoken, etc. Essentially anything that can be counted uses this type of coding. For example, we can define a blink as when the eye opens and closes in less than 1 second. Then we just count all the times a player does it. We have analyzed the blink rates of a large part of our sample. Currently we are at over 65,000 blinks and growing.
Duration measures how long something takes and is measured in either frames or seconds. For example, how long you look at your cards or how long your smile lasts on your face.
You can check your cards twice in a hand which would be frequency and each card check can last a different amount of time. This would be the combination of frequency and duration.
Measuring Frequency and Duration is a time intensive task but relatively simple to do. It starts to get difficult when you try to articulate the frequency at which a player does something like play with his chips. How do you define “play with chips?” Does it include shuffling the chips, doing a coin roll, tossing them around? There are a lot of ways to define chip playing. Let’s say a player tends to roll the chips on his fingers when he’s strong, and he shuffles chips when he’s weak. If you were to label all chip plays the same, you would miss that data. This is why descriptions--operational definitions of behavior--which are qualitative in nature are so important and why you really need to combine qualitative and quantitative methods to understand the whole picture.
This is where what I call Style comes in. Labeling the style of a behavior is designed to decrease the subjective nature of behavior at the table.
Chip playing becomes: Chip Fidgeting, Chip Rolling, Chip Tapping, Chip Touching, Chip Shuffling. All of these categories have video examples and descriptions of what those styles actually mean.
Instead of saying, “The player played with his chips.”
We say, “The player exhibited a chip roll lasting 3.5 seconds.”
The goal of behavioral coding is to measure behavior consistently in order to perform behavioral analysis. Descriptions of behavior are so important because information is often lost in language.
Once you have accurate behavioral descriptions you need to find out if they actually mean something. To do this we created a massive database to input all of the contextual action of the game, what everyone has at all times, what the board texture is, stack sizes, bet amounts, etc. Our database is a lot like PokerTracker or Hold’Em Manager, which are databases for online hand histories. The only difference is that in our database we can input behavioral information alongside contextual information. This allows us to conduct every kind of analysis and find where specific behaviors exist in our sample. In a couple of seconds I can display all the times players in our sample have Aces or Kings, and then go over those moments looking for differences. Or compare the average card check of the top 10% of hands to the bottom 10% across a specific sample within our data set.
Some of you might be asking, “What's more valuable, Frequency, Duration, or Style?”
Looking at stats in a database is interesting, but sitting down and watching every single moment of every single video over the course of a year gives you a level of insight that I would never trade for raw data analysis.
The primary goal of Beyond Tells was to create a product and I can say with confidence that the majority of insights came from more casual observations of behavior. I am not just some researcher studying poker. I am a player and a very practical person. When playing poker you won’t have the ability to speed up, slow down, and count everything that happens. The true benefit of behavioral coding was the realization that the poker community has no intelligent way of discussing behavior at the poker table.
Advancements in poker came from language. If you say, “I was the short stack and my opponent was 3-betting very light so I shoved,” the person you’re talking to can get an understanding of what is going on.
When behavior is described what it is written and passed on as advice is often useless. Expanding your behavioral language and improving your ability to identify behavioral information and tells at the table is really about establishing a system for understanding, remembering and using nonverbal information at the table. And that is exactly what the Beyond Tells training is all about.
Our coding and analysis are far from done. Research is always ongoing. Currently, all of this is done manually, meaning it requires someone to actually sit down and code all of the information. I am going to eventually publish our behavioral coding guidelines so if you are crazy enough you can do your own research. If you have any questions or comments or you are conducting research on behavior, I would love to help out in anyway I can. Between my research on Beyond Tells and work at the Nonverbal Group, I have many ways of recording and breaking down behavior and this really is just a small portion of what is possible.