Forecasting An Internet Shutdown: An Exercise With Indicators
The original article was published on April 19, 2017 on Bellingcat.com.
Purpose Of The Research
Internet shutdowns happen — more frequently than we often hear. An impressive 56 countrywide shutdowns were documented in 2016, according to Access Now’s campaign #KeepItOn, which also defines a shutdown as “an intentional disruption of internet or electronic communications, rendering them inaccessible or effectively unusable, for a specific population or within a location, often to exert control over the flow of information.”
A rich amount of academic literature and journalism exists reporting on shutdowns and individual website blocks, and their connection with the real-world political events unfolding. Professionals and activists in that space have conducted an impressive amount of research, and discussion continues on what triggers an internet shutdown, as well as on its effectiveness in serving its intended purpose.
A variety of justifications has been given by governments worldwide for their decision to shut down internet services. A very common trait among them, however, is the stated or implied intention to prevent the spreading of physical unrest (which could be happening for political, religious, economic, or ethnic reasons).
The question that generated this research, therefore, is on whether we could identify, analyze, and monitor indicators of a potential full-scale or selective shutdown — before it happens. The article wants to be an analytical experiment to begin filling that gap, and to suggest some ideas on potential indicators to locate and monitor in the open.
Testing An Analytical Model
Ultimately, the decision to take drastic repressive measures on a country’s internet traffic rests with its government. While human unpredictability is a factor to consider, particularly for leaders whose authority is increasingly undermined by the events, we can try to formulate a model to assess the likelihood of such a decision through some key factors:
- Online mobilization is successful in triggering, increasing, or giving global exposure to civil unrest;
- Opposition parties and leaders successfully use social forums to spread their message — otherwise silenced through traditional media;
- The diaspora community of the country’s citizens — including their foreign-based media, NGOs, and other entities — effectively distributes propaganda language, giving further momentum to the unrest.
- The internet infrastructure is simple, state-owned, poorly developed, and/or easily controllable by the authorities;
- Internet filtering and monitoring technology is known as existing in the country;
- Online surveillance campaigns are suspected and/or proven to be deployed by the authorities.
And, as a negative factor:
- The online community in the country is sizeable, and internationally connected or exposed;
- The “digital economy” covers a considerable portion of the country’s economy, and/or of strategic national assets;
- The country is sensitive to the international pressure due to its geopolitical context.
Case Study: Paraguay
The small, impoverished country of Paraguay is undergoing a (new) constitutional crisis. The current president, Horacio Cartes, leader of a tiny elite which dominates the power nodes of the country and elected in 2013, is attempting to force a constitutional amendment that will allow him to be re-elected in 2018. This is perceived by protesters as a marked U-turn towards authoritarian rule in the country, after nearly three decades of democracy.
In fact, Paraguay was ruled with an iron fist for 35 years by General Alfredo Stroessner, whose dictatorship ended in 1989. Since then, its democracy has been shaky, and plagued by corruption and political unrest.
On March 31, 2017, protesters descended en masse onto the Congress and set the building on fire. Police reaction eventually led to the storming of the headquarters of one of the opposition parties, the PLRA, and the death of a young party member and activist, Rodrigo Quintana.
Unrest was ongoing as of April 5, 2017.
Internet blockages, even when targeted at specific websites, are not necessarily rational decisions based on strategic thought. They are very often knee-jerk reactions by autocratic governments, or military juntas, to the loss of control over the society they rule.
For example, the Thai military junta briefly blocked Facebook in the entire country on 28 May 2014 to repress the spreading of protest content in the wake of the 22 May coup d’état, and to then pressure social media companies for “collaboration”; and Indian state authorities quite frequently switch off the entire connectivity in their states to limit the spreading of “rumors” — essentially, incitement to protests and unrest — like in this recent case in Odisha.
With that in mind, to assess — to the best of our capabilities — how likely it is that a government will take that drastic measure, we need to get a good picture of what is happening in the related online community. In order to do that, and before we dive into the the three categories for the proposed analytical model, it might be important to look at a crucial part of the context: the typical behavior for the local online community.
How Do People Access The Internet?
This is a key question to understand the leverage, and the incentive, a government would have in restricting or disrupting access to internet services. We need to look at the current online community’s behavior:
- Are landline broadband connections the most popular way to access the web, or are mobile services predominant?
- Are broadband services — and the internet in the first place — the most popular means of social communication, or are phone communications (including SMS) more common? This is often true in countries with significantly underdeveloped internet infrastructure, particularly in Sub-Saharan Africa.
In the context of Paraguay:
- Landline connections are minimal. Of the highly popular mobile internet services, two providers dominate the market (see later in this report).
- While phone services remain an important means of communication, due to the lack of development for the country’s internet services, social media usage is pervasive. With a very large portion (approximately 70%) of the population under 40 years of age, the UNICEF had estimated in 2014 that some 30.000 new users were created on social networks in Paraguay every month.
What Are The Main Communication Platforms In The Target Country?
Social media usage varies from country to country. Likewise, instant messaging has developed rapidly over the past decade at least and sees important differences depending on the community in question.
A source often quoted is the World Map of Social Networks by Vincenzo Cosenza. Updated to January 2017 (at the moment of this writing), it shows that:
- The top social platform in Paraguay (as well as the rest of Latin America) is Facebook;
- The runner up is Instagram.
According to the available public data, Twitter comes as a very distant second social platform, as for active users. In March 2017, only a 1.5% of social media usage happened via Twitter — as opposed to the 97% of Facebook. This appears confirmed by other sources, including statistics produced locally that estimate Paraguayan Twitter users at 400.000 as of 2016.
Instant messaging usage by country is more difficult to gauge. WhatsApp remains an ubiquitous mean of communication globally, particularly within developing countries, but Skype continues to be the main messenger in several regions of the world. Telegram is quickly running up, especially in politically-sensitized communities. Finally, we should not ignore the elephants in the room — Facebook Messenger, utilized by one billion of users globally, and WhatsApp, on the same levels of global diffusion. Given how prominent mobile connections are in Paraguay (as we will see later in this case study), it appears reasonable to consider them as probably highly popular means of online communication in the country.
As a matter of fact, WhatsApp’s voice calling feature was briefly blocked in 2015 by the two main ISPs, Tigo and Personal, likely due to the ability of users to circumvent paid phone call plans through the free VoIP service provided by the platform — giving us a hint of the possible size of its Paraguayan community.
Once we have completed a basic evaluation of the internet footprint for the community in question, we’re ready to look into what could constitute the first section in our analytical model: the motive.
What Is The Current Sentiment, And The Trajectory, Of The Online Political Discourse?
Generating The Leads: Twitter Trends
In Paraguay, as we have mentioned, Facebook and Instagram lead the market. Given that both present some significant challenges in the collection, extraction, and manipulation of data for external research, a good practice can be to obtain initial leads from a looser social environment — Twitter.
Several tools are available that allow for a quick capture of the main Twitter trends in a certain geographic area. For this example, we will use a commercial platform (offering a 7-day free trial period), Trendsmap.
Note: we are not endorsing the use of a particular commercial website, application, or software. While we have decided to use the demo version of this particular service in the case study for a few reasons — easy to use, good visualization capabilities — several pure open source services exist offering at least a part of its functionalities at no cost.
A zoom-in around the Paraguayan capital, Asunción, on 2 April reveals the following top trends in the area:
Note that we have decided to display both hashtags and keywords, but filtered out trending users at this stage — we just want to understand what are the most discussed topics at this stage.
Some terms are decisively self-explanatory. #prayforparaguay shows the community’s consternation for the violent incidents, plus by being in English, seems to try to appeal to the international community — which could catch the attention of an authoritarian government; “violencia” is clearly a leading theme too.
What stands out are two hashtags with the potential of becoming slogans. As the “Arab Spring”, the #Euromaidan, and numerous other uprisings in the digital era show us, these are potential key aggregators, and regardless of the fact that their translation into actual urban unrest is still being analyzed and discussed, an increasingly authoritarian and paranoid government could see their rapid propagation as a potential existential threat, triggering an internet shutdown.
In this case, #golpeparlamentario (Parliamentary coup) and #enmienda (amendment) have gained significant traction in Asunción, Paraguay’s capital city, and surrounding areas. The latter refers to the amendment being protested as unconstitutional and overreaching by the oppositions.
#golpeparlamentario, as of April 2, had been used in 26.000 tweets over the previous 7 days; #enmienda in 22.000. These are notable numbers for a small internet community the size of Paraguay’s.
However, a look at the general keywords used in tweets coming out of Paraguay in the same span of time paints a slightly more diverse picture. It reinforces the focus on “enmienda” (used 44.000 times), but it also shows some seemingly more neutral terms related to the probable reporting of the events. “Incendio” and “fuego” (both mean fire), or “bomberos” (firefighters), all most likely relate to the Congress being set on fire by protesters after the amendment was approved.
A look at the trends map also shows a direct mention to the PLRA, which, as we know through initial background research on the Paraguayan crisis, is the opposition party leading the protests. Their headquarters were attacked by the police, leading to the death of one young party member.
At this stage, we have some strong enough leads to move our research onto other social platforms.
Locating Online Aggregation: Facebook
Facebook, being an ubiquitous mean of aggregation for communities, is particularly conducive to the coordination and promotion of political protests. This is a fact widely documented in various settings, from major revolutionary trends such as the “Arab Spring,” to relatively localized and country-specific unrest.
The first logical destination for locating protests coordination are Facebook Events. This product essentially consists of calendar appointments, either public or private, that can be shared with a large number of people — and more importantly, be re-shared by invitees, so to grow the invitees list exponentially. Given these features, they offer a natural platform for coordinating street demonstrations, together with the broad variety of leisure uses the product is mainly intended for.
An initial look via the Facebook search bar of some of the trending hashtags and keywords located on Trendsmap brings up very little relevant content. The current version of the search functionality on Facebook doesn’t necessarily prioritize trending topics, but possibly responds to different dynamics.
We are then going to try a crossover through a simple Google search. The advanced search operator “site:” can be pretty powerful for social platforms, so we’ll simply try:
site:facebook.com/events paraguay AND manifestación (“demonstration” in Spanish)
This leads to a few irrelevant results, and a promising one:
The “Announcement of a demonstration for the defense of the (National) Constitution” turns out to be the social media aggregation and coordination hub for the massive protest event that brought thousands of people on the streets of Asunción on March 30:
The event was created by a Facebook page named Museo de la Corrupción. The page sports some 7.000 followers, and displays slogan imagery used to push protests further (such as the #no hashtag picture, a simple and effective symbol for the protest agenda).
What is critical to understanding potential future developments is that the page in question is actively inviting followers to a new demonstration within a page post, rather than by using a Facebook Event. This new protest is scheduled for April 3, 7pm, in Plaza de Armas (in front of the Congress). Participants are encouraged to bring white flags as a protest symbol, as well as a copy of the Paraguayan constitution:
The protest will eventually unfold peacefully.
As of April 5, 2017, no further demonstrations were being observed as publicly advertised via social networks, but the protest movement appeared mostly united (despite some interesting splinter groups calling the PLRA board members as “traitors”), still unsatisfied by the government response, and probably capable to rapidly schedule and promote new street demos.
Confirming Through Visual Content: Instagram
Instagram can be useful now to further gauge the sentiment and intentions of the opposition movement. One quick look at another main protest hashtag, #AsiNo (“not that way”), surfaces more than 40.000 posts, the most popular ones of which are related to the Paraguayan demonstrations:
We have now completed a basic mapping of the social media footprint for the protest movement, and will be able to both monitor developments and expand research into other linked avenues, later on.
How Can We Monitor The Evolution Of The Online Protest?
As open source researchers know, and as mentioned previously in this case study, social media platforms come with vastly different ease of access to their data. Facebook’s API makes it practically impossible to extract data for external manipulation. Twitter, however, presents interesting opportunities to explore topics by automating the data collection via API. This can be done through a variety of means — most notably custom Python scripts — but for the sake of simplicity, and ease of visualization, this time we will use a platform readily available: TAGS (Twitter Archive Google Spreadsheets), by Martin Hawksey.
TAGS, after a quick set up (you need an active Twitter account to link to every new spreadsheet), allows the researcher to explore tweets around specific hashtags, keywords, or from target users. For this case study, we will focus on one of the four main hashtags located via Trendsmap, and we’ll try to understand its pattern of diffusion.
Pulling data related to #golpeparlamentario, the most popular one, reveals that it has losing traction since its peak in the early days of the protest:
The top 10 tweeters for the hashtag in question, with an aggregate total of 245 related tweets, have seen #golpeparlamentario spiking at the beginning of the examined period — a time span between 6–9 days, unfortunately hard to better define due to limitations inherent to the Twitter API.
However, only a few profiles continue using it — and certainly not the high-diffusion media accounts, such as the ABC ones. A look at the hashtag’s diffusion over the previous three days shows that it is still being used, although at a much slower pace:
The most retweeted tweets using #golpeparlamentario over the previous 2 days focus on the allegedly slow movement that the inquiry on the police’s actions is seeing, particularly in contrast with the fast repression of the protests:
This is the most popular tweet based on retweets:
- Repeated the same analysis on the remaining hashtags;
- Verified that new protest hashtags are not picking up steam within the community;
- Manually researched Facebook, and Instagram, for key indicators of renewed incitement to protest (FB events, content shares such as viral photos or memes, or other ways to trigger street demos);
we can conclude that after the Chamber of Deputies suspended the constitutional amendment on April 5th, the protest lost momentum except for individual activists.
This may relieve the authorities from some immediate pressure, potentially downsizing the perception of an existential threat, and decreasing the likelihood of a shutdown as a reaction.
In the following section we will look at the second element of our analytical model: capability.
How Easy Is It For The Government To Shut Down The Internet?
In a landlocked country, with significantly underdeveloped landline internet infrastructure — a monopoly of the state-owned Corporación Paraguaya de Comunicaciones (Copaco) — Paraguay’s netizens have turned themselves to the exploding mobile market since its liberalization in 2009.
Market penetration for mobile internet services is estimated at about 110% as of 2016, versus a meager 2.6% for fixed broadband. The main mobile providers are Tigo, considered the market leader, Personal (owned by Telecom Argentina), Claro, and Vox.
The relative fragmentation of the widespread mobile internet market could make it slightly less controllable by the central government if the different privately owned operators would oppose some resistance to shutdowns that can have catastrophic financial consequences. However, that is far from removing risks of surveillance, censorship, and even full shutdowns of the networks.
As for surveillance, Paraguay, under the current president Horacio Cartes, has already tried to implement controversial measures. For example, one bill aimed at forcing ISPs into long-term data retention, known as “Pyrawebs”, was eventually rejected by both the Chamber of Deputies and Senate in mid 2015 after a virtual uprising from worldwide activists, and the local internet community.
Around the same time, Paraguay’s government was among the countries cited in a report by the Citizen Lab as likely users of the FinFisher spyware technology, infamously used not only for law enforcement investigations, but also for political espionage.
Online censorship has also been an issue in recent years. In 2012, for example, the two market-dominant mobile operators (Tigo and Personal) blocked access to a local website for almost an entire day, possibly in response to political commentary through user-generated content, and without a judicial order.
Internet shutdowns, the ultimate drastic measure, are still unprecedented in Paraguay.
Overall, and at a high level, we get the picture of an immature internet infrastructure, in a small country, the government of which has already shown concerning signs of at least attempts at surveillance and control, and where technology for targeted censorship exists.
Should the political unrest further degenerate, and risk the political status quo, a full-scale internet block can be considered as probable.
Finally, we need to assess what could constitute deterrent in rolling out an internet block.
What Could Keep The Government From Implementing Or Extending A Shutdown?
Cost is a key factor for the duration of a shutdown, and it’s more difficult to forecast: famously, the five-day Egyptian internet block in 2011 was estimated as having generated direct costs for at least 90 million USD. With the “internet economy” averaging at about 5% of the GDP for developing countries, and a nominal GDP estimate for 2016 at the 111th position in the world ranking for 2016, it’s hard to imagine that the Paraguayan government would, for more than a very short period of time, carelessly inflict a damage to its country’s economy that can nowadays easily amount to hundreds of millions of US dollars.
Additionally, international pressure — particularly from Western countries, where a diaspora community from the country in question could also be pushing local governments to speak up on internet restrictions — can be an important deterrent factor. Linkages to the West, their influence on the local communities, and international pressure are a very nuanced and fluid area that for example Steven Levitsky and Lucan A. Way have researched thoroughly, showing that proximity to Western democracies historically pushed autocratic governments against applying crackdowns on their citizens.
An opposite outcome is observed in a country like Cameroon — where the internet is blocked in an entire region since January 2017 for political reasons — which could separately prompt further analysis on the existence of such links to developed democracies, and their power (or will) to influence the local government’s action.
A more likely action by the authorities in Paraguay could be the selective block of the social media (or instant messaging) platforms where the political discourse happens. This could last longer, but would carry the risk of further frustrating protesters, eliminating an important relief valve, and, ultimately, that of increasing pressure on the government.
For A Final Recap
While accurately predicting an internet shutdown remains an exercise requiring in-depth analysis, monitoring of technical signal, and quality, often proprietary information, open source research can provide important hints to draw a tentative trajectory of a government’s intention to restrict access to the internet.
With a marked decrease in possible motive since early April, though with strong capability, and medium deterrent, an internet shutdown in Paraguay can be still considered as unlikely at the time of publishing this article. Given the enduring negative sentiment towards the government, and the potential for quick aggregation demonstrated by protesters, the motive remains highly volatile, however.
In the process of researching which factors could influence the outcome, we have outlined a tentative research and monitoring model that could be used as an initial template for analysis and forecasting, hopefully facilitating further research, and ultimately, timely preparation to an impending shutdown.