Free riding in climate protests

The study investigates whether and how individuals’ decisions to participate in climate protests depend on their expectations about others’ participation. The central research question is whether higher expected turnout makes a given individual more (strategic complementarity) or less (strategic substitutability) likely to attend. The context is the global Fridays for Future (FFF) movement, where broad adult participation is pivotal for political impact but little is known about adult mobilization dynamics. Leveraging the Third Global Climate Strike on 20 September 2019—a large, coordinated event explicitly inviting adults—the authors pre-registered the hypothesis that participation decisions would be strategic complements, given arguments about perceived protest success and image concerns. They then test this hypothesis causally using a randomized information intervention that shifts beliefs about others’ turnout, tracking adult respondents in Berlin, Hamburg, Munich, and Cologne across three survey waves (two pre-event, one post-event).

Prior work offers mixed evidence on strategic interdependence in protests. Researchers have conjectured strategic complementarity in both general populations and social networks. However, experimental evidence has documented strategic substitutability in Hong Kong’s Umbrella Movement and among right-leaning protesters in Germany, while complementarity was found among left-leaning counter-protesters—suggesting context-specific dynamics. The climate protest literature, particularly on mobilization, has been largely qualitative or non-causal; causal evidence on climate protest participation is scarce. This study contributes by providing causal evidence on strategic interdependence in a major global climate movement and by focusing on the general public beyond activist cores and tight social networks, complementing research on social ties and movement diffusion.

Design: Three-wave online panel survey and randomized controlled trial (RCT) in Germany’s four largest cities (Berlin, Hamburg, Munich, Cologne). Waves: (1) Sep 6–11, 2019; (2) Sep 16–20, 2019 (until noon of protest day); (3) Dec 5–16, 2019. Pre-registration: AEARCTR-0004583. Ethics: Approved by Universität Hamburg’s Faculty of Business, Economics and Social Sciences. Sampling: Quota samples from Kantar’s local online panels of adults aged 18–69; invitations framed as a three-wave study on environmental and climate protection. Compensation per wave plus €2 completion bonus. Wave completion counts: 2,574 (Wave 1), 1,879 (Wave 2), 1,510 (Wave 3; final analysis sample). Attrition from Wave 1→2: 27%; Wave 2→3: 19.6% (no difference by treatment; probit P=0.86). Sample approximately matches city populations on quotas; balance checks show no significant control–treatment differences at 5%. Key variables and timing:

• Wave 1: Own participation intention (Q9); beliefs about others’ planned participation b (Q10); beliefs about others’ actual participation b′ (Q13). City-specific signal s computed from Wave 1 intentions (share answering 1 or 2 on Q9): s = 0.325 (Berlin), 0.367 (Hamburg), 0.367 (Munich), 0.366 (Cologne); overall 0.353 (based on n=2,574).
• Wave 2: Random assignment to control or treatment. Treatment groups received truthful information (signal s) about others’ planned participation in their city; a subset also received age-group signals (18–29, 30–69). All then stated post-intervention beliefs about others’ actual participation b (Q19).
• Wave 3: Actual participation in the local event on Sep 20 (Q25; five options). Outcome a coded 1 if response (1) “Yes, I was there as a participant,” and 0 otherwise. Intervention and first stage: Treatment varied beliefs exogenously. OLS of Δb = b − b′ on treatment indicator z, condition indicator c (c=0 if b below s; c=1 if b≥s), and z×c shows:
• Below-signal group (c=0): Average treatment effect on beliefs +4.25 percentage points (pp), P=0.000.
• Above-signal group (c=1): Average treatment effect −5.54 pp, P=0.000. Robustness: Effects stable when excluding near-signal beliefs, re-weighting to match Wave 1 quotas, and excluding failed attention checks. Estimation strategy: Instrumental-variable models estimate the causal effect of belief changes on participation. Endogenous regressor: Δb = b − b′. Instruments: z, c, and z×c. Models: IV probit (primary, given binary outcome) and 2SLS (reference). Average marginal effects computed to interpret a 1 pp increase in post-intervention beliefs (equivalently in Δb). Assumption: Treatment affects participation only via beliefs (exclusion restriction). Additional heterogeneity analysis exploits age-group signals for treated subset.

• Participation rates: By city—Berlin 9.2%, Hamburg 11.3%, Munich 10.4%, Cologne 13.8%; overall 11.0% (control 10.7%, treatment 11.1%).
• Causal effect of beliefs on participation: Average marginal effect of a 1 pp increase in beliefs is approximately −0.7 pp. • IV probit: coefficient on Δb = −3.272 (SE 0.803), average marginal effect −0.679 (SE 0.208), n=1,510, Wald χ²=16.62, P<0.01. • 2SLS: coefficient on Δb = −0.699 (SE 0.212), average marginal effect −0.699 (SE 0.212), n=1,510, Wald χ²=10.89, P<0.01. • Interpretation: Protest participation decisions are strategic substitutes—higher expected turnout reduces one’s own participation probability.
• Robustness: Effect sizes range from −0.625 to −0.727 across sensitivity checks; tests reject exogeneity of beliefs without instrumenting (Wald tests), supporting the necessity of the RCT.
• Heterogeneity: • By direction of update: Below-signal group AME −0.598 (SE 0.161, P=0.000); above-signal group AME −0.808 (SE 0.284, P=0.004); difference marginally significant (P=0.094), indicating somewhat stronger substitutability for downward updates. • By age-group beliefs (treated subset, n=496): Beliefs about 30–69-year-olds significantly reduce participation (AME −0.493, SE 0.238, P=0.038); beliefs about 18–29-year-olds not significant (AME −0.235, SE 0.217, P=0.278); no interaction with respondent’s own age group (P>0.4).
• First-stage shifts in beliefs confirm strong, directionally consistent treatment effects: +4.25 pp (below-signal), −5.54 pp (above-signal), both P=0.000.

The study provides causal evidence that, among adults in the general population, climate protest participation decisions are strategic substitutes: expecting higher turnout reduces an individual’s propensity to participate. This aligns with models of green transitions that allow for free-riding behavior and with experimental findings from other contexts (for example, Hong Kong’s Umbrella Movement) while contrasting with contexts where social ties and identity/image concerns can produce complementarity. Potential mechanisms consistent with the findings include: (1) individual costs rising with protest size (logistical or psychological, even if FFF events are typically peaceful); (2) diminished signaling value of participation as crowds grow, especially salient for adults joining a youth-led movement; (3) critical-mass dynamics around a target number of adult participants, after which marginal participation is perceived as less pivotal; and (4) the pursuit of relatively modest, specific policy goals encouraging free riding, particularly among strangers lacking strong social ties. The heterogeneity result—effects driven by beliefs about adults (30–69)—supports mechanisms tied to adult participation salience. Practically, decentralized, local mobilizations can mitigate substitutability by lowering logistical barriers, fostering social ties, and potentially turning smaller expected turnouts into a motivator. Communication strategies should avoid overpromising turnout; conservative pre-event estimates may improve mobilization when strategic substitutability prevails.

This paper delivers pre-registered, causal evidence that climate protest mobilization among adults in Germany exhibits strategic substitutability: a 1 pp increase in expected turnout reduces individual participation probability by about 0.7 pp. The results extend the literature by documenting strategic interdependence in the general public (beyond activist networks) within a major global climate movement and by providing actionable implications for movement structure and communication (favoring decentralized, locally organized actions and conservative turnout messaging). Future research could: (1) examine how social ties and network structure shift interdependence from substitutability to complementarity; (2) test generalizability across countries, issues, and protest types (including more radical movements); (3) explore dynamic thresholds/critical mass and framing of goals (modest vs. systemic) on strategic behavior; and (4) investigate micro-mechanisms (image signaling, perceived success probabilities, logistic costs) using enriched experimental designs and field data.

• External validity: Quota-based online panels approximate but do not constitute probabilistic samples; results pertain to adults (18–69) in four German cities during a specific event.
• Attrition: 27% (Wave 1→2) and 19.6% (Wave 2→3); although balanced across treatment and addressed with robustness checks and re-weighting, selective attrition by age was detected.
• Timing: The third wave was delayed to December due to technical issues; outcome is a factual report of participation, but recall bias cannot be entirely ruled out.
• Identification assumptions: Causal interpretation depends on the exclusion restriction that the information intervention affects participation only through beliefs; estimates identify a local average treatment effect for compliers.
• Measurement: Participation is self-reported; beliefs elicited without incentive-compatible scoring to avoid known downsides, which may still introduce reporting noise.
• Context-specificity: Strategic interdependence may differ in settings with stronger social ties, different political stakes, or more radical movement tactics, limiting generalizability.