Cheap Talk and Credibility in Vee Communication

From Costly Signals to Cheap Talk

In signaling games, the message is the action, and it has a direct impact on payoffs. Proposing to move in together is costly in terms of commitment and potential awkwardness if rejected. In contrast, cheap talk refers to communication that is payoff-irrelevant on its own. The words "I'm feeling sad" cost nothing to say and don't change anyone's utility directly. Their only potential effect comes from their ability to change the listener's beliefs, which in turn may change their actions.

As the notes explain, any cheap talk game has a "babbling equilibrium." This is the equivalent of a pooling equilibrium where communication is completely ignored. If a Hinge believes a Partner will say whatever gets them the best immediate result, regardless of the truth, the Hinge will learn to disregard their words and act only on their prior beliefs. In this state, communication is just noise.

The crucial question is: under what conditions can we move beyond babbling and have informative communication?

The Credibility Condition: Alignment of Interests

Unlike costly signals, where credibility is enforced by the cost of the signal itself, the credibility of cheap talk depends entirely on the alignment of preferences between the sender and the receiver.

To understand this, let's look at the special cases mentioned in the MIT lecture notes.

Let's translate this into the context of a vee relationship:

• Common Interests (Fully Revealing Equilibrium): Imagine a Hinge and one of their partners are deciding on a restaurant for a date night. Both want to pick a place they will both genuinely enjoy. The partner (Sender) has private information about their craving ($\theta$). When they say, "I'm in the mood for Thai," the Hinge (Receiver) has every reason to believe them. Their interests are perfectly aligned—maximizing joint enjoyment. The sender has no incentive to lie, so their words are fully credible.

• Opposing Interests (Babbling Equilibrium): Now imagine the Hinge has one free Saturday to spend with either Partner A or Partner B. This is a zero-sum situation regarding that specific resource (the Hinge's time). If Partner A sends a message saying, "My need to see you this weekend is a 10/10, it's critically important," this is not credible. Partner B would say the same thing. Because their interests are in direct conflict, their words convey no information. The Hinge is left to make a decision based on other factors, not the cheap talk messages.

Most situations in relationships fall somewhere in between these two extremes. This is where communication becomes partially informative.

Partially Revealing Equilibria: The Crawford-Sobel Model

The most famous model of cheap talk, by Vincent Crawford and Joel Sobel, shows how a systematic bias between the sender and receiver leads to "partitioned" communication. The message doesn't reveal the exact state, but it credibly places the state within a certain range.

Let's build a model for a vee:

• Players: A Partner (Alex, the Sender) and the Hinge (Blair, the Receiver).
• State of the World ($\theta$): Alex's true level of satisfaction with the time they are getting in the relationship, on a scale from $\theta =0$ (completely dissatisfied) to $\theta =1$ (perfectly satisfied). This is Alex's private information.
• Action ($a$): The amount of change Blair makes to the time allocation, from $a=0$ (no change) to $a=1$ (a major reallocation of time towards Alex).
• Preferences & Bias:
  • Blair wants their action to match Alex's true satisfaction level. Blair's ideal action is $a=\theta$. If Alex is perfectly happy ($\theta =1$), Blair wants to make no changes. If Alex is very unhappy ($\theta =0.1$), Blair wants to make significant changes. Blair's utility can be modeled as $U_R(a,\theta )=-(a-\theta {)}^2$.
  • Alex appreciates Blair's response but always has an incentive to get a little more time than is strictly "necessary" to reach satisfaction. This systematic desire for "more" is a bias, $b>0$. Alex's ideal action from Blair is always $a=\theta +b$. Alex's utility is $U_S(a,\theta )=-(a-(\theta +b){)}^2$.

This bias $b$ is the measure of preference divergence. A large bias means Alex has a strong incentive to exaggerate their dissatisfaction to get more time. A small bias means their interests are more closely aligned with Blair's. The size of this bias determines how much information can be credibly communicated.

The image below, illustrating the Crawford-Sobel model, shows exactly how this works.

Let's interpret these graphs in terms of our scenario:

• N=1 (Large Bias): If Alex's bias $b$ for more time is very large, Blair knows that whatever Alex's true satisfaction level is, they will always have a strong incentive to claim they are dissatisfied to get Blair to act. Alex might say "I'm at a 0.1!" even if they are really at a 0.6. Anticipating this, Blair decides to ignore Alex's specific words. Blair will simply choose an action based on their average expectation of Alex's satisfaction. This is a babbling equilibrium. No information is transmitted. All states ($\theta$) are pooled into one uninformative message.

• N=2 (Medium Bias): If the bias $b$ is smaller, some communication becomes possible. Alex can't credibly report their exact satisfaction level, but they can credibly partition the space. For example, they can send one message for "things are generally okay" (e.g., $\theta >0.4$) and another for "I'm genuinely unhappy" (e.g., $\theta <0.4$). Alex wouldn't be tempted to say "I'm unhappy" if their satisfaction was just over the 0.4 threshold, because Blair's resulting large over-correction would be too much even for Alex's biased preference. Blair trusts these coarse signals and takes one of two actions in response. This is a partially revealing equilibrium.

• N=3 (Small Bias): If the bias $b$ is very small, their interests are closely aligned. Communication can become more fine-grained. Alex can now credibly partition their feelings into three ranges: "I'm unhappy," "I'm okay," and "I'm very satisfied." The messages are still not perfectly precise, but more information gets through.

The key takeaway is that the credibility of cheap talk is not all-or-nothing. It's a gradient. The more aligned the players' interests are, the finer the partitions of the state space can be, and the more information is transmitted.

Advanced Communication Dynamics in Vees

The basic sender-receiver model can be extended to capture more complex dynamics relevant to your work.

Multiple Rounds of Communication: As the MIT notes allude to, a back-and-forth conversation can sometimes achieve better outcomes than a single declaration. This process can help "co-create" a shared understanding and expose a sender who is bluffing. In a vee, this is the difference between a partner making a single statement and the Hinge engaging in a longer dialogue to probe and clarify, which can lead to more efficient and truthful outcomes.

Mediated Communication: What if communication doesn't happen directly, but through a third party? This is a concept known as mediated communication. In a vee context, this could be a relationship coach, a therapist, or even a trusted mutual friend. A skilled mediator can strategically filter or rephrase messages to manage the sender's incentive to lie, potentially allowing for more information to be transmitted than would be possible in a direct, face-to-face talk. Given your background as a relationship analyzer, this framework provides a formal way to think about the value of such intervention. A mediator can be designed to relax the incentive compatibility constraints that prevent truthful revelation.

Conclusion

In this lesson, we have moved from costly signals to costless "cheap talk," exploring the very different foundation of its credibility.

• Key Takeaways:
  • Cheap talk is costless, payoff-irrelevant communication whose power lies only in its ability to alter beliefs.
  • The default outcome is a "babbling equilibrium" where nothing is communicated, as there is no cost to lying.
  • Credible cheap talk is only possible when the sender's and receiver's preferences are sufficiently aligned. The degree of alignment determines the precision of communication.
  • With misaligned preferences, communication isn't necessarily useless. It often results in partially revealing equilibria, where information is conveyed in coarse categories (e.g., "high/low," "good/bad").
  • The credibility of a partner's words in a vee depends on the perceived alignment of their goals with the listener's goals for that specific issue.

So far in this module, we have focused on the informed player trying to send information. But what if the tables are turned? In our next and final lesson for this module, we will investigate how an uninformed player can proactively design situations or questions to "screen" the other person and compel them to reveal their private information.