This website uses cookies so that we can provide you with the best user experience possible. Cookie information is stored in your browser and performs functions such as recognising you when you return to our website and helping our team to understand which sections of the website you find most interesting and useful.
This blog is based on the paper “A Bayesian Analysis of Collective Action and Internet Shutdowns in India”. [1]
Introduction
Since 2011, internet shutdowns have steadily become an increasingly popular form of digital repression, especially in India – which accounted for more than 50% of global recorded shutdowns from 2016 to 2019. Within India, Jammu and Kashmir (J&K) has the highest number of recorded shutdowns. Common shutdown justifications include ‘ensuring public safety’ in order to curb the prevalence of collective action in the form of protests and riots.
As such, we conduct a quantitative exploration of the relationship between collective action and internet shutdowns in India (with a case study of J&K) through a two-step analysis. We first analyse the prevalence of collective action and the presence of a shutdown on any given day, and sec- ondly quantify the occurrence of protests and riots during a sustained shutdown to answer the following key questions:
1) To what extent does an increase in collective action impact the probability of observing a shutdown on any given day
2) To what extent does the duration of an internet shutdown impact the occurrence of riots and protests?
Methods
We apply Bayesian inference via Generalised Linear Models (GLMs) implemented using the Stan probabilistic programming language in R to estimate correlates of shutdown behaviour. We adopt a Bayesian approach for a number of reasons. Firstly, the Bayesian approach allows us to incorporate pre-existing knowledge and expertise in the form of priors on the fixed effects, (𝛽), in our model. This, in turn, improves inference and provides a shrinkage effect that mitigates the impact of sampling variation. Secondly, the Bayesian approach produces a full posterior distribution over parameters rather than a point estimate. This explicit quantification of uncertainty in the results provides additional insight into the effects of shutdowns. These methods have not been applied to the areas of collective action and shutdowns to our knowledge.
Data
We primarily use publicly available count data from Access Now and the Integrated Conflict Early Warning System (ICEWS) for the time frame 2016 to 2019. While the Access Now dataset provides a relatively comprehensive overview of internet shutdowns, ICEWS consists of data surrounding early crisis and conflict data between certain socio-political actors, for example protesters and a local government. We primarily make use of recorded Shutdown Occurrences, which is daily time stamped data of recorded shutdowns from the Access Now dataset, and Protests and Riots from the ICEWS dataset, also daily time stamped.
Results
In answering the first research question we conclude that, in the case of India, the data is consistent with riots having an effect at 0.09 with a 95% credible interval of 0.04 – 0.15. For Jammu and Kashmir, the data is consistent with riots having an effect of 0.15 with a 95% credible interval of 0.03 – 0.26, and protests having an effect of 0.07 a 95% credible interval of 0.01 – 0.13. Therefore, the India-wide data shows that an increase in riots increases the probability of observing a shutdown on the same day. For Jammu and Kashmir, the data shows that an increase in both riots and protests increase the probability of observing a shutdown, although riots have double the predictive value than protests.
In answering the second research question we conclude that, in the case of India, while the day of a shutdown is not a meaningful predictor for protests, it is for riots with the occurrence of riots decreasing by 0.08 with each subsequent day (e.g. a 1-point increase in the shutdown day leads to a decrease of 0.08 of riot count). For Jammu and Kashmir the day of a shutdown is not a meaningful predictor for protests nor riots.
The result suggest that riots are a better predictive parameter for both India and Jammu and Kashmir in comparison to protests. And that there is only a marginal negative effect of ongoing riots in relation to the length of a shutdown.
Discussion
One reason riots appear to be a better predictor variable could be explained by riots being an extreme form of collective action. As riots occur less frequently, some governments may choose to impose more shutdowns. Furthermore, it may be that those deciding to order a shutdown are either unaware of the degree of impotence of the hierarchical structure of a riot in comparison to a protest, or are overconfident in the ability of a shutdown to stop riots. Alternatively, if severe social unrest occurs, governments may be willing to use any means necessary, including a shutdown. Given the negative consequences of a shutdown alternatives to shutdowns are an important consideration. While these may include, for example, legislative changes to address societal factors that ignite shutdowns, future research may look to compare and contrast government tools and assess their effectiveness in maintaining information and social control. Ultimately, it should never be overlooked that protest, riots, and the state’s responses to these, represent an ongoing and complex relationship between citizens and the state. These incorporate a wide range of social, economic, and political factors that should not be blindly abstracted into a simplistic technical set of solutions or responses.
We acknowledge there are a range of potential confounding factors, and encourage future studies to incorporate and control for additional variables.
Conclusion
Internet shutdowns are a growing issue. Given their economic, legal and political consequences require further research. Efforts from all parts of society, including government, industry and academia, are necessary to mitigate the negative consequences of shutdowns.
The analyses presented here, relying on the existence of datasets such as those gathered by Access Now and the ICEWS, combined with new developments in statistical and computational methods for their analysis, provided a lens into understanding the motivations of censors and the responses of citizens to shutdowns. This, in turn, provides an evidential basis both for policy development surrounding information controls, but also to identify the potential for interventions, both by government and civil society, to combat future shutdowns and preserve individuals’ freedoms with respect to human rights.