← Wiki Index / Business & Economics Theory / Crawford–Sobel Model
📈 Business & Economics Theory

Crawford–Sobel Model

The Crawford–Sobel model is a foundational game-theoretic model of strategic information transmission (cheap talk) in which an informed sender communicates with an uninformed decision-maker without binding verification costs.

The Crawford–Sobel model, introduced by economists Vincent P. Crawford and Joel Sobel in their landmark 1982 Econometrica paper "Strategic Information Transmission," is the canonical game-theoretic framework for cheap talk—costless, non-binding, and unverifiable communication between an informed Sender and an uninformed Receiver. Unlike Spence's job-market signaling model, where messages carry direct differential costs, or verifiable-disclosure games, where claims can be audited, messages in the Crawford–Sobel model enter neither player's payoff function directly.

Despite the absence of lying penalties or signaling costs, Crawford and Sobel demonstrated that meaningful communication can still survive in equilibrium—provided the Sender's and Receiver's preferences are partially aligned—and proved that all equilibria take a partitional (interval-based) form whose informativeness declines as preference divergence grows.

Model Setup and Equilibrium Structure

In the canonical uniform-quadratic specification:

  1. State of the world: Nature draws a continuous private state θ uniformly on [0, 1], observed only by the Sender.
  2. Communication: The Sender transmits a costless message m from a message space M to the Receiver.
  3. Action: Observing m, the Receiver chooses a continuous action y ∈ ℝ.
  4. Payoffs: The Receiver's payoff is −(y − θ)2, so the Receiver's optimal action equals the posterior expectation E[θ | m]. The Sender's payoff is −(y − (θ + b))2, where b > 0 is a constant bias parameter measuring conflict of interest (the Sender always prefers an action b units higher than the Receiver's ideal point).

Because the Sender wants to shift y upward by b, full revelation (m = θ) cannot be an equilibrium whenever b ≠ 0: if the Receiver believed every message literally, the Sender would profitably exaggerate θ by b. Instead, in any Perfect Bayesian Equilibrium, the state space [0, 1] is partitioned into a finite number N of contiguous intervals [a0, a1], [a1, a2], …, [aN−1, aN] (with a0 = 0 and aN = 1). The Sender reports only which interval contains θ, and the Receiver responds with the midpoint of that interval, (ai−1 + ai) / 2.

Bias Parameter (b)Maximum Equilibrium Partitions (Nmax)Communication OutcomeInformation Transmitted
b = 0Unbounded (N → ∞)Full revelation (first-best)Complete state θ revealed
0 < b < 1/4Finite Nmax ≥ 2Coarse interval cheap talkPartial, noisy information
b ≥ 1/4Nmax = 1Babbling equilibrium onlyZero information transmitted

The Indifference Condition and Partition Bounds

At every interior partition boundary ai, the Sender must be indifferent between inducing the lower interval's action and the upper interval's action:

ai + b − (ai−1 + ai)/2 = (ai + ai+1)/2 − (ai + b)

Simplifying gives the fundamental second-order difference equation ai+1 − ai = (ai − ai−1) + 4b. Thus, successive intervals must widen by 4b from left to right. Because the sum of N widening intervals must fit inside [0, 1], the maximum number of informative partitions satisfies 2N(N − 1)b < 1. For b ≥ 1/4, even a two-partition equilibrium fails to fit inside [0, 1], leaving only the uninformative babbling equilibrium (N = 1) in which the Receiver ignores all messages.

Economic and Institutional Applications

The Crawford–Sobel framework underlies modern economic analysis of legislative committee hearings, central bank forward guidance, analyst stock recommendations, medical second opinions, and organizational delegation. It also illuminates why delegation (giving decision authority directly to a biased expert within a constrained range) can sometimes outperform cheap talk when b is moderate, and why multi-sender competition or repeated reputation building can expand the set of credible messages.

Another key theoretical implication is equilibrium multiplicity: whenever an equilibrium with N > 1 informative intervals exists, every coarser equilibrium with 1, 2, …, N − 1 intervals—including the uninformative babbling equilibrium—also exists simultaneously. Standard welfare comparisons under ex ante Pareto optimality typically focus on the most informative equilibrium (Nmax), because finer partitions strictly raise both the Receiver's and the Sender's expected payoffs prior to learning θ.

Frequently Asked Questions

Why is full truth-telling impossible in the Crawford–Sobel model when bias b > 0? If the Receiver believed every message truthfully, the Sender would always report θ + b to induce their own preferred action, destroying the credibility of point-value reports.

What is a babbling equilibrium? A babbling equilibrium is an uninformative equilibrium—which always exists in cheap-talk games—in which the Sender's messages are uncorrelated with the true state and the Receiver ignores them, choosing the prior expected value.

How does equilibrium precision change as bias decreases? As the Sender's bias b shrinks toward zero, the maximum number of equilibrium intervals Nmax grows, intervals become finer, and communication approaches full information revelation.

References

  1. Crawford, V. P., & Sobel, J. (1982). Strategic Information Transmission. Econometrica. https://doi.org/10.2307/1913390
  2. Ogawa, H. (2018). Receiver's Sensitivity and Strategic Information Transmission in Multi-sender Cheap Talk. SSRN Electronic Journal. https://doi.org/10.2139/ssrn.3252283
  3. Kono, H., & Kandori, M. (2021). Corrigendum to Crawford and Sobel (1982) “Strategic Information Transmission”. Econometrica. https://doi.org/10.3982/ecta17617
🧬 Support Open Science
Explore interactive science games on LowKey Games or browse scientific apparel & accessories in the CRV Science Shop.
💡 Spot an inaccuracy or have primary literature to expand this article?
Submit a citation or revision note using your shared CRV Science / LowKey Games profile ID.
Suggest Revision