Contribution
It made media capture endogenous in a political agency model. Whether the government silences the press is decided by a bargaining game, so the features of the media market (how many outlets, how they are owned, how much they earn from audiences) determine whether voters get the news, and through that, corruption and turnover. It gives “independence” of the media a precise sense: the difficulty, τ, of transferring resources from the government to an outlet.
Key insight
Plurality protects the press because news is non-rival. If one outlet publishes, all voters learn. So silencing the press means silencing every outlet, and each outlet’s price is what it would earn as the only one to break the story, a, not its share a/n. “If the government wants to buy out all the media, it has to pay each of them as if it were a monopoly provider of unbiased information.” The cost of capture rises linearly in the number of outlets while the benefit, the rent of office, does not.
Assumptions
- Two-period retrospective voting. The incumbent is good with probability γ and gives voters 1, or bad and gives 0. A challenger is drawn from the same distribution. Voters do not observe their payoff before voting.
- Hard news only. If the incumbent is bad, all n outlets receive the same verifiable signal with probability q. “news cannot be fabricated”, signals can only be bad, and only verifiable information is printed.
- Audience revenue. An outlet that reports news earns a/m when m outlets report it, and 0 otherwise. If at least one outlet reports, all voters are informed.
- Transfers. The incumbent sees the signal and makes simultaneous, private offers tᵢ ≥ 0. A transfer tᵢ costs her tᵢ and yields tᵢ/τ to the outlet, τ ∈ [0, ∞). She gets r − Σtᵢ if re-elected and −Σtᵢ if not. Footnote 7: the result survives if offers are observed by other outlets.
- Equilibrium. Pure-strategy perfect Bayesian equilibrium in which voters do not play weakly dominated strategies.
Key results
- Proposition 1. If n < r/(τa), the media are captured: each outlet suppresses its signal for a bribe tᵢ = τa. If n ≥ r/(τa), each reports truthfully. The appendix proves this is the unique equilibrium of the class, with the incumbent indifferent at nτa = r. Re-derived in THEORY-837.
- Consequences. Under capture, voters’ expected utility is γ in both periods and turnover is zero (“voters are indifferent between the incumbent and the challenger but they vote for the incumbent”). Without capture, a bad incumbent is found out with probability q, turnover is q(1 − γ), and second-period welfare is γ + q(1 − γ)γ. Total audience revenue is qa when the media are free and 0 when captured.
- Proposition 2. Turnover, voter welfare and audience revenue are non-decreasing in q, n, a and τ. q acts only when the media are free; n, a and τ act through whether they are.
- Table 1, suggestive only. In 88 countries (Djankov et al. data), high state ownership of newspapers goes with 7.21 more years of a chief executive’s tenure (s.e. 1.82) and 1.24 more on the ICRG corruption scale (s.e. 0.26). High ownership concentration goes with 5.46 years (1.78) and 1.16 (0.25). “While such evidence crude and cannot be interpreted causally, it is encouraging that these facts are consistent with Propositions 1 and 2.”
- Proposition 3, moral hazard. With rent extraction y and detection probability Ψ(y)q, more media activity lowers rent and raises voter welfare, but its effect on turnover is ambiguous. Screening raises turnover, while discipline makes bad incumbents harder to catch. Turnover must rise in q near q = 0.
- Proposition 4, free entry at cost c. The media are free if the free-entry number of outlets, ⌊qa/c⌋, exceeds the number the incumbent is willing to buy, ⌊r/(τa)⌋. Barriers to entry therefore invite capture.
- Lemma 5 and Proposition 6, chosen monitoring quality. In equilibrium M = ⌊r/(τa)⌋ outlets choose the same high quality q̂(1) and are “potentially captured”. The rest form a tail with quality q̂(i) decreasing in rank. A bad incumbent is removed with probability q̂(M + 1). Turnover and welfare rise with a and τ. The authors read the steep vertical differentiation of the US and UK press as a sign of a press “working for the audience rather than for the government”.
- Proposition 7, access instead of money. If the government can only grant interviews worth Φ(m̂) to each of m̂ favoured outlets, with Φ decreasing, the media are captured iff Φ(n) > a.
- Propositions 8–11, segmented audiences. When a share 1 − φ of viewers read only their own outlet, the incumbent needs to silence only enough outlets to keep half the voters uninformed. The cost of capture rises in φ, τᵢ and aᵢ. It is strictly increasing in n when φ ≥ 1/2. For φ < 1/2 it is non-decreasing in n only once integer effects are set aside.
Claims
| id | claim | strength | support |
|---|---|---|---|
| C1 | In the baseline model the media are captured iff nτa < r, with each outlet paid τa | strong | Proposition 1 and its proof; re-derived in THEORY-837 |
| C2 | Capture lowers turnover and voter welfare | strong in the model | Propositions 1–2; with moral hazard, welfare holds and turnover becomes ambiguous (Proposition 3) |
| C3 | Plurality, high transaction costs (independent, dispersed or foreign ownership) and commercial audience revenue protect against capture | strong in the model, as comparative statics | Propositions 2, 4, 6, 7, 9 and 10 |
| C4 | State ownership and concentration of newspapers go with longer tenure and more corruption across countries | weak | Table 1, raw differences in means, “cannot be interpreted causally” |
| C5 | Formal press freedom is no guarantee of a press free from political interference | moderate | the model’s equilibrium capture under no censorship; illustrated by Russia, Peru, Italy |
Concepts
- Captured: “In situations where the media receives a transfer in exchange for silence we say that the media is captured, referring otherwise to the media being independent.” A transaction, not a disposition.
- Transaction cost τ: how hard it is for the incumbent to move value to an outlet. Lowest for state-owned media, higher for widely held, foreign or independently owned media.
- Commercial motive a: the most an outlet can earn from audiences by reporting news.
- Screening and discipline effects of media activity, in the moral hazard extension.
Connections
It builds on the agency models of Barro (1973) and Ferejohn (1986). It cites McMillan and Zoido on Montesinos’s payments to Peruvian media, Djankov et al. on state ownership, and Gentzkow and Shapiro, Mullainathan and Shleifer and Baron for media bias from sources other than capture. The record’s readings of Di Tella and Franceschelli (LIT-1319) and Szeidl and Szucs (LIT-1315) take it as their model. Szeidl and Szucs equate their δ, the government’s ability to reward slant, with τ.
Bearing on the record
- CLAIM-404, evidence route. The model is a clean case of an evaluator governed through its evidence alone. The voters’ decision rule is correct Bayesian updating throughout, and nothing is decided for them. Capture removes the evidence they would update on. In the captured equilibrium voters are not deceived: they know silence means nothing, and their posterior stays at the prior γ. That separates the evidence route from persuasion (THEORY-404), where a receiver who knows the design is moved. Here the receiver is not moved; it is left at its prior, which is enough for the incumbent, since the tie goes to her.
- CLAIM-404, the loop between institutions. The model has the press-as-evaluator structure of the claim’s third part, with the evaluated party able to affect the evaluator’s resources. But it is static. Capture is decided once by comparing nτa with r, and nothing feeds back from the outcome to the next period’s n, τ, a or r, so it contains no amplification loop. The record’s loop condition (CLAIM-330) has no counterpart here. The model supports the claim that formal independence (no censorship) does not make the evaluation independent. It does not support the claim’s dynamic part.
- CASE-123, press variant. The variant’s “a newsroom whose conditions of production the evaluated party’s decisions affect” is this model’s outlet, with transfers as the channel. The case’s sentence that Besley and Prat rest on an unread work can now be replaced: the model is read, it is static, and its mechanism is purchase of silence, not a drift in how the newsroom reads evidence.
- TERM-066. Besley and Prat’s “captured” is the purpose-relative, transactional sense the term set aside (“a decision bought outright is captured in it”). It is a single bargain, not a durable regime reproduced through the outlet’s own processes, and it needs no outside standard, since suppression of a true verifiable signal is defined in the model. If the owner adopts the term as drafted, this paper’s capture falls outside it. That is a reason to file the institutional sense as a narrower term, as the term’s own body suggests.
No instruction for machine-learning practice; nothing for the anthology.
Limitations
- Hard, homogeneous news. All outlets hold the same verifiable signal, which only a bad type generates. With soft news the authors expect a reputation game they do not solve. Di Tella and Franceschelli argue that with non-verifiable news partial bribing can pay, which the model rules out.
- Indifferent voters. In the captured equilibrium the posterior after silence equals the challenger’s, so voters are exactly indifferent. That they re-elect is forced by the pure-strategy restriction (fn. 9). A mixed equilibrium with random voting exists and is called “analogous”.
- Static. No dynamics of capture, entrenchment or reputation.
- A notational slip in the working paper. The appendix proof says an outlet “accepts tᵢ > a and rejects tᵢ < a”, while the strategies and the proposition use τa. The conclusion, nτa against r, is consistent with τa. The published version was not checked.
- Evidence. Table 1 is raw differences in means across 88 countries.
Open questions
- With soft or heterogeneous news, can an incumbent profitably buy some outlets and not others? That would break the all-or-nothing logic that makes plurality protective.
- What happens when capture today changes tomorrow’s n, a or τ, for example by bankrupting critics or favouring friendly entrants? That is the dynamic the record’s claim needs and the model omits. Szeidl and Szucs’s endogenous ownership is one version.