Contribution
It replaces the annual, country-level expert score of media freedom with a content-based score for each outlet at high frequency. Each nominally independent outlet is placed on a scale whose ends are fixed by outlets of known allegiance: regime-owned outlets mark full capture, and opposition outlets in exile or online mark full independence. It is validated on one country by five checks the authors set in advance.
Key insight
A regime’s editorial preferences are not published, but the outlets it owns reveal them, and outlets it cannot reach reveal what reporting free of it looks like. Capture can then be measured as relative position, with no need to know the “true” agenda: an outlet is as captured as its coverage is closer to the regime’s own outlets than to the uncoercible ones. The measure is a distance ratio. What makes a movement toward the regime pole capture, rather than a change of mind, is not in the measure.
Assumptions
- Definition. Media capture is “a situation in which news media fail to provide the public with scrutiny of government performance”, narrowed by “Another condition for media capture is therefore that an outlet has to be coerced by an external actor to alter its editorial choices in his or her favour.”
- Regime-owned outlets express the regime’s preferences. “Being under complete control of the regime, these outlets can be assumed to adhere to its editorial preferences.” The regime’s preference is assumed coherent.
- Opposition outlets mark full independence, and their changes in coverage reflect the changing importance of issues.
- Agenda and tone are the levers. Agenda bias is grounded in agenda-setting research, and tone bias in effects of tone on attitudes. Framing is left out deliberately, for want of an agreed definition and measure.
- Topics relevant to legitimacy are chosen in advance: welfare, security, democratic procedure, and, for Nicaragua, COVID-19, the 2018 protests, the OAS dispute, elections, repression and censorship, and the revolution.
Key results
- Topic independence score (Eq. 1). i = [(d(P_N, P_Reg) − d(P_N, P_Opp)) / d(P_Opp, P_Reg) + 1] / 2, with 1 for an outlet that covers a topic like the opposition and 0 for one that covers it like the regime. By the triangle inequality the numerator’s difference lies within ±d(P_Opp, P_Reg), so the score lies in [0, 1] (checked here). It is undefined when the poles coincide.
- Outlet agenda score (Eqs. 2–3). A weighted sum of topic scores, with weights proportional to how far apart the poles are on each topic. The paper’s example, distances 0.2, 0.1 and 0.05 giving weights 0.57, 0.28 and 0.14, was rechecked.
- Tone independence score (Eq. 4). The same ratio on mean sentence polarity (−1, 0, 1 from pysentimiento) over sentences mentioning Ortega.
- Data. Twelve of 27 national outlets with an active web presence, and 144,700 articles. Six are regime-owned, three nominally independent (Radio Corporación, Canal 10, Canal 14) and two opposition (Confidencial, 100% Noticias). A 30-topic STM with outlet and date as covariates.
- Validity checks I–V, all reported passed.
- I: topics track known events (COVID-19 cases near 6% prevalence in 2020; elections in 2016–2017 and 2021).
- II: opposition outlets cover the OAS, COVID-19, protests and human-rights topics more than regime outlets, and regime outlets cover welfare and infrastructure more.
- III: opposition tone averages about −.4 and regime tone about .2.
- IV: all three independent outlets’ agenda scores fall over 2018–2022.
- V: agenda and tone agree that Canal 14 moved most and Radio Corporación least. Canal 14’s tone score fell from about .8 to below .5 by early 2022. Canal 10 had too few mentions of Ortega for a tone score.
- Against expert indices. RSF and V-Dem series show a general decline since 2018 and cannot show that the outlets responded differently.
Claims
| id | claim | strength | support |
|---|---|---|---|
| C1 | The method gives outlet-level, high-frequency scores of alignment with a regime’s editorial preferences | strong, as a construction | Eqs. 1–4; the Nicaraguan application |
| C2 | In Nicaragua, 2018–2022, nominally independent outlets moved toward the regime pole, Canal 14 most and Radio Corporación least | moderate | Figures 3–5; no statistical tests or uncertainty for the scores are reported in the main text |
| C3 | The scores measure media capture | weak to moderate | the five checks are consistent with it; whether movement is coerced is argued from the context of harassment, and the authors grant that “it is impossible to definitively reject” a free change of editorial line |
| C4 | The method improves on annual expert surveys | moderate | it resolves outlets and months, which the surveys do not; it has not been compared with them on the same quantity |
Concepts
- Media capture (regime-driven): an outlet’s editorial choices “forcibly substituted by the regime’s editorial preferences in terms of tone when addressing the regime leadership, and media agenda”.
- Nominally independent outlet: not overtly owned by the regime and still operating legally in the country.
- Agenda independence score, tone independence score: an outlet’s relative position between the regime and opposition poles, 0 to 1.
- Agenda bias, tone bias: unequal attention to topics, and evaluative slant toward an actor.
Connections
It grounds agenda bias in agenda-setting (McCombs and Shaw, LIT-1335, through the Luo et al. meta-analysis) and takes “capture” from Stigler’s regulatory capture. It cites Guriev and Treisman on informational autocracy and Glasius on authoritarian practices. It does not cite Besley and Prat or Szeidl and Szucs. Its method is close in spirit to Szeidl and Szucs’s use of a critical reference outlet to net out the neutral level of coverage, with two poles where they have one.
Bearing on the record
- TERM-066. The paper’s definition adds a condition the term lacks and the term adds one the paper lacks. The paper requires coercion by an external actor, so an outlet that agrees with the regime is not captured. The term requires a standard set apart from the process judged, and the paper supplies exactly such a standard, made operational: the uncoercible outlets’ coverage. A25’s caution holds for this paper’s own measure: “Agenda and tone are observable indicators of possible capture. They are not themselves a complete causal identification of capture.” The paper agrees in its validity section. Its observable signature is one of the three things the term says to keep apart, and the paper shows how to measure it, not how to tell coerced from agreed alignment.
- CLAIM-404. The measure detects influence on an evaluator’s output without access to its decisions or procedures, which fits the claim’s starting point that control need not issue decisions. It does not tell the evidence route from the criteria route: a falling agenda score could be either.
- TERM-064, the direction “institution legible to society”. The method is a way to make an outlet’s governance legible to outsiders from its output alone, with the caution the term already carries: whoever chooses the poles and the topic list controls what counts as captured. The topics here are chosen by the researchers as relevant to the regime’s legitimacy.
No instruction for machine-learning practice; nothing for the anthology.
Limitations
- The poles are assumed, not measured. If opposition outlets change their niche, for example becoming more critical after being driven into exile, a stable independent outlet loses score. The authors raise this and say there is “no obvious better gold standard”.
- Coercion is not observed. The score measures alignment. The authors propose correlating score drops with data on attacks on journalists and have not done it.
- One country, three outlets. Canal 10 has no tone score.
- No uncertainty. The main text reports no statistical tests or intervals for the scores or the pole differences, though it calls the differences significant. The classifier’s validation is in an annex not read here.
- Requires a fully captured outlet, which limits it to regimes already far along. Tone uses untargeted sentence sentiment.
Open questions
- Do score drops follow specific regime measures, such as the cybercrime law, and acts of harassment? That would be the causal test the paper proposes.
- Does the measure agree with Szeidl and Szucs’s single-reference slant in a case where both can be computed, such as Hungary, which the authors name as the next application?