Modelling temporal dynamics: does internet use fuel anti-government protests?

ABSTRACT

The past three decades have witnessed a rapid global uptake of digital media. Does an increase in internet access lead to more anti-government protests globally, in both democracies and non-democracies? Has the role of the internet changed over time from benefiting the opposition to benefiting the regime? We use time-series cross-national data and negative binomial regressions to model protest data in 151 countries from 1990 to 2020. By leveraging change in the development of digital information globally, the results show that increases in internet penetration and mobile cellular subscription rates increase the number of anti-government protests in non-democracies in the period from 1990 to 2010, but not among a subsample of democracies. After 2010, increases in internet penetration rates did not affect the number of protests in either democracies or non-democracies. The use of cellular internet continues to have a small positive effect on protest frequency after 2010. We also test the government's internet censorship efforts as a mechanism for decreasing information access and diminishing mobilization. Results suggest authoritarian regimes modified their strategies over time, more effectively using information and communications technologies (ICTs) to quell anti-government protests using digital repression and information control consistent with the theory of informational autocracy.

KEYWORDS:

• Social media
• internet use
• authoritarian regimes
• political protest
• non-democracies
• democracies

Introduction

Does an uptake in digital media lead to more anti-government protests globally, in both democracies and non-democracies? The past three decades have witnessed the rapid expansion of internet use around the world. In 2019 more than 56% of the global population used the internet, according to the World Bank (see Figure 1). Social media use has also grown exponentially. As of January 2022, 62.5% of the world’s population were internet users (4.95 billion), 58.4% used social media platforms (4.62 billion), and 67% had a mobile phone (5.31 billion).¹ Despite these trends in communication technologies, a global divide in internet freedom has emerged between autocracies and democracies. Freedom House’s Freedom of the Net report finds global internet freedom declined for a twelfth consecutive year in 2022, as authoritarian governments sought to wall off the open internet into a patchwork of repressive areas.² Today more governments are censoring what information people can access, blocking foreign websites, collecting personal data, strategically using digital platforms for social control, and centralizing their countries’ technical infrastructure. This study focuses on the impact of internet access rates on political protests, comparing democracies and non-democracies and measuring change over the past three decades.

Figure 1. Changes in internet penetration rates over time (World Bank).

Source: ITU World Telecommunication/ICT Indicators Database, World Bank.

The growing literature emphasizes digital media’s potential for enabling social movements and democratic protests³ by lowering the barriers to collective action, facilitating information dissemination, and decreasing coordination costs.⁴ As well as being tools for democratic activists and protesters, digital information technologies are increasingly a means of repression, censorship, surveillance, and disinformation by authoritarian governments.⁵

There is limited cross-national longitudinal evidence on how the uptake in internet penetration affects political protests.⁶ Most studies on the effects of digital media on political participation focus on a single country or region and use a cross-sectional (that is, single time period) research design.⁷ Out of 36 studies analysed by Boulianne, only six studies used a panel research design (that is, repeated observations of the same case), and the other two employed a cross-national design (that is, comparing across countries). A recent systematic review of 496 articles of causal and correlational evidence on the relationship between digital media and different political variables in Nature Human Behavior⁸ includes just two studies using causal inference research designs to provide evidence for digital media effects on political protest participation.⁹ In addition to these, to the best of our knowledge, two more studies using causal inference research designs analyse the effect of digital media on mobilization and protests specifically.¹⁰

Causal inference involves the identification of a cause of a specific phenomenon by establishing covariation of cause and effect, time-order with the cause preceding the effect, and the elimination of plausible alternative causes.¹¹ Commonly used methods include statistical matching, panel data, differences-in-differences, instrumental variables, and regression discontinuity designs that help ensure that the explanatory variable causes a change in the outcome variable. This study uses panel data with country and year random effects that account for unobserved (that is, unmeasured) factors. Panel designs with unit/country and time-fixed (or random) effects are common causal inference techniques used with observational data that allow measurement of change in both the explanatory variable (internet access) and the outcome variable (political protests) over time between the treatment and control groups.¹²

Additionally, the majority of studies on digital media’s impact on political participation focus on democracies.¹³ Boulianne’s meta-analysis found that research in democratic countries dominates compared to non-democracies or young democracies.¹⁴ Others in their search were not able to identify any causal inference studies in authoritarian regimes in Africa or the Middle East.¹⁵

We seek to improve our understanding of the effects of internet and digital media use on political protest in both democracies and non-democracies by employing time series methods and using cross-national panel data spanning three decades. We analyse the effects of internet use and mobile connectivity across time and types of political regimes (that is, time and space).

Several cross-national studies highlight that the impact of internet use and digital media on anti-government protests varies based on political regime. In their innovative book, Weidmann & Rød ¹⁶ use city-year and city-week observations and random effects multilevel models. The results show that between 2005 and 2012 higher internet traffic rates in cities (1564 cities in 61 countries) are associated with lower rates of protest occurrence in autocracies. In contrast, Ruijgrok ¹⁷ covers protests in more than 130 countries between 1990 and 2013, finding that higher internet penetration rates in a country increase the expected number of protests in autocracies but not democracies.

The divergent results can be attributed to the use of different measures of internet use, as well as different samples of countries, units of analysis (city vs. country), and time frames (2005–2012 vs. 1990–2013). Weidmann & Rød use internet traffic, measured as the number of internet addresses (Internet Protocols) in a geographic location as a proxy for internet penetration.¹⁸ Ruijgrok measures internet penetration as the percentage of the population using the internet within the last 3 months.¹⁹ While internet traffic provides more granular real-time data, it measures the availability of internet infrastructure, not individual internet users. Further, reliable data on internet traffic is not available for countries globally over three decades. Studies show that measures of individual internet adoption rates outperform measures of internet deployment or broadband infrastructure.²⁰ Mossberger et al. also find that higher internet adoption rates lead to improved local economic outcomes but not broadband infrastructure.²¹ As our interest is to evaluate the relationship between internet penetration and political protests and mobilization using annual data for countries, we follow Ruijgrok ²² in measuring the percentage of the population online using World Bank data.

The puzzle emerging from the literature is whether growing internet access does or does not lead to more anti-government protests. Given the relatively few studies of non-democracies and mixed previous results, this study takes a closer look at the temporal dynamics of the role of information and communications technologies (ICTs) in protests around the globe, comparing democracies to non-democracies. Building on Ruijgrok's research, we investigate the relationship between changes in internet penetration and the number of anti-government protests cross-nationally over the past three decades. Change over time is key to solving this puzzle. We propose that initially, internet penetration leads to more protests as it spreads faster than authoritarian regimes can control. But as regimes catch up in digital sophistication, the “repression” side of the technology dominates. Internet access is expected to have a larger effect on protests in authoritarian than democratic regimes.

Several contributions extend Ruijgrok’s analysis. First, we use two alternative measures of internet access rates. The first is mobile cellular subscriptions per 100 inhabitants (including broadband and non-broadband subscriptions) from the World Telecommunication/ICT Indicators Database. During protests, protesters access digital information primarily through mobile cellular networks, not wired broadband. For consistency with past research, we also analyse internet penetration rates, defined as the number of individuals using the internet in the past 3 months as a percentage of the population from the World Telecommunication/ICT Indicators Database (see Figure 1).

Second, we extend the time frame analysed by Ruijgrok through 2020. In 2013, which is the last year included in Ruijgrok's analysis, less than 36% of the world’s population used the internet, on average, with lower rates in authoritarian regimes than in democracies. In 2020, it is estimated that 60% of the world population were internet users. This is a significant increase from 2013. This rapid growth in digital information access suggests that updated analyses are needed. Taking advantage of the 30-year time series, we also analyse the period before and after the Arab Spring protests and the widespread use of social media (2010) separately, to detect potential changes over time. Given the rapid adoption of the internet globally, the three decade analysis and the evaluation of the change in the effect size over time are the main contributions of this study. Third, departing from earlier work, we use the Mass Mobilization dataset that covers anti-government protests in more than 150 countries to quantify the number of protests.²³

We use time-series cross-national data and negative binomial regressions to model protest data in 151 countries from 1990 to 2020, as internet access rates increased. As a measure of regime type, we use democracy indices from the Varieties of Democracies project. Results corroborate the finding from Ruijgrok and show that an increase in internet penetration rates increases the number of protests in non-democracies in the period from 1990 to 2010, but not among a subsample of consolidated democracies (V-Dem electoral democracy index > 0.7). After 2010, increases in internet penetration rates do not affect the number of protests in either democracies or non-democracies. Increases in mobile internet subscriptions continue to have a positive effect on the number of protests in autocracies in the later period. This suggests authoritarian regimes modified their strategies over time, more effectively using ICTs to quell anti-government protests using censorship and digital repression,²⁴ in line with the theory of informational autocracy.²⁵

Considering temporal dynamics: digital media and democratic protests

Consistent with Deibert and Rohozinski,²⁶ the recent literature suggests that ICTs can benefit either political elites or opposition movements, depending on a variety of factors such as government censoring capacity and disinformation campaigns, the strength of civil society, the nature of the regime ²⁷ and temporal dynamics of the protests.²⁸ Few studies have empirically evaluated temporal dynamics in the relationship between internet penetration and political protests cross-nationally over an extended time period. This study takes a closer look at the temporal dynamics of the role of the internet and ICTs in protests around the globe.

The Arab Spring uprisings in 2010–2012 popularized the idea of the democratizing power of the internet and digital media for political organization and coordination of protests. A decade later, Starlink satellite broadband terminals may help galvanize anti-government protests in Iran.²⁹ The idea in each case is that the new technologies promote development and democracy, similar to the revolutionary impacts of the invention of the telegraph, telephone, and printing press.³⁰ Each of these new technologies transformed society and structurally altered how information was produced, circulated, and consumed.

Earlier studies, especially those focused on the Arab Spring, argued that digital media globalizes local struggles, makes collective action possible, and is a necessary condition for regime change.³¹ Others pointed out that new technologies can be used by authoritarian states to distract the public from politics and, instead, provide more entertainment.³² Popular bloggers, for example, can be turned into state agents who promote the regime and spread pro-state propaganda messages. As Morozov states “[s]ome bloggers are simply too useful to get rid of”.³³ Others argued that considering new technology as a dichotomy – liberation technology for activists or technology of information control for the state – is flawed. Rather, we should consider them as “complicated and continuously evolving manifestations of social forces at a particular time and place”.³⁴ By studying a 30-year time frame this study seeks to parse out these evolving forces over time.

In her 2020 book, Vanderhill emphasizes that ICTs are neither necessary nor sufficient for regime overthrow and democratization.³⁵ Nevertheless, digital technology can aid the opposition in their collective action through several mechanisms. ICTs can be used in challenging state information monopoly by providing alternative channels for the uncensored information flow. Often social media and messaging apps are the only political communication channels available for the opposition as authoritarian governments tend to control other media including radio, newspapers, and television. Second, aside from facilitating the spread of uncensored information, ICTs can facilitate the diffusion of democratic ideas and values.³⁶ Third, digital technology reduces the barriers to collective action and mobilization.³⁷ Lastly, ICTs enable people to increase international pressure on elites by spreading information about protests, violence from the state, and human rights abuses.³⁸

At the same time, authoritarian governments use digital media as a part of digital repression to prevent or stop collective action. Earl et al. argue that traditional state repression has evolved into digital repression where the goals are similar in terms of maintaining power and regime survival, but they are reached through new, different means of using digital tools.³⁹ Literature on authoritarian politics shows the significance of information control and propaganda for regime stability and survival.⁴⁰ Among the information control tools⁴¹ available for autocrats are control of the content of the internet (censorship, filtering, blocking of websites), surveillance, and cyberattacks. Some countries such as China and Russia are highly efficient in their information control models. The governments in these countries are able to monitor and restrict the flow of information that may endanger the survival of the regime via regulations, ownership of infrastructure, and self-censorship induced by surveillance and repression.⁴² The 2022 Freedom on the Net report shows rising digital repression in many countries mirrored broader crackdowns on human rights, especially in Russia, Myanmar, Libya, and Sudan, which experienced the world’s sharpest declines in internet freedom.⁴³

Have ICTs always been tools that benefited both the state (in terms of reducing the chances of protest occurrence) and opposition and activists (in terms of organizing and sustaining protests)? Or has the role of the internet changed over time from benefiting the opposition more to becoming increasingly the tool for the regime?

We test several hypotheses. Due to the internet and digital technology's ability to enable social movements as discussed above, we expect that an increase in internet access rates will lead to an increase in the frequency of anti-government protest occurrence. We expect this effect to vary based on the level of democracy in a country as previous research has shown that the internet and access to digital information and networks help to fuel protests in autocracies but not democracies.⁴⁴

H1: An increase in internet access rates is associated with an increase in the number of anti-government protests between 1990 and 2020. The effect of internet penetration is conditional on the level of democracy because autocracies face fewer constraints in repressing and controlling the internet. Thus as access rates rise more democratic countries will have fewer anti-government protests than autocratic countries.

Second, we test whether internet censorship efforts in authoritarian countries affect the number of anti-government protests. If information censorship efforts by authoritarian regimes help to prevent or minimize the occurrence of protests, there should be a decline in the number of protests as regime information censorship grows even if internet access rates are expanding.

H2: As information censorship efforts increase in authoritarian states, the effect of internet access on the number of protests will decrease.

Lastly, we argue that a wave of large anti-government protests enabled via digital and social media, which includes but is not limited to the 2009 protests in Iran, the 2009 and 2010 protests in Kazakhstan, the Arab Spring uprisings, and the 2011–2012 protests in Russia, was a pivotal moment that intensified the state’s desire to control information and information communication technology to prevent future protests from occurring. As popular uprisings are a common means of ending dictatorships,⁴⁵ authoritarian regimes should be interested in more information control that can minimize the occurrence of collective action. Therefore, as authoritarian governments increasingly employ censoring and disinformation campaigns, we expect a declining association between internet access rates and the frequency of protests over time.

We identify the year 2010 as an average cut-off point cross-nationally (we also test 2009 and 2011 as alternative cutoff points with the results reported in the Appendix; the same results are found). In addition to the intensifying digital information control after 2010, the period of the 2010s was also a period when social media platforms became increasingly popular with more individuals joining them.⁴⁶

H3: The effect of expanding internet access rates on the number of anti-government protests decreases over time, especially in non-democracies. Pre-2010, an increase in internet access leads to more protests as the internet spread faster than authoritarian regimes could control. Post-2010, once authoritarian regimes master the “repression” side of the technology, the relationship between internet access and anti-government protests diminishes.

The pivotal moment can be illustrated with the example of Russia. In Russia, an initiator of the quarter of all cyber-enabled information operations,⁴⁷ domestic internet access was relatively unregulated and free up until 2012.⁴⁸ Zassoursky argues that when the internet developed, the state was more interested in how it would affect the economy, not politics.⁴⁹ This lack of control of digital information in the 2000s was corrected by the control of traditional media (television, radio, print) via ownership, subsidy, or political favors.⁵⁰

The Russian state began a more aggressive campaign to censor the internet after the 2011–2012 protests and demonstrations at Bolotnaya Square in Moscow, protests that were digitally enabled via Facebook.⁵¹ Oates argues that ICTs played a vital part in informing the public and organizing the protests on the ground.⁵² The passage of the 2019 federal “Sovereign Internet” law marked another stage of the state’s desire to limit the flow of online information and isolate the domestic internet from foreign information. Although the implementation of the law is costly, it signals the desire of the state to consolidate its control over ICTs. Today, Russia has one of the most restrictive internet environments in the world, scoring 30 out of 100 on the Freedom of the Net report by Freedom House.⁵³ After Russia’s military invasion of the sovereign country of Ukraine in 2022, the Kremlin intensified its efforts to suppress domestic dissents and closed the country’s remaining independent media outlets.⁵⁴

Data and methodology

To test whether an increase in internet access rates impacts the number of protests in democracies and non-democracies, we model the effects of internet access rates (two measures) on the annual number of protests in 151 countries from 1990 to 2020. The dependent variable is the count of the number of anti-government protests from the Mass Mobilization Protest Data dataset ⁵⁵ that records individual protests against governments. Anti-government protests are defined as events where 50 or more protesters publicly demonstrate against the government. The data on events are collected via the LexisNexis news database, one of the largest online news databases, that includes major world publications, such as the New York Times, Washington Post, Christian Science Monitor, Times of London, Jerusalem Post, and regional newspapers. While other protest datasets, such as the Mass Mobilization in Autocracies Database, provide related data, these data are only available for a shorter time period.⁵⁶ Given our focus on temporal dynamics, a 30-year time frame is an important element of our study.

The main independent variable is the rate of internet access measured by mobile cellular subscriptions or alternatively internet penetration rates. The first indicator measures the number of mobile cellular telephone subscriptions – that is, subscriptions to a public switched telephone network using cellular technology per 100 people, including both analog and digital cellular systems (IMT-2000 3G and 4G subscriptions); 3G and 4G are wireless mobile telecommunications technologies that provide mobile Internet access. Data on mobile cellular subscriptions per 100 inhabitants is from the World Telecommunication/ICT Indicators Database. The second indicator is the percentage of individuals using the internet as shown in the global maps (Figure 1) based on the data from the World Bank.⁵⁷ As defined by the World Bank, internet users are individuals who have used the internet (from any location) in the last 3 months including via a computer, mobile phone, games machine, etc. This measure is non-stationary. We take a natural log of the internet penetration rate to make the measure stationary and include the logged variable in the statistical models reported below. See Table A1 in the Appendix for panel unit-root tests which indicate that some panels are stationary after the log-transformation. The use of these two indicators allows us to better measure internet access globally as developing nations primarily rely on mobile cellular networks.

Over the past three decades, the rapid expansion of digital media has fundamentally transformed how information is consumed and spread. Figure 2 shows the dramatic rise in internet adoption rates cross-nationally over the past three decades broken down for non-democracies (left panel) and democracies (right panel). Worldwide internet use increased from 6.7% in 2000 to 56.7% in 2019, according to data from the World Bank. Beyond logging our measure of internet penetration, to account for these time trends, we include a covariate for time, time-squared, and time-cubed in our multivariate regression models.

Figure 2. Internet penetration rates between 1990 and 2020 by regime type.

To analyse whether an increase in internet access rates affects the number of protests in democracies and non-democracies differently, we use the electoral democracy index from the Varieties of Democracy (V-Dem) project.⁵⁸ Higher values of the index reflect higher democracy levels. We are interested in the interaction effect between the levels of internet access and the levels of democracy in a country. We also created a binary version of the index to model and compare the effects between democracies and non-democracies. If the value of the electoral democracy index is greater than 0.5, the country-years are coded as democratic, otherwise as autocratic. Table 1 reports summary statistics for the outcome and predictor variables in the statistical models.

Table 1. Summary statistics for democracies and non-democracies.

	N	Mean	SD	Min	Max
Democracies
Annual number of protests (1990–2020)	2300	3.543	6.692	0	143
Internet penetration rate (internet users)	2348	29.947	31.138	0	98.822
Internet penetration rate (internet users, logged)	2160	2.248	2.392	−9.103	4.593
Mobile-cellular subscriptions, per 100 inhabitants (including multiple lines)	2348	63.082	52.775	0	187.417
Total population (logged)	2348	16.064	1.542	12.755	21.025
GDP per capita (logged)	2295	8.653	1.469	4.956	11.725
Percent of population ages 20–24	2348	8.026	1.456	3.721	11.633
Internet censorship effort	2108	1.383	.689	−1.443	2.1
Non-democracies
Annual number of protests (1990–2020)	2347	2.864	6.123	0	91
Internet penetration rate (internet users)	2395	14.135	23.307	0	100
Internet penetration rate (internet users, logged)	1908	0.903	2.952	−10.953	4.605
Mobile-cellular subscription, per 100 inhabitants (including multiple lines)	2395	40.706	51.310	0	200
Total population (logged)	2383	16.198	1.525	12.731	21.061
GDP per capita (logged)	2289	7.386	1.431	3.127	11.351
Percent of population ages 20–24	2386	8.985	1.112	4.533	12.544
Internet censorship effort	1936	−0.446	1.359	−4.025	2.037

Regime information censorship is another mechanism that might contribute to political protests. To measure state censorship of information that might contribute to a decrease in the frequency of anti-government protests, we use an index of government internet censorship efforts available in the V-Dem dataset. The index quantifies the internet censorship attempts including Internet filtering (blocking access to certain websites or browsers), denial-of-service attacks, and partial or total Internet shutdowns. Higher values are associated with unfiltered and unrestricted Internet access. Additionally, we include control variables that are associated with the number of protests at the country level, including total population, GDP per capita, and the percentage of young people (ages 20–24) with data from the World Bank.

Because our dependent variable is a count with overdispersion (variance is greater than the mean, see Table 1 for descriptive statistics), we estimate random-effects negative binomial pooled cross-sectional time series models. We alternatively estimate dynamic panel models by including a one-year lag of the outcome variable. Lagged variables are commonly used in response to endogeneity concerns with observational data, by modelling short-term (year over year) changes in the outcome variable, or, in our case, the number of anti-government protests in a country. Dynamic panel models also address autocorrelation concerns.

Part 1 results: times series cross-national data predicting protests

Table 2 presents the results of the negative binomial models for the annual number of anti-government protests for 1990–2020, using the two measures of internet access rates as main predictors. Columns 1–2 present estimates from models that include a variable for the percentage of the population that are internet users (internet penetration). Columns 3–4 present estimates from similar statistical models, but with internet access measured as the number of mobile cellular subscriptions per 100 inhabitants. Table 2 columns 1 and 3 report the base model estimates with the main explanatory variables and the interaction between different internet access rate measures and regime types. Columns 2 and 4 report estimates for models that include the lagged number of anti-government protests, as the number of protests in a year might be correlated with the number of protests in the following year (autocorrelation). The results are robust to changes in the set of covariates (adding one year lag of the outcome variable) and model assumptions (using fixed effects). The negatively signed and statistically significant interaction terms in all models suggest that more democratic countries with higher internet penetration rates have fewer anti-government protests, all else equal. These results are consistent with Ruijgrok’s research.⁵⁹

Table 2. Negative binomial random effects overdispersion model estimates for the annual number of anti-government protests 1990–2020, cross-nationally.

	(1)	(2)	(3)	(4)
	Number of protests	Number of protests	Number of protests	Number of protests
Number of protests at t − 1		0.015***		0.015***
		(0.001)		(0.001)
Electoral democracy index	0.906***	0.948***	1.445***	1.439***
	(0.149)	(0.149)	(0.142)	(0.145)
Mobile cellular subscriptions			0.005***	0.005***
			(0.001)	(0.001)
Mobile cellular subscriptions # Electoral democracy index			−0.011***	−0.010***
			(0.001)	(0.001)
Internet penetration rate (logged)	0.116***	0.102***
	(0.021)	(0.021)
Internet penetration rate (logged) # Electoral democracy index	−0.107**	−0.105**
	(0.034)	(0.035)
Total population (logged)	0.227***	0.215***	0.217***	0.206***
	(0.021)	(0.021)	(0.020)	(0.020)
GDP (logged)	−0.131***	−0.112***	−0.075**	−0.067*
	(0.033)	(0.033)	(0.026)	(0.026)
Percent of population ages 20–24	−0.073***	−0.067**	−0.072***	−0.066**
	(0.022)	(0.022)	(0.020)	(0.021)
Year	−0.190***	−0.163***	−0.104***	−0.091***
	(0.045)	(0.048)	(0.022)	(0.025)
Year squared	0.012***	0.011***	0.009***	0.008***
	(0.002)	(0.003)	(0.002)	(0.002)
Year cubed	−0.000***	−0.000***	−0.000***	−0.000***
	(0.000)	(0.000)	(0.000)	(0.000)
Constant	−1.930**	−2.140**	−3.123***	−3.126***
	(0.660)	(0.666)	(0.447)	(0.460)
Observations	3984	3961	4520	4382
Countries	151	151	151	151

*p < 0.05; **p < 0.01; ***p < 0.001.

Note: Standard errors are in parentheses. The random effects apply to the distribution of the dispersion parameter. The dispersion is the same for all observations in the same panel. It varies randomly from group to group, such that the inverse of one plus the dispersion follows a Beta distribution.

Because the interpretation of the negative binomial model coefficients is challenging, we plot the average marginal effects of internet penetration on the number of anti-government protests based on Models 1 and 3 from Table 2 (see Figure 3). This helps us determine if internet access rates are statistically significant for democracies and autocracies.

Figure 3. Average marginal effects of internet penetration on the number of protests for democracies and autocracies.

Note: Shaded areas represent a 95% confidence interval. The average marginal effects are calculated assuming the random effect is zero.

The left side of Figure 3 plots the average marginal effect of internet penetration on the predicted number of protests. The effect is positive and statistically significant for values of the electoral democracy index from 0 to 0.6 (that is, non-democracies), providing support for H1. For example, for a country with a democracy index of 0.5, a 1% increase in internet penetration is associated with an increase in the number of anti-government protests by 0.23.⁶⁰

For consolidated democracies (index values 0.7–0.9), the effect is insignificant and is not different from zero. Previous research has found digital political activism can be used to pacify the masses and as a substitute for on the ground activism in democracies.⁶¹ Therefore, one possible explanation for the null effect is that it is easier for people to participate online than on the ground.

The right side of Figure 3 shows the average marginal effect of mobile cellular subscriptions on the predicted number of protests. The effect is positive and statistically significant for non-democratic countries with a democracy index between 0 and 0.3. The effect becomes negative for countries with a democracy index between 0.6 and 0.9. Substantively, for an autocracy with a democracy index of 0.1, a 10-point increase in the rates of mobile cellular subscription is associated with an increase in the number of anti-government protests by 0.12.

Overall, the results from Table 2 and Figure 3 suggest that across the three-decade time series, an increase in internet penetration plays a modest but positive role in increasing the number of anti-government protests in non-democracies but not democracies, consistent with H1. In non-democracies, an increase in internet access rates, whether it is measured as internet penetration or mobile cellular subscriptions, leads to, on average, an increase in the number of anti-government protests.

Part 2 results: internet censorship efforts

As discussed above, one way for authoritarian countries to minimize the effects of internet access on protests is to censor the online media space via filtering efforts, distributed denial-of-service (DDoS) attacks, or internet shutdowns. We test whether the internet censorship levels interact with changes in internet access, per H2. Due to a high correlation between the democracy index and internet censorship level variables from V-Dem, we test the interaction on a subsample of non-democratic countries (V-Dem democracy index is less than 0.5). Table 3 provides estimates for the negative binomial random effects overdispersion models for the annual number of anti-government protests from 1990 to 2020 for the subsample of non-democratic countries. The interaction of internet access and internet censorship effort is statistically significant and negatively signed, suggesting as internet access rises along with internet censorship efforts in a country, the number of protests declines. But the substantive effects show little impact of internet censorship effort on protests as internet access rises over time.

Table 3. Negative binomial random effects overdispersion model estimates for the annual number of anti-government protests 1990–2020 for subsamples of democratic and non-democratic countries.

	(1)	(3)
	Number of protests (non-democracies)	Number of protests (non-democracies)
Mobile cellular subscriptions		0.005***
		(0.001)
Internet penetration rate (logged)	0.079*
	(0.035)
Internet censorship effort	0.259***	0.290***
	(0.033)	(0.037)
Mobile cellular subscriptions # Internet censorship effort		−0.001*
		(0.000)
Internet penetration rate (logged) # Internet censorship effort	−0.028**
	(0.009)
Total population (logged)	0.237***	0.238***
	(0.031)	(0.031)
GDP (logged)	−0.106*	−0.100*
	(0.049)	(0.042)
Percent of population ages 20–24	−0.063	−0.039
	(0.032)	(0.033)
Year	−0.475***	−0.178**
	(0.101)	(0.066)
Year squared	0.030***	0.014***
	(0.005)	(0.004)
Year cubed	−0.001***	−0.000***
	(0.000)	(0.000)
Constant	−0.813	−2.851***
	(1.103)	(0.757)
Observations	1773	1854
Countries	103	104

*p < 0.05; **p < 0.01; ***p < 0.001.

Note: Standard errors are in parentheses. The random effects apply to the distribution of the dispersion parameter. The dispersion is the same for all observations in the same panel. It varies randomly from group to group, such that the inverse of one plus the dispersion follows a Beta distribution.

Figure 4 plots the average marginal effects of internet access (for mobile cellular subscriptions left and internet penetration rates rate) on the number of anti-government protests in non-democracies as regime censorship effort increases. The marginal effect of an increase in the internet penetration on the number of protests does not depend on the value of regime censorship for values of censorship between −4 and 1.5 for mobile subscriptions and −4 to ∼0.25 for internet penetration (that is, most of the range of both predictor variables). The marginal effect is positive and statistically significant (internet access is driving more protests) and substantively of the same size for the ranges of censorship efforts specified above. The graphs suggest that direct internet censorship efforts have little effect in quelling anti-government protests in non-democracies over the past three decades.

Figure 4. Average marginal effects of mobile cellular subscriptions and internet penetration on the number of protests, for varying levels of internet censorship efforts for non-democracies.

One potential explanation for this is the ability of opposition actors and protesters to circumvent government censorship online by using virtual private networks (VPNs), Tor browser, or other tools that counteract censorship. This suggests that there are different mechanisms at play with the potential to decrease the number of protests in non-democracies.

The V-Dem index measuring regime internet censoring may not be sufficient to measure information control. In addition to censorship discussed above, digital repression takes the form of information control, and information channelling which has become more prominent.⁶² While control of online content is meant to limit access (via filtering, censoring, restricting, or removing content, launching distributed denial of service (DDoS) attacks, installing deep packet inspection systems, slowing traffic, or shutting down the internet), information channelling, on the other hand, influences information production, dissemination, and consumption. Disinformation campaigns seek to modify public attention to content, its diffusion, and, ultimately, the behaviour of the population. Information channelling is used to bury information challenging the regime or information that calls for collective action by diverting people from useful information and redirecting them.⁶³ Such distractions might be more efficient for information control than filtering and censoring and are clearly being used extensively.⁶⁴

Part 3 results: change in the effects over time

To analyse these changing dynamics, we analyse how the effect of internet access on the number of protests changes over time to test H3. To do this we compare digital media use in early and later time periods. Table 4 presents the results of the random-effects negative binomial pooled cross-sectional times series models comparing two time periods: 1990–2010 (low information control period) and 2011–2020 (high information control period).

Table 4. Negative binomial random effects overdispersion model estimates for the number of protests, by time periods.

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
	Years 1990–2010	Years 1990–2010	Years 2011–2020	Years 2011–2020	Years 1990–2010	Years 1990–2010	Years 2011–2020	Years 2011–2020
Number of protests at t − 1		0.036***		0.007**		0.038***		0.007***
		(0.003)		(0.002)		(0.004)		(0.002)
Electoral democracy index	0.757***	0.868***	2.378***	2.325***	0.310	0.506*	2.075*	1.975*
	(0.189)	(0.190)	(0.642)	(0.638)	(0.216)	(0.207)	(0.868)	(0.864)
Mobile cellular subscriptions	0.004	0.004*	0.010***	0.010***
	(0.002)	(0.002)	(0.003)	(0.002)
Mobile cellular subscriptions # Electoral democracy index	−0.010***	−0.009***	−0.017**	−0.016**
	(0.003)	(0.003)	(0.005)	(0.005)
Internet penetration rate (logged)					0.159***	0.121***	0.183	0.180
					(0.026)	(0.026)	(0.110)	(0.109)
Internet penetration rate (logged) # Electoral democracy index					−0.134**	−0.129**	−0.420	−0.391
					(0.041)	(0.042)	(0.223)	(0.222)
Total population (logged)	0.256***	0.255***	0.309***	0.298***	0.280***	0.263***	0.304***	0.294***
	(0.028)	(0.028)	(0.032)	(0.032)	(0.031)	(0.030)	(0.032)	(0.032)
GDP (logged)	−0.015	−0.054	−0.041	−0.035	−0.125**	−0.127**	0.022	0.017
	(0.035)	(0.035)	(0.051)	(0.050)	(0.046)	(0.045)	(0.067)	(0.066)
Percent of population ages 20–24	−0.070*	−0.052	−0.019	−0.018	−0.074*	−0.061	−0.025	−0.023
	(0.030)	(0.030)	(0.037)	(0.037)	(0.031)	(0.031)	(0.037)	(0.037)
Year	−0.075*	0.028	−16.612***	−16.909***	−0.109	0.004	−16.595***	−16.830***
	(0.038)	(0.048)	(3.402)	(3.410)	(0.069)	(0.073)	(3.406)	(3.412)
Year squared	0.009*	−0.001	0.651***	0.662***	0.005	−0.003	0.652***	0.660***
	(0.004)	(0.005)	(0.130)	(0.130)	(0.005)	(0.006)	(0.130)	(0.130)
Year cubed	−0.000*	−0.000	−0.008***	−0.009***	−0.000	0.000	−0.008***	−0.009***
	(0.000)	(0.000)	(0.002)	(0.002)	(0.000)	(0.000)	(0.002)	(0.002)
Constant	−3.476***	−3.752***	134.369***	137.202***	−2.003*	−2.643**	133.910***	136.336***
	(0.592)	(0.614)	(29.614)	(29.691)	(0.865)	(0.859)	(29.629)	(29.693)
Observations	3036	2899	1484	1483	2500	2478	1484	1483
Countries	151	151	151	151	151	151	151	151

*p < 0.05; **p < 0.01; ***p < 0.001.

Note: Standard errors are in parentheses. The random effects apply to the distribution of the dispersion parameter. The dispersion is the same for all observations in the same panel. It varies randomly from group to group, such that the inverse of one plus the dispersion follows a Beta distribution.

The results reported in columns 1–4 from Table 4 include mobile cellular subscriptions as a measure of internet access. The coefficients for the mobile cellular subscriptions are positive between 1990 and 2010 and statistically significant when including a lag of the dependent variable; countries with higher rates of mobile cellular subscriptions have more anti-government protests year over year, all else equal. Consistent with the previous analysis, the interaction term between rates of mobile subscriptions and the level of democracy is negative and statistically significant, meaning non-democratic countries have more protests as mobile subscription rates increase, all else equal.

In the later period (2011–2020), the results suggest a similar trend: non-democratic countries experience more protests as their rates of mobile subscriptions increase. The interaction terms of mobile subscription rates and the electoral democracy index are negative and statistically significant (columns 3–4 of Table 4).

To better understand these trends and whether they are statistically significant, Figure 5 plots the average marginal effects of mobile cellular subscriptions on the number of protests across the range of non-democratic to democratic countries (X-axis). In the earlier period (graph on the left) when cellular subscriptions were not widespread, the effect of mobile subscriptions is only statistically significant for countries with a democracy index of 0.6 and higher and the direction is negative. In the later period (after 2010) with greater adoption of mobile technology globally, in non-democracies with an index score between 0 and 0.45, an increase in mobile cellular subscriptions is associated with a small but statistically significant increase in the number of anti-government protests. For countries with an index score above 0.45, the effect of mobile cellular subscriptions on the number of protests is not statistically significantly different from zero at the 0.05 level given the data and model.

Figure 5. Average marginal effects of mobile cellular subscriptions on the number of protests, by time periods (pre and post-2010).

An explanation for the null effect for non-democracies before 2010 is that mobile phones were not in widespread use at this time in non-democratic countries (an average of 16.8% in non-democracies compared with 37.2% in democracies during 1990–2010). A positive effect after 2010 is consistent with the Arab Spring protests where mobile phones were used extensively for protest organization.⁶⁵ Despite digital repression, an increase in digital connectivity continues to lead to modestly more protests in autocracies.

A different pattern emerges when we look at the effects of internet penetration measured by internet users over time. In the earlier period (1990–2010), an increase in rates of internet penetration (that is, internet users) is associated with an increase in the number of anti-government protests, all else equal (columns 5–6 of Table 4). The interaction between internet penetration rates and the level of democracy in a country is negative and statistically significant, suggesting that between 1990 and 2010, an increase in internet penetration led to an increase in the number of protests in non-democratic countries. In the later period between 2011 and 2020 (columns 7–8 of Table 4), the coefficients for the internet penetration rates and the interaction terms are insignificant, pointing to no statistically significant effect of growing internet adoption on the number of protests at the 0.05 level. To better illustrate these trends, Figure 6 plots the average marginal effects of internet penetration on the number of protests.

Figure 6. Average marginal effects of internet penetration on the number of protests, by time periods (pre and post-2010).

As Figure 6 shows (left side), between 1990 and 2010, for countries with a democracy index below 0.7, the average effect of an increase in internet penetration is positive and statistically significant. For example, for a country with a democracy score of 0.5, a 1% increase in internet penetration is associated with an increase in the number of anti-government protests by 0.31. This positive relationship corroborates the findings from Ruijgrok's (2017) study that covered the period 1990–2013.

In the later period (right side of Figure 6), between 2011 and 2020, the average marginal effect of internet penetration is insignificant for all the values of the democracy index. This suggests that after 2010, the effect of an increase in internet penetration on the number of anti-government protests in non-democracies is not statistically different from zero at the 0.05 level given the data and model, suggesting little effect.

This change over time suggests authoritarian governments became more sophisticated in their control of information technology, which perhaps contributed to a decrease in the occurrence of protests. If the earlier studies found that internet use has facilitated the occurrence of protests in authoritarian states,⁶⁶ the inclusion of the most recent data (for example, Russia began its crackdown on the internet domestically, became sophisticated in cyber and information warfare, and started to export censorship technologies oversees approximately in 2012) shows that the internet in authoritarian (and democratic) countries becomes a less effective tool for protest organization. The results suggest that authoritarian states successfully use censorship and information control to manipulate public opinion and ensure regime survival, in line with the literature on information autocrats.⁶⁷

Conclusion

Previous research pointed to conflicting results as to whether growing internet adoption leads to more anti-government protests globally, especially in non-democracies. Given the relatively few causal inference studies of non-democracies, this study takes a closer look at the temporal dynamics of the role of internet access rates on the frequency of protests around the globe between 1990 and 2020, comparing democracies to non-democracies. We argue that the effect of the increases in internet penetration on the frequency of anti-government protests decreases over time. We propose that initially, before 2010, internet penetration leads to more protests as it spreads faster than authoritarian regimes can control. But as regimes catch up in digital sophistication, after 2010, the “repression” side of the technology dominates.

The empirical analysis explores whether internet access rates fuel anti-government protests comparing countries globally using panel data and time series analysis. We build on previous work and argue that the authoritarian states tilted the playing field and became more efficient and effective in controlling the internet and limiting anti-government movements after 2010. The evidence finds as levels of internet penetration increase, the predicted number of protests in non-democracies increases modestly (but not in democracies), on average over 30 years.

More importantly, the effects change over time. While earlier studies found that internet use has facilitated the occurrence of protests in authoritarian states,⁶⁸ the inclusion of the most recent data suggests that the internet in authoritarian regimes may be a less effective tool for the opposition forces after around 2010. Results show that in the earlier years (1990–2010) an increase in internet access measured as internet penetration rates increased the number of anti-government protests in non-democracies. In the later period (2011–2020), an increase in internet penetration rates is not connected to the occurrence of protests in either democracies or non-democracies. Mobile subscriptions still exert a positive impact on more protests in non-democracies, even in the later period (and did so in the earlier period for democracies). This finding suggests that despite government digital repression policies, digitally sophisticated opposition movements can still effectively use the technology for political mobilization. Using panel data (countries) and time series analysis helps mitigate concerns of endogeneity between internet penetration when predicting protests in countries globally.

We also show that internet censorship efforts play a minimal role in the number of anti-government protests in non-democracies. Regardless of the level of censorship, our results indicate that the effect of the increase in internet access rates remains positive and statistically significant. This finding suggests that there might be a different mechanism at play that allows non-democratic regimes to minimize the occurrence of protests in the later years. One possible explanation is the reliance on the information autocrats ⁶⁹ on information manipulation, rather than information filtering and censorship, to convince the public of the government's competency rather than on force and ideology as in the past. Information channelling strategies⁷⁰ include disinformation campaigns and information flooding. Future research might concentrate on testing whether these mechanisms are at play and have the potential to diminish political mobilization.

Disclosure statement

The authors report there are no competing interests to declare.

Data availability statement

Data used in this analysis and statistical code will be available on Harvard’s Dataverse.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Notes on contributors

Daria Kuznetsova

Daria Kuznetsova is a Ph.D. candidate in Political Science at the University of Iowa. Her research lies at the intersection of comparative politics and mass communication and investigates how different political actors exploit, leverage, and seek control of digital communication spaces during contentious politics. Additionally, Daria concentrates on the role social media, disinformation, and propaganda play in authoritarian politics.

Caroline Tolbert

Caroline J. Tolbert is a University Distinguished Professor of Political Science at the University of Iowa. Her work is driven by an interest in strengthening democracies and increasing political participation. With Karen Mossberger and colleagues, she is coauthor of four books on digital media and society including Choosing the Future: Technology and Opportunity in Communities (2021, Oxford) and Digital Citizenship: The Internet, Society and Participation (2008, MIT). Choosing the Future was the winner of the 2022 Goldsmith Award for the best academic book from the Shorenstein Center on the Media, Politics and Public Policy, Harvard.

Notes

1 Kemp, “Digital 2022.”

2 Shahbaz, Funk, and Vesteinsson, “Freedom on the Net 2022: Countering an Authoritarian Overhaul of the Internet.”

3 Manacorda and Tesei, “Liberation Technology”; Zhuravskaya, Petrova, and Enikolopov, “Political Effects of the Internet and Social Media.”

4 Boulianne, “Twenty Years of Digital Media Effects on Civic and Political Participation”; Enikolopov et al., “Social Image, Networks, and Protest Participation”; Ruijgrok, “From the Web to the Streets”; Shirky, Here Comes Everybody.

5 Chadwick, The Hybrid Media System; Morozov, “Liberation Technology”; Vanderhill, Autocracy and Resistance in the Internet Age; Earl, Maher, and Pan, “The Digital Repression of Social Movements, Protest, and Activism.”

6 Lorenz-Spreen et al., “A Systematic Review of Worldwide Causal and Correlational Evidence on Digital Media and Democracy.”

7 Boulianne, “Social Media Use and Participation”; Skoric et al., “Social Media and Citizen Engagement.”

8 Lorenz-Spreen et al., “A Systematic Review of Worldwide Causal and Correlational Evidence on Digital Media and Democracy.”

9 Bekmagambetov et al., “Critical Social Media Information Flows”; Enikolopov et al., “Social Image, Networks, and Protest Participation.”

10 Manacorda and Tesei, “Liberation Technology”; Qin, Strömberg, and Wu, “Social Media, Information Networks, and Protests in China.”

11 Angrist and Pischke, Mostly Harmless Econometrics; Pearl, Causality.

12 Lorenz-Spreen et al., “A Systematic Review of Worldwide Causal and Correlational Evidence on Digital Media and Democracy.”

13 Larson, “Controlling Social Desirability Bias”; Theocharis and Lowe, “Does Facebook Increase Political Participation?”; Tolbert and McNeal, “Unraveling the Effects of the Internet on Political Participation?”; Vaccari and Valeriani, Outside the Bubble.

14 Boulianne, “Social Media Use and Participation.”

15 Lorenz-Spreen et al., “A Systematic Review of Worldwide Causal and Correlational Evidence on Digital Media and Democracy.”

16 Weidmann and Espen Geelmuyden Rød, The Internet and Political Protest in Autocracies.

17 Ruijgrok, “From the Web to the Streets.”

18 Weidmann and Espen Geelmuyden Rød, The Internet and Political Protest in Autocracies.

19 Ruijgrok, “From the Web to the Streets.”

20 Whitacre, Gallardo, and Strover, “Broadband׳s Contribution to Economic Growth in Rural Areas.”

21 Mossberger, Tolbert, and LaCombe, Choosing the Future.

22 Ruijgrok, “From the Web to the Streets.”

23 Clark and Regan, “Mass Mobilization Protest Data V4.”

24 Earl, Maher, and Pan, “The Digital Repression of Social Movements, Protest, and Activism.”

25 Guriev and Treisman, “Informational Autocrats.”

26 Deibert and Rohozinski, “Liberation vs. Control.”

27 Vanderhill, Autocracy and Resistance in the Internet Age.

28 Weidmann and Espen Geelmuyden Rød, The Internet and Political Protest in Autocracies.

29 Washington Post, “Opinion | Can Elon Musk’s Satellites Beat Iranian Internet Blackouts?”

30 Chadwick, The Hybrid Media System.

31 Howard, The Digital Origins of Dictatorship and Democracy.

32 Morozov, The Net Delusion.

33 Ibid., 89.

34 Deibert and Rohozinski, “Liberation vs. Control,” 44.

35 Vanderhill, Autocracy and Resistance in the Internet Age.

36 Ibid.

37 Shirky, Here Comes Everybody.

38 Vanderhill, Autocracy and Resistance in the Internet Age.

39 Earl, Maher, and Pan, “The Digital Repression of Social Movements, Protest, and Activism.”

40 Feldstein, The Rise of Digital Repression; Guriev and Treisman, “Informational Autocrats”; Lankina, Watanabe, and Netesova, “How Russian Media Control, Manipulate, and Leverage Public Discontent”; Roberts, Censored; Vanderhill, Autocracy and Resistance in the Internet Age; Way and Levitsky, “The Dynamics of Autocratic Coercion after the Cold War”; Zhuravskaya, Petrova, and Enikolopov, “Political Effects of the Internet and Social Media.”

41 Earl, Maher, and Pan, “The Digital Repression of Social Movements, Protest, and Activism.”

42 Howells and Henry, “Varieties of Digital Authoritarianism”; Keremoğlu and Weidmann, “How Dictators Control the Internet”; Vendil Pallin, “Internet Control through Ownership.”

43 Shahbaz, Funk, and Vesteinsson, “Freedom on the Net 2022: Countering an Authoritarian Overhaul of the Internet.”

44 Ruijgrok, “From the Web to the Streets.”

45 Feldstein, The Rise of Digital Repression; Geddes, Wright, and Frantz, How Dictatorships Work.

46 Pew Research Center, “Social Media Fact Sheet.”

47 Foote et al., “Cyber Conflict at the Intersection of Information Operations: Cyber-Enabled Information Operations, 2000–2016.”

48 Howells and Henry, “Varieties of Digital Authoritarianism.”

49 Zassoursky, Media and Power in Post-Soviet Russia.

50 Sivetc, “State Regulation of Online Speech in Russia”; Vartanova, “Media Ownership and Concentration in Russia.”

51 Howells and Henry, “Varieties of Digital Authoritarianism.”

52 Oates, Revolution Stalled.

53 Freedom House, “Russia.”

54 Shahbaz, Funk, and Vesteinsson, “Freedom on the Net 2022: Countering an Authoritarian Overhaul of the Internet.”

55 Clark and Regan, “Mass Mobilization Protest Data V4.”

56 Weidmann and Espen Geelmuyden Rød, The Internet and Political Protest in Autocracies.

57 The World Bank, “Individuals Using the Internet (% of Population).”

58 Coppedge et al., “V-Dem Dataset 2021 V11.1.”

59 Ruijgrok, “From the Web to the Streets.”

60 The value is based on the average marginal effects graphs.

61 Lorenz-Spreen et al., “A Systematic Review of Worldwide Causal and Correlational Evidence on Digital Media and Democracy.”

62 Earl, Maher, and Pan, “The Digital Repression of Social Movements, Protest, and Activism.”

63 Ibid.

64 Hobbs and Roberts, “How Sudden Censorship Can Increase Access to Information.”

65 Howard and Hussain, Democracy’s Fourth Wave?

66 Ibid.; Oates, Revolution Stalled; Ruijgrok, “From the Web to the Streets.”

67 Guriev and Treisman, “Informational Autocrats.”

68 Howard and Hussain, Democracy’s Fourth Wave?; Oates, Revolution Stalled.

69 Guriev and Treisman, “Informational Autocrats.”

70 Earl, Maher, and Pan, “The Digital Repression of Social Movements, Protest, and Activism.”