Election-related internet-shutdowns in autocracies and hybrid regimes

ABSTRACT

In autocratic and hybrid regimes, the internet plays an ambiguous role. It simultaneously possesses the quality of a liberation technology and opportunities for digital authoritarianism. Although autocrats have learned to utilize the internet to their advantage, they are aware of its liberation potential. Whenever the political survival of the regime is challenged, they use manipulative tools to avert uncertainty. Shutting down the internet has become one of these tools. This articleanalyzes the occurrence of internet shutdowns during national elections in autocratic and hybrid regimes. We show that internet shutdowns do not occur erratically but rather follow strategic decisions. Although capacity to control access to the internet is crucial, we highlight the importance of incentives to shut down the internet provided by different degrees of uncertainty during the election.

KEYWORDS:

• Internet
• Electoral Integrity
• Autocracy
• Hybrid Regimes
• Communication

Acknowledgements

Former versions of this article have been presented at the 2021 seminar at the Université de Montréal Electoral Chairs Tuesday Seminar and the 26th IPSA World Congress. We thank all panelists and three anonymous Reviewers for very helpful suggestions. Further, we would like to thank the AccessNow network for providing the necessary data on internet shutdowns.

Disclosure statement

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

Notes

1 For example, the 5-day outage in 2011 in Egypt has caused approximately a loss of $90 million (Wagner 2018).

2 Liberal democracies are excluded from the analysis. Although internet shutdowns occur also in liberal democracies (e.g. Howard, Agarwal, and Hussain 2011), the room for maneuver is institutionally limited and we assume the rationale behind shutdowns in liberal democracies to differ substantially. We identify and exclude liberal democracies using the regimes of the world coding from V-Dem [v2x_regime_amb]. Autocracies are included in the analysis only, if formally competitive elections are held. Cases in which despite this formal criterion no oppositional party or candidate received votes are deemed closed autocracies and excluded from the analysis.

3 A list of all election-related internet shutdowns is provided in the online appendix A1. In total, Access Now recorded 684 internet shutdowns between 2016 and 2020. Most of these shutdowns were executed locally (N = 314) and lasted only a few hours. If official justifications for the shutdowns are provided, these refer mostly to public safety (N = 159), fake news and hate speech (N = 121), or national security (N = 67).

4 Importantly, the operationalization of content control and shutdown capacity refers to the theoretic capacity to control the technology. Yet, there may be incidences where the actual occurrence of shutdowns may be perceived by experts as indicator of the theoretic capacity. We therefore use lagged variables for capacity and content control.

5 For intuitive interpretation, the variables of content control and electoral violence are inverted from the original data. Thus, higher values indicate a higher degree of content control or electoral violence.

6 For conceptual reasons, we decided not to include the overall degree of freedom or dummy variables for regimes. Firstly, liberal democracies and closed autocracies are excluded from the sample. Second, any measure of democraticness is conceptually linked to the freedom of expression and the conduct of free and fair elections and thus correlated to our IVs.

7 Descriptive data for all independent variables is summarized in the online appendix A1.

8 The boxplots show the median and interquartile range (IQR) through the boxes. The whiskers show minimum and maximum observations within the 1.5* IQR. The notches around the median show the 95%-confidence interval of the median.

9 In both types of elections, countries that ordered internet shutdowns show statistically significant higher levels of capacity. In presidential elections, t(39) = −5.3, p<0.001. In parliamentary elections, t(27) = −4.9, p<0.001.

10 Although the pooling of both types of elections would allow us to increase the statistical power of each regression model due to higher numbers of observations, this approach would require the inclusion of further controls (dummy for type of elections) and modelling of autocorrelation between observations. Keeping the rarity of shutdowns in mind, we decided to use rather slim model in split samples.

11 Correlations and model statistics are summarized in the online appendix A1.

12 These initial models remain largely robust when controlling for further modernization factors or for military personnel, press freedom, and oil rents. Results of these tests are summarized in the online appendix A3.

13 The capacity levels are calculated on basis of the average capacity and ±1 standard deviation (lower and upper bound).

14 For lower levels of capacity, the difference between lower and average content control of content are even larger: 8 times higher for average capacity [dotted line], and 11 times higher for lower capacity [solid line].

15 Controlling for rare-event logistic regression (King and Zeng 2001), we find our coefficients to be stable but with slightly larger p-values.

Additional information

Notes on contributors

Kristin Eichhorn

Kristin Eichhorn is a PhD Candidate at the Chemnitz University of Technology. She holds an MA in East European Studies from FU Berlin. Her research interests include authoritarian regimes and transformations, elections and digitalization.

Eric Linhart

Eric Linhart is Professor for Political Systems at the Chemnitz University of Technology. His research interests include electoral systems, voting behavior and coalition theory. His publications have appeared, among others, in the European Political Science Review, Party Politics, Political Studies and West European Politics.