Imagine you’re playing a high-stakes card game, but you can’t see your opponent’s cards. You know they have a hand, but you don’t know whether it’s strong or weak. Every move they make gives you a clue, and with each new piece of information, you adjust your thinking about what they might be holding. This process of updating what you think based on what you observe is central to strategic decision-making in situations where information is incomplete-and it’s exactly what we mean by “beliefs” in game theory.
Table of Contents
- What are beliefs in game theory?
- Beliefs versus information sets: knowing versus thinking
- Why this distinction matters
- Bayesian updating: revising beliefs with new information
- The logic behind rational updating
- The mechanics of Bayes’ rule
- Walking through an example
- Applying Bayes’ rule in strategic settings
What are beliefs in game theory?
In game theory, a belief isn’t just a vague hunch or gut feeling. It’s a technical term with a precise meaning. A belief represents a player’s subjective probability assessment about the likelihood of being at a particular decision point (called a “node”) within a set of possible situations (called an “information set”) that they cannot distinguish between. Think of it as your best educated guess, expressed as a probability, about where you actually are in the game when you can’t see everything.
For example, suppose you’re negotiating a business deal, and your counterpart makes an aggressive first offer. You might think there’s a 70% chance they’re a tough negotiator who won’t budge much, and a 30% chance they’re actually flexible but starting high as a tactic. That 70-30 split is your belief-a probability distribution over the different “types” your opponent might be.
Beliefs versus information sets: knowing versus thinking
It’s crucial to understand the difference between what a player knows and what they believe. An information set captures what a player knows-specifically, it’s the collection of decision nodes (points in the game) that the player cannot tell apart. If you’re at an information set with multiple nodes, you know you’re somewhere in that set, but you don’t know exactly which node you’re at.
A belief, on the other hand, represents what you think about where you are within that information set. It’s expressed as a probability distribution across the nodes you cannot distinguish. For instance, if there are three possible scenarios in your information set, your belief might assign probabilities of 0.5, 0.3, and 0.2 to each scenario respectively. The information set tells you the boundaries of your uncertainty; the belief tells you how you’re distributing your confidence within those boundaries.
Consider a simple example: You’re player 2 in a game where player 1 can choose either action A or action B privately, and then you must respond. If you cannot observe which action player 1 chose, both resulting decision points are in the same information set for you. Your information set is the fact that you’re at one of those two points. Your belief is your assessment-perhaps “I think there’s a 60% chance player 1 chose A and a 40% chance they chose B.” That’s the difference: the set defines your ignorance, while the belief quantifies your judgment within that ignorance.
Why this distinction matters
This distinction is not just academic hairsplitting. It fundamentally affects how we analyze strategic behavior. In perfect Bayesian equilibrium, players must choose strategies that are optimal given their beliefs, and those beliefs must be consistent with the strategies being played whenever possible. If we only knew what players knew (their information sets) without understanding what they believed about where they were within those sets, we couldn’t predict or prescribe rational behavior.
Bayesian updating: revising beliefs with new information
Rational players don’t hold static beliefs-they update them as new information arrives. This process is called Bayesian updating or belief updating. It’s the mechanism by which a player modifies their initial beliefs (called “prior beliefs”) when they receive new evidence, forming updated beliefs (called “posterior beliefs”).
Let’s say you’re considering whether to invest in a startup. Initially, based on industry data, you believe there’s a 40% chance the company will succeed. That’s your prior belief. Then you learn that the company just secured a major partnership. This new information should change your assessment. Using Bayesian updating, you would combine your prior belief with the likelihood of observing this partnership under different success scenarios to form a new, higher probability of success-your posterior belief.
The logic behind rational updating
The key insight is that rational belief revision isn’t arbitrary. Bayesian updating uses the rules of conditional probability to ensure that beliefs change in a logically consistent way. When you observe an action or outcome, you should update your beliefs about the underlying state of the world by considering how likely that observation would be under each possible state.
In dynamic games with incomplete information, players continuously engage in this updating process. Every move by an opponent potentially reveals information about their type or their private knowledge. If a rival firm cuts prices aggressively, you might update your belief about whether they have lower costs than you thought. If a poker player raises heavily, you adjust your assessment of their hand strength. This iterative updating of beliefs is what allows strategic interaction to unfold rationally over time.
The mechanics of Bayes’ rule
At the heart of Bayesian updating lies a mathematical formula known as Bayes’ rule. While the formula itself might look intimidating, the logic is straightforward. Bayes’ rule provides a systematic way to update the probability of an event A given that you’ve observed event B.
The formula works like this: The probability of A given B equals the probability of B given A, multiplied by the prior probability of A, all divided by the overall probability of observing B. Mathematically, it’s written as P(A|B) = [P(B|A) × P(A)] / P(B).
Let’s break this down with a concrete example from medicine, which helps illustrate the concept clearly. Suppose a certain disease affects 1% of the population-that’s your prior probability, P(Disease) = 0.01. There’s a test for this disease that’s 95% accurate when someone has the disease (the test correctly identifies 95% of sick people), and 90% accurate when someone doesn’t have it (the test correctly identifies 90% of healthy people as healthy, meaning it has a 10% false positive rate).
Walking through an example
Now imagine someone tests positive. What’s the probability they actually have the disease? Intuitively, many people would guess around 95%, but Bayes’ rule reveals something surprising. We need to calculate: P(Disease|Positive Test) = [P(Positive Test|Disease) × P(Disease)] / P(Positive Test).
We know P(Positive Test|Disease) = 0.95 and P(Disease) = 0.01. To find P(Positive Test), we need to consider both ways someone can test positive: they have the disease and test positive (0.95 × 0.01 = 0.0095), or they don’t have the disease but get a false positive (0.10 × 0.99 = 0.099). So P(Positive Test) = 0.0095 + 0.099 = 0.1085.
Therefore, P(Disease|Positive Test) = 0.0095 / 0.1085 = 0.0875, or about 8.75%. Despite a positive test result on a 95% accurate test, the probability of actually having the disease is less than 9%! This counterintuitive result occurs because the disease is rare-there are many more healthy people getting false positives than sick people getting true positives.
Applying Bayes’ rule in strategic settings
In game theory, Bayes’ rule ensures that beliefs are updated consistently with observed actions and equilibrium strategies. When you observe an opponent’s move, you use Bayes’ rule to update your belief about what type of opponent they are. If certain types are more likely to choose certain actions, observing those actions should shift your beliefs accordingly.
For instance, in a signaling game where educated workers send signals to potential employers, an employer observes a worker’s education level and uses Bayes’ rule to update their belief about whether the worker is high-ability or low-ability. If high-ability workers are more likely to obtain higher education (because it’s less costly for them), then observing a high education level should increase the employer’s posterior belief that the worker is high-ability.
The beauty of Bayes’ rule is that it provides a unique, logically consistent way to update beliefs. It removes arbitrariness from belief revision and ensures that rational players all follow the same updating logic when they receive the same information. This consistency is essential for analyzing strategic equilibria in games with incomplete information, where beliefs and strategies must align in a mutually consistent way.
What do you think? How might understanding Bayesian updating change the way you interpret signals in everyday strategic situations, such as negotiations or competitive business decisions? Can you think of a situation where failing to properly update beliefs might lead to poor strategic outcomes?
Leave a Reply