Emergence of protests during the COVID-19 pandemic: quantitative models to explore the contributions of societal conditions

The paper investigates why and how civil resistance movements emerged during the COVID-19 pandemic, focusing on the roles of societal conditions that evolve at different time scales. It situates the study within theories of grievances, political opportunity, and mobilisation, noting that pandemics cause uncertainty, economic disruption, and psychological burdens that can propagate through social interactions to create societal tension. The authors aim to integrate literature-based insights with quantitative analysis to determine which fast-changing variables predict the timing (when) of protest outbreaks and which slow-moving structural factors explain the severity (how intense) of protest waves. They propose and test three hypotheses: H1, that protest outbreaks relate to fast variables; H2, that the acceptability/legitimacy of containment measures relates to protest outbreaks; and H3, that slow variables influence the intensity of protest peaks.

The theoretical framework synthesizes grievance theories (structural deprivation, inequality, repression), political opportunity (changes in state capacity, legitimacy, rights and freedoms), and mobilisation capacity (organisation, resources, demographics, urbanisation). Prior work shows correlations between deprivation (poverty, group grievances, limited civil liberties) and uprisings; the capacity of governments to adapt can either mitigate grievances or create opportunities for resistance. The authors emphasise multi-timescale social dynamics: demographic conditions change slowly; political-economic stability varies at intermediate pace; individual emotions shift rapidly. They develop a causal loop diagram (CLD) mapping pathways from pandemic-driven shocks (morbidity, containment measures, uncertainty) through grievances, opportunity structures, and mobilisation processes to civil resistance, including reinforcing/balancing feedbacks (e.g., uncertainty-fake news loops, legitimacy effects, repression’s nonlinear impacts). This literature highlights that fast shocks may trigger protests, while slow structural conditions set the stage for their severity.

The study uses a mixed-methods approach combining: (1) Construction of a causal loop diagram (CLD) from literature across grievances, political opportunity, and mobilisation, tailored to pandemic contexts. (2) Data assembly for 27 countries (unbalanced panel, 7545 country-days in 2020), compiling proxies for fast, intermediate, and slow factors: excess mortality (The Economist), unemployment and GDP (OECD), mobility (Google), government response indices including containment, stringency, healthcare and economic support (OxCGRT), protests/riots and repression (ACLED), political rights and civil liberties (Freedom House), state capacity/legitimacy and group grievance (Fragile States Index), urbanisation (World Bank), and youth bulge (UN). Daily variables were smoothed using 7-day rolling averages to reduce weekday effects. (3) Statistical modelling: zero-inflated negative binomial (ZINB) models for fast variables to account for overdispersion and excess zeros in daily protest counts, and negative binomial models (NBM) for slow variables (available only at annual resolution in 2020) to assess cross-national differences. Model selection used AIC and log-likelihood; multicollinearity checks used VIF; maximum likelihood estimation was applied. (4) System dynamics modelling (SDM): The CLD was operationalised into a stock–flow model capturing two core processes—societal tension (sigmoid function of legitimacy of measures vs. threshold; influenced by mobility restrictions and excess mortality) and mobilisation (transition from calm citizens to potential activists to activists, with recruitment and disengagement dynamics). The SDM includes parameters such as tension velocity, threshold, pressure velocity, mobilisation tendency and threshold, and fatigue velocity. Equations specify flows (CU, CR, CM, CD) and the protest output as a linear function of activists. Calibration was performed in Vensim on four countries (Italy, Spain, USA, Netherlands) with a daily time step; parameters were optimised to fit observed protest dynamics (details in Supplementary Information).

- Fast variables and outbreak timing (H1): ZINB models fitted better than standard NBM for daily counts, indicating distinct triggering and count-generating processes. Low containment levels were associated with an absence of COVID-19-related protests; higher unemployment correlated with increased likelihood of protests; economic support measures correlated with reduced likelihood; limited evidence for a consistent repression effect, aligning with literature on its nonlinear impact. These results support H1 that fast-changing variables relate to the onset of protest outbreaks. - Acceptability/legitimacy of measures (H2): Legitimacy was modelled as a function of excess mortality and mobility restrictions. Empirically, phases with stricter measures and higher mortality showed relationships with protest activity; however, the clearest statistical result reported is that low containment is associated with absence of protests, while unemployment raises and economic support lowers protest likelihood. The evidence is mixed but indicates that perceived acceptability/legitimacy is implicated in triggering dynamics. - Slow variables and intensity (H3): NBM models with slow structural variables (e.g., group grievance, political rights, state legitimacy) provided a better fit for the intensity of protests than fast-variable ZINB models, supporting H3. Countries with high grievances and lower political rights experienced more severe waves of protests. Contrary to some literature, higher youth bulge correlated with fewer protests in this pandemic context, possibly because COVID-19 disproportionately affected older populations. - System dynamics model: Calibration yielded good fits for the Netherlands and Spain but weaker fits for the USA and Italy. Differences likely stem from spatial heterogeneity (e.g., US state-level variation), country-specific tension–legitimacy dynamics, and distinct motivations for protests. The SDM captured bursty patterns via tension and mobilisation feedbacks but requires finer-grained data for broader generalisability. - Overall predictive power: Log-likelihoods and AICs indicate only modest predictive performance; latent processes (e.g., unobserved societal tension), nonlinearities, and data limitations constrain forecasting accuracy.

Findings indicate that incorporating fast-changing variables helps identify when protest waves are likely to start, while slow-moving structural conditions better explain how severe those waves become. The study demonstrates the importance of modelling multiple time scales and feedback processes—tension, recruitment, and disengagement—in understanding civil resistance during a pandemic. The SDM framework connects disruptive shocks (mortality and restrictions) to legitimacy, societal tension, and mobilisation dynamics, reproducing bursty protest patterns in some contexts. Mixed calibration results suggest country-specific mechanisms and spatial heterogeneity must be represented. The results highlight the limits of purely structural or purely short-term models and argue for hybrid, dynamical approaches integrating both fast and slow processes.

The paper contributes a holistic framework—combining literature-derived causal mapping, cross-national statistical analysis, and a system dynamics model—to study the emergence of protests during COVID-19. It shows that fast variables improve understanding of protest onset, while slow structural variables better estimate protest intensity. The SDM provides an explicit representation of tension and mobilisation feedbacks, illustrating pathways through which pandemic shocks can lead to civil resistance. Future research should: expand fine-grained, multi-year event and attendance data; integrate additional factors (e.g., specific legislation, varying political opportunity perceptions); exploit individual-level or agent-based models for contagion and behavioural change; incorporate media and social-media sentiment to capture perception and disinformation dynamics; and model spatial heterogeneity (e.g., subnational variation) to enhance predictive power and generalisability.

- Data scope and granularity: Analysis largely limited to 2020 due to daily ACLED coverage; reliance on OECD and related sources biases toward WEIRD countries; limited attendee counts and protest typologies constrain mobilisation modelling. - Measurement and selection biases: News-based event data (ACLED) may have coverage biases; concurrent movements (e.g., Black Lives Matter) may confound COVID-19 protest signals; legitimacy and political opportunity perceptions are difficult to capture with available proxies. - Model constraints: Slow variables are annual, preventing longitudinal within-country modelling; ZINB/NBM models show modest predictive power, indicating unobserved latent processes and nonlinearities; SDM calibration mixed across countries, suggesting need for finer spatial resolution and additional mechanisms. - Generalisability: Country-specific differences (e.g., US state heterogeneity, Italy’s early high mortality) limit model transferability without context-specific adjustments.