Leaving messages as coproduction: impact of government COVID-19 non-pharmaceutical interventions on citizens' online participation in China

The study investigates how government-implemented non-pharmaceutical interventions (NPIs) during COVID-19 affected citizens’ online participation in China in 2020. While NPIs (e.g., curfews, travel restrictions, closures, lockdowns) effectively reduced transmission, they also substantially altered social and economic life. Prior work has emphasized NPIs’ health and environmental impacts and individual lifestyle changes, but little is known about their socio-political effects, particularly online interactions with government when offline channels were constrained. The authors posit that NPIs, by limiting offline access to services and information, could shift and increase citizen-government online engagement. They focus on China due to early and widespread NPI adoption across many cities in 2020 and to address a research gap in developing-country contexts. The core research question is whether and how NPIs influenced the volume and nature of online participation with local leaders, and what mechanisms (heightened needs vs. coproduction behaviors) explain any observed changes.

The paper reviews evidence that NPIs reduced COVID-19 transmission and had broad macro-level environmental and societal effects (e.g., air/water quality improvements, reduced noise, supply chain disruptions, increased inequalities). At the micro level, NPIs affected mental health, sleep, diet, and exercise. Yet socio-political impacts, especially online citizen-government interaction during crises, remain underexplored. The authors highlight the importance of communication for NPI compliance and effectiveness amid complex, evolving policies. Online participation is framed as a distinct, accessible political mode, especially for disadvantaged groups. Two strands of prior research are noted: (1) whether traditional predictors (socioeconomic status, education, social networks) extend to online participation; and (2) online-specific factors (digital skills, social media use, government media strategies, e-government maturity). There is limited examination of exogenous shocks (like pandemics) on online participation. The Chinese context offers a relevant case due to early NPIs, nationwide scope, and unique political-cultural conditions.

Data and setting: The authors compiled a day-by-city panel (323 prefecture-level cities in mainland China, excluding Hong Kong, Macau, Taiwan, and cities lacking data) from 2020-01-01 to 2020-12-31. Online participation was measured via the Local Leadership Message Board (LLMB) hosted by People.cn, an official nationwide platform. Using a Python crawler, they collected 108,981 messages addressed to City Party Secretaries in 2020 and aggregated them to city-day counts. They also consider messages to mayors in robustness checks.

Treatment (NPIs): City-day NPI status was coded from municipal/provincial government websites, transport department announcements, bus/passenger company notices, and media reports. Strict NPIs required all three: (1) prohibition of any congregation; (2) restrictions on intercity traffic and individual movement; (3) restrictions on intra-city traffic and individual movement. Implementation dates are listed in Supplementary Table 1. A relaxed NPI definition used in robustness required (1) prohibition of social gatherings and (2) intercity travel restrictions but allowed intra-city movement.

Controls: Local COVID-19 severity (city-day) from the Sina Real-time COVID-19 Platform and official health commission sources: newly confirmed cases, new deaths, current confirmed cases, cumulative confirmed cases, cumulative deaths.

Empirical strategy:

• Difference-in-differences (DiD): Y_it = β·NPIS_it + α·X_it + μ_i + π_t + ε_it, where Y_it is LLMB message count to the City Party Secretary in city i on day t; NPIS_it indicates NPI implementation; X_it are COVID-19 severity controls; μ_i and π_t are city and date fixed effects. Event-study (leads/lags) assesses parallel trends, using the day before implementation as reference.
• Interactive Fixed Effects counterfactual estimator (IFEct; Liu et al., 2020): Y_it = β·NPIS_it + f(X_it) + h(U_it) + ε_it, where f(·) captures observable controls and h(·) captures unobserved factors with heterogeneous time trends. Steps: (1) estimate f and h using control cities; (2) construct counterfactual Y(0) for treated cities; (3) compute treatment effects as Y − Y(0) for each treated city-day; (4) average across treated units. Validity was checked with placebo tests (TOST equivalence and t-test) using the week before NPI implementation.

Robustness checks:

• Placebo DiD with 1000 random treated groups to assess distribution of fake effects.
• Winsorization of outcome at 1% and 5% tails.
• Subsamples: first wave period (2020-01-01 to 2020-05-31); pre–Spring Festival window (2020-01-01 to 2020-01-24); excluding Spring Festival period (2020-01-24 to 2020-02-02).
• Alternative treatment definition (relaxed NPIs) and outcome including both Party Secretaries’ and mayors’ messages.

Heterogeneity analysis: Cities split by above/below mean GDP per capita (economic conditions), mobile phone subscribers per capita (telecommunication foundation), and share of residents with college degree (education level). Sources: 2020 China City Statistical Yearbook, city Statistical Communiqués, and the Seventh National Population Census.

Mechanism analysis: LLMB requires categorizing messages into: seeking help, consulting, providing suggestions, expressing thanks, complaining. Separate regressions by message type assess two mechanisms: (1) inconvenience in daily life (seeking help, consulting, thanking); (2) coproduction behaviors (providing suggestions, complaining). Topic-field composition (e.g., transportation, healthcare, administration, etc.) was examined via text analysis (Supplementary Table 3).

• Main effect (DiD): Implementing NPIs increased daily LLMB messages to City Party Secretaries. With COVID-19 controls and city/date fixed effects, coefficient ≈ 0.166 (Model 2), about 18% of the treated cities’ pre-intervention mean (0.932). Using levels without controls showed 0.217; log outcome models also positive (e.g., ≈ 0.053 in Model 4). Event-study confirmed pre-trends were parallel (no significant differences for k ≤ −2) and showed growing positive effects after implementation (k > 1).
• IFEct estimator: Treatment effect remained positive and significant; with controls and fixed effects, treated cities had ≈ 0.743 more messages than their counterfactuals (Model 6), aligning with DiD magnitudes and validating robustness.
• Robustness: Placebo DiD showed null effects centered at zero across 1000 random assignments. Winsorized outcomes (1% and 5%) preserved significance (coefficients ≈ 0.111–0.157). Subsamples (first wave; pre–Spring Festival; excluding Spring Festival) all yielded positive significant effects (e.g., ≈ 0.148–0.376 across Models 11–16).
• Alternative treatment and outcome definitions: Relaxed NPIs also increased participation with larger effects (e.g., ≈ 0.329–0.397 to Party Secretaries; ≈ 0.409–0.429 when combining Party Secretaries and mayors; Models 17–20).
• Heterogeneity (Table 8): Effects were larger/significant in cities with: • Higher economic conditions: coefficients ≈ 0.327–0.467 (levels) and ≈ 0.102–0.115 (logs), all p<0.05; not significant in lower economic conditions. • Stronger telecommunication foundations: larger coefficients (e.g., levels 0.294 vs. 0.134; logs 0.055 vs. 0.013). • Higher education levels: significant positive effects (levels ≈ 0.409–0.563; logs ≈ 0.115–0.131), but not significant in less-educated cities.
• Mechanisms (Table 9): • Inconvenience/need: seeking help increased (0.126, p<0.05; 0.094, p<0.1); expressing thanks increased (0.006, p<0.05). Consulting not significant. • Coproduction: providing suggestions increased (0.087, p<0.01; 0.053, p<0.01); complaining not significant. • Topic composition during NPI stages shifted toward transportation, healthcare, and administration, consistent with mobility restrictions and governance concerns.
• Temporal dynamics: Online participation peaked about three weeks after NPI implementation, suggesting a tolerance threshold beyond which adverse effects (e.g., life difficulties, psychological stress) become more pronounced.

The findings directly answer whether NPIs affected citizens’ online engagement: NPIs increased online participation with local leaders, indicating that when offline channels were constrained, citizens shifted to digital channels for information, services, and problem-solving. This validates the importance of communication infrastructures and digital competencies for crisis governance. The heterogeneity results underscore that cities with stronger economic resources, telecommunication infrastructures, and higher education levels experience larger boosts, highlighting inequality risks in digital access and capacity. Mechanism results show both demand-driven engagement (seeking help, thanking) and proactive coproduction (providing suggestions), indicating citizens are not merely passive policy recipients but active contributors to crisis governance. The temporal pattern (peaking at ~3 weeks) implies policy timing matters: prolonged strict NPIs may erode benefits and intensify negative social effects, necessitating adaptive calibration of restrictions. Overall, the results reveal socio-political externalities of NPIs and the complementarity of online and offline participation: digital channels can compensate when physical participation is limited, enhancing responsiveness and potentially compliance.

This study provides causal evidence that COVID-19 NPIs in China elevated citizens’ online participation with local leaders, robust across multiple identification strategies and checks. It contributes by: (1) identifying socio-political impacts of NPIs beyond health/environmental outcomes; (2) demonstrating the role of resources, digital infrastructure, and education in shaping online engagement; and (3) uncovering mechanisms of increased demand and coproduction behaviors. Policy-wise, results advocate strengthening e-government and online participation systems to maintain governance responsiveness during crises and to encourage citizen coproduction. Future research directions include: (1) expanding data sources beyond LLMB (e.g., Weibo/WeChat, government sites, 12345 hotlines) to better capture online participation; (2) incorporating individual-level data to assess heterogeneous impacts across demographic groups; (3) evaluating persistence of behavior changes post-interventions, which current data limitations preclude; and (4) moving beyond participation levels to analyze content and affective dimensions of online engagement.

The authors note several constraints: (1) Online participation is measured via a single nationwide platform (LLMB) due to data availability; other channels were not incorporated. (2) Lack of individual-level data limits assessment of heterogeneous effects across demographic groups and potential exclusion of disadvantaged populations. (3) Difficulty determining whether participation changes persist after NPIs because interventions continued over multiple years, preventing clean post-period observation. (4) The analysis focuses primarily on participation levels; deeper content and affective analyses were not conducted within this study.