# Digital Censorship

## **What this driver is**

Digital Censorship is a continuing, time-varying factor within rights, social stability, and the information environment. It is modeled as a Driver because its intensity, capacity, or prevalence can change across reporting periods and can transmit the effects of multiple events, decisions, and institutions to later outcomes. It is not a dated event, a person, an organization, a welfare score, or an assertion that every movement in a correlated series was caused by the same mechanism. The relevant scope is the population, market, institution, infrastructure, or ecological system actually exposed to Digital Censorship; a national or sectoral observation must not be silently generalized to the whole world.

The boundary of Digital Censorship is narrower than its category and broader than one headline. It covers the durable condition named by the title, but excludes downstream welfare outcomes that must be measured separately. It also remains distinct from [Disinformation Prevalence](/drivers/disinformation-prevalence): the two may interact or share a proxy, yet they represent different causal questions. Digital Censorship is therefore a persistent analytical node, not a label for every adjacent development.

Digital Censorship concerns deliberate restrictions on access to online information or communication through blocking, filtering, throttling, shutdowns, removals, licensing, intermediary liability, or compelled platform action. It must be distinguished from ordinary moderation under a transparent rule and from connectivity failures without restrictive intent. The concept is narrower than [State Repression](/drivers/state-repression) and differs from Disinformation Prevalence: censorship limits what can be reached or said, whereas disinformation changes the content and credibility of what circulates.

The temporal record attached to Digital Censorship uses Voice and Accountability - Governance score (0-100) (GOV_WGI_VA_SC) as a disclosed proxy for 2020-2024. The proxy is an observable lens, not a complete operational definition. The stored scale is inverted because a higher raw observation represents less of Digital Censorship; after inversion, a higher normalized value means a stronger level of the named factor. Current weight is derived from the latest real normalized observation rather than entered as an editorial score, probability, forecast, or confidence estimate.

## **Mechanism**

The mechanism for Digital Censorship begins with a change in the factor itself and then moves through civic participation, access to trustworthy information, exposure to coercion, social trust, and the ability of groups to organize. The first step is exposure: the change must reach identifiable households, firms, public bodies, infrastructure, ecosystems, or security actors. The second step is transmission through prices, incentives, rules, information, physical constraints, organizational capacity, or behavior. The third step is adaptation: exposed actors may substitute, relocate, delay decisions, change compliance, invest, seek protection, or pass costs to others. The final welfare effect is the net result after those responses, not the initial movement alone.

For Digital Censorship, direct and indirect effects must be separated. A direct effect changes safety, access, income, health, legal protection, service continuity, or environmental exposure without a long chain of assumptions. An indirect effect passes through fiscal space, expectations, legitimacy, supply networks, knowledge, or capital formation. Each extra link makes timing and magnitude more conditional and raises the evidentiary burden for a graph relation.

Distribution is part of the mechanism. The same movement in Digital Censorship may help one group and harm another because exposure, geography, wealth, age, legal status, occupation, insurance, and institutional quality differ. Analysis must identify who is exposed, who can adapt, and who bears transition costs; an aggregate average can conceal opposing effects.

Timing also matters for Digital Censorship. Safety, availability, prices, or service interruption can move quickly; budgets, investment, reform, demography, diffusion, and capital replacement take longer. Feedback may reinforce the move, while substitution, policy response, learning, or resilience may offset it. The timeline records movement, but every DriverIndicatorImpact must separately state lag, direction, strength, and evidence.

## **Evidence base**

The primary quantitative record is the [Voice and Accountability - Governance score (0-100) dataset](https://api.worldbank.org/v2/country/all/indicator/GOV_WGI_VA_SC?date=2020:2025&format=json&per_page=20000). It supplies the raw dated observations, provider unit, reporting scope, and aggregation note used for the 2020-2024 Driver series. Across that available record, Digital Censorship finished above its first normalized observation: 0.43523966 in 2020 versus 0.44970528 in 2024. The minimum was 0.43523966 in 2020, the maximum was 0.44970528 in 2024, and the observed sequence contained 4 increases and 0 decreases. These are descriptive facts about the selected proxy; they are not an estimated causal effect on welfare.

The zero-to-one conversion is mechanical and preserved in lineage: Published percentage divided by 100 and clamped to [0,1]. For Digital Censorship, the scale is inverted because higher raw values indicate less of the named factor. Adjacent real observations create a segment only when the value changes; direction follows the sign and strength is the absolute change. The latest point sets current weight. Normalization preserves reproducibility but neither makes unlike concepts interchangeable nor turns a proxy into a complete index.

For Digital Censorship, the provider describes the selected series in these terms: Voice and Accountability (VA) captures perceptions of the extent to which citizens can participate in selecting their government including electoral integrity, and of accountability mechanisms for citizens—reflected in the ability to access information, governmental oversight bodies, and a robust traditional/digital media landscape. Governance score is a linear transformation of the governance estimate using hypothetical worst-case and base-case countries. It is an absolute score ranging from 0-100. Larger values correspond to better governance. The (country-series-time) metadata include the (a) number of sources, and (b) confidence intervals (lower and upper bound). The number of sources indicates the number of underlying data sources on which the governance score (respectively the underlying governance estimate) is based. The confidence intervals include the lower and upper bound of th... The cited producer is The Worldwide Governance Indicators: Revised Methodology for Measuring Governance Using Perception Data., World Bank Group (WBG), uri: https://www.worldbank.org/en/publication/worldwide-governance-indicators, note: Detailed documentation of the WGI, interactive tools for exploring the data, and full access to the underlying source data available at www.govindicators.org. Worldwide Governance Indicators, 2025 Revision, World Bank (www.govindicators.org), Accessed on 12/15/2025. And: World Bank (2025). “The Worldwide Governance Indicators: Revised Methodology for Measuring Governance Using Perception Data.” Washington, DC: World Bank Group. The 2025 revision builds upon the original WGI methodology which was introduced in 1999 and summarized by the following paper: Kaufmann, Daniel & Aart C. Kraay. (2024). “The Worldwide Governance Indicators: Methodology and 2024 Update.” Policy Research Working Paper. Washington, DC: World Bank Group., date accessed: 2026-06-03, date published: 2025. That qualification controls interpretation: coverage gaps, aggregation choices, revisions, reporting incentives, and the distance between the series and the Driver concept can all limit inference. Where the exact requested metric lacked sufficient observations and a documented fallback was used, the fallback remains visibly identified in the research artifact rather than presented as an exact measurement.

Separate context comes from [Universal Declaration of Human Rights](https://www.un.org/en/about-us/universal-declaration-of-human-rights). It supports domain vocabulary and setting, not hidden annual values or a manufactured coefficient. Evidence is strongest for observed proxy movement, weaker for the claim that Digital Censorship caused an outcome, and weakest where the proxy captures a neighboring facet. Coincidence or association remains a hypothesis until analysis addresses confounding, reverse causality, selection, and measurement error.

## **Effect on welfare indicators**

The following indicators are the concrete welfare-sensitive endpoints against which Digital Censorship should be evaluated. Their links resolve to existing Factrail Indicator records. They do not create a causal graph edge by themselves: an active DriverIndicatorImpact is warranted only when a source supports the specific pathway, direction, lag, and scope.

- [WJP Rule of Law Index](/indicators/justice-wjp-rule-of-law-index-global). For Digital Censorship, expected first-order direction is **lower**. The channel to test is changes in civic freedoms, equal treatment, due process, and effective remedies. Horizon: months to several years. Confidence: medium. A same-period correlation would not prove this impact.
- [Internet penetration](/indicators/tech-internet-penetration-global). For Digital Censorship, expected first-order direction is **lower**. The channel to test is network shutdowns, affordability, platform access, censorship, and willingness to participate online. Horizon: months to years. Confidence: low to medium and conditional on access restrictions. A same-period correlation would not prove this impact.
- [People internally displaced by conflict and violence](/indicators/sec-conflict-violence-internal-displacement). For Digital Censorship, expected first-order direction is **higher**. The channel to test is escalation from exclusion or repression into direct threats, communal violence, and flight. Horizon: days to years. Confidence: low for information effects alone and higher when coercion escalates. A same-period correlation would not prove this impact.

First-order effects for Digital Censorship follow directly through named channels. Second-order effects depend on responses by governments, markets, organizations, communities, or households and may arrive later or with the opposite sign. Missing coefficients must not be replaced with round numbers; the relation stays draft or absent until evidence narrows the range.

## **Interactions and contested points**

Digital Censorship can reinforce or offset other Drivers in rights, social stability, and the information environment, including Disinformation Prevalence, but shared timing is not enough to establish an interaction. Reinforcement is plausible when both factors act on the same bottleneck, exposed population, institutional rule, price, or infrastructure network. Offsetting is plausible when adaptation, substitution, redundancy, legal safeguards, fiscal support, or technological learning weakens the pathway. Conditional interaction is the default where one Driver changes the exposure or response to another rather than moving the welfare indicator independently.

The strongest skeptical interpretation is that the selected series is too remote from Digital Censorship, that observed movement is driven by omitted factors, and that the broad causal narrative cannot identify a stable sign or magnitude across countries and periods. That objection is especially important when multiple Drivers share the same public dataset or when the series measures an outcome adjacent to the concept rather than the concept itself. The strongest competing interpretation is that a transparent proxy, clear boundaries, and explicit uncertainty are still more useful than an unmeasured label, provided the proxy is never mistaken for proof and the graph does not fabricate unsupported edges.

The record therefore separates what is established from what remains contested. Established here are the English canonical identity, the sourced 2020-2024 proxy observations, the disclosed normalization, and the distinction between Driver movement and welfare impact. Plausible but not automatically established are the indicator pathways listed above. Unresolved are universal causal magnitudes, responsibility, forecasts, and any claim outside the source scope. Publication or verification must not erase those boundaries; later evidence should update the relations and content through the governed pipeline rather than silently rewriting the history of Digital Censorship.

## **What is documented**
Turkey supplies multiple dated mechanisms for digital censorship. Wikipedia was blocked nationwide in April 2017. On 26 December 2019, Turkey’s Constitutional Court found that the blocking violated freedom of expression, and the Wikimedia Foundation described the ruling as ending a two-and-a-half-year access restriction. Law No. 7253, published in the Official Gazette on 31 July 2020, created new obligations for large social-network providers, including local representation and mechanisms connected to content removal and access blocking. ARTICLE 19’s legal analysis described the amendment as adding removal of content to the existing blocking architecture and setting escalating sanctions for non-compliance. In October 2022, legislation commonly called the disinformation law created a criminal offence concerning dissemination of allegedly false or misleading information; Venice Commission opinion CDL-AD(2022)034 assessed the provision and warned of its implications for freedom of expression. Saudi Arabia supplies a punitive online-expression case through Salma al-Shehab, whose 2022 sentence was based on peaceful Twitter activity according to Human Rights Watch and Amnesty International.

## **Who moves this driver**
[**Recep Tayyip Erdoğan**](/people/recep-tayyip-erdogan) is linked through the Wikipedia block, the 2020 social-media law and the 2022 disinformation law. [**Mohammed bin Salman**](/people/mohammed-bin-salman) is linked through the Salma al-Shehab prosecution. The Turkish facts distinguish access blocking, platform regulation and criminal liability rather than collapsing them into one censorship event.

## **Welfare pathways**
Digital censorship can reduce access to reference material, inhibit political discussion and increase compliance costs for platforms and publishers. Criminal exposure for ambiguous speech can produce self-censorship before any formal case is filed. Nationwide blocking also affects students, researchers and businesses that rely on open information resources. At the same time, internet regulation can pursue legitimate aims such as privacy, unlawful-content enforcement and user remedies, so welfare analysis depends on legal precision, proportionality and safeguards rather than on the existence of regulation alone.

## **What is uncertain**
The 2020 Turkish law contains several obligations with different purposes, and not every requirement is itself censorship. The relevant concern rests on the powers and sanctions that can pressure removal or restrict access, as documented by the cited legal analyses. The Wikipedia case has a later constitutional judgment that clarifies the rights violation, while the 2022 criminal provision remains sensitive to implementation and judicial interpretation. The Saudi case documents an extreme sentence but does not establish that all users face the same enforcement risk.

## Dated facts

- 2024: Journalist Mech Dara is charged with inciting social unrest; a prison-uniform apology names Hun Manet. https://factrail.ai/facts/kh-mech-dara-2024. Sources: https://www.reuters.com/world/asia-pacific/cambodia-arrests-journalist-known-scam-compound-reporting-2024-10-01/, https://pressocm.gov.kh/.
- 2024: Parliament approves the RTVS-to-STVR public-broadcaster reconstruction. https://factrail.ai/facts/sk-fico-stvr-2024. Sources: https://www.reuters.com/world/europe/slovak-parliament-approves-governments-contested-public-broadcaster-revamp-2024-06-20/, https://www.nrsr.sk/.
- 2022: Turkey adopts criminal provision on false or misleading information. https://factrail.ai/facts/tr-disinformation-law-2022. Sources: https://www.coe.int/en/web/venice-commission/-/cdl-ad-2022-034-e, https://www.article19.org/resources/regulation-of-social-media-platforms-in-turkey-internet-law/.
- 2022: Saudi appeals court gives Salma al-Shehab a 34-year sentence over Twitter activity. https://factrail.ai/facts/sa-salma-al-shehab-sentence-2022. Sources: https://www.hrw.org/news/2022/08/19/saudi-arabia-woman-sentenced-34-years-tweets.
- 2020: Turkey adopts social-media platform obligations. https://factrail.ai/facts/tr-social-media-law-2020. Sources: https://www.article19.org/resources/regulation-of-social-media-platforms-in-turkey-internet-law/, https://www.erdem-erdem.av.tr/en/insights/regulations-on-social-media-and-internet-contents-through-law-no-7253.
- 2017: Turkey blocks access to Wikipedia. https://factrail.ai/facts/tr-wikipedia-block-2017. Sources: https://www.anayasa.gov.tr/en/news/individual-application/press-release-concerning-the-judgment-finding-a-violation-of-the-freedom-of-expression-due-to-blocking-of-access-to-wikipedia/, https://wikimediafoundation.org/news/2019/12/26/turkish-constitutional-court-rules-that-the-two-and-a-half-year-block-of-wikipedia-is-unconstitutional/.
- 2017: Government orders internet shutdown in North-West and South-West Anglophone regions. https://factrail.ai/facts/cm-anglophone-shutdown-2017. Sources: https://www.accessnow.org/press-release/access-now-files-supporting-intervention-in-renewed-legal-challenge-to-internet-shutdown-in-cameroon/, https://www.ohchr.org/sites/default/files/2022-06/ANNEXE.%20Summary%20report%20final.November%202021.pdf.

## Methodology

https://factrail.ai/methodology

Canonical: https://factrail.ai/drivers/digital-censorship