The influence of the 2021 European flooding on pro-environmental attitudes and partial behaviour transition

The study investigates whether direct experience with extreme weather events, specifically the July 2021 European floods, causally increases pro-environmental attitudes and translates into pro-environmental behaviours. While prior research often finds associations between extreme events and environmental concern, causality is unclear due to self-selection and observational designs. Leveraging a natural experiment in Luxembourg during an ongoing national survey of 16–29-year-olds, the authors assess how exogenous exposure to flooding during the fieldwork period influenced environmental attitudes and behaviours, and whether effects are stronger for those living closer to flood zones. The work addresses research gaps on: a) the relationship between personal experiences of extreme weather and climate attitudes, b) whether such events help close the attitude–behaviour gap, and c) whether environmental attitudes mediate effects on behaviour. The study contributes by exploiting quasi-random timing of exposure, analysing subnational geographic proximity, and using validated, multidimensional instruments for attitudes and behaviour.

Prior studies document links between extreme weather experiences and climate change concern, risk perception, and environmental behaviours, including voting, yet findings are mixed and often noncausal. Some research suggests beliefs about climate change shape interpretations of extreme events rather than vice versa. Many studies rely on single-item measures and lack longitudinal or quasi-experimental designs, limiting causal inference. Psychological distance theory posits that direct, proximal experiences can reduce abstractness and increase perceived risk, potentially fostering pro-environmental responses. Evidence on geographical proximity suggests heterogeneity: effects can be stronger among those living closer to disasters. However, validated multidimensional measures of environmental attitudes and behaviour are rarely used in such analyses, underscoring the need for rigorous designs to test whether extreme events shift attitudes and narrow the attitude–behaviour gap.

Design: Natural experiment exploiting the exogenous timing of severe flooding in Luxembourg during the field period of a national survey, creating a quasi-random division of respondents into pre-event (control) and post-event (treated) groups. Treatment is a binary indicator for being surveyed before vs. after the floods (which began the night of July 14–15, 2021). Spatial proximity to floods is a binary variable indicating residence in one of 102 communes directly affected by flooding/extreme water levels (coded using official hydrometric data from the Luxembourg Land Registry and Topography Administration). Sample: Cross-sectional data from the “Young people and COVID-19” (YAC+) survey, a stratified random sample of residents aged 12–29 in Luxembourg. Fieldwork: 13 July–1 October 2021. Approximately 500 respondents were surveyed before the floods and ~2000 after; the analytical sample includes 2058 respondents aged 16–29 (younger respondents excluded due to missing key variables), after further exclusions for missingness. Ethical approval was granted by the University of Luxembourg Ethics Review Panel (ERP20-041-C-A; amendment 1); electronic consent obtained. Measures: - Environmental attitudes: 9-item scale by Diekmann & Preisendörfer capturing affective, cognitive, and conative dimensions. Factor analysis conducted; factor scores computed as an environmental scale (Cronbach’s α = 0.82). Affective, cognitive, and conative subscales were standardized and averaged due to low item counts. - Pro-environmental behaviour: Self-reported frequency of seven behaviours: (1) buying organic food; (2) buying products with an environmental seal; (3) eating less meat; (4) using the car less for short trips; (5) conscious heating use at home; (6) paying attention to energy efficiency when buying appliances; (7) using a reusable water bottle. - Covariates/adjustments: Sociodemographics and known predictors, including gender, age, education, income/wealth proxies, household size, dwelling type; social trust (World Values Survey-based scale measuring specific and generalized trust), institutional trust (trust in a range of institutions, 5-point Likert), and fairness perceptions (six-item scale of distributive principles: equality, need, equity, entitlement). Analytical strategy: Multivariate linear models with cluster-robust standard errors clustered at the commune level; two-sided t-tests. Models adjust for pre-treatment characteristics for efficiency. Spatial heterogeneity examined via stratification by proximity and interaction models testing moderation of the attitude–behaviour link by flood exposure and proximity. Causal effects estimated via two-stage instrumental variable (IV) regression using treatment timing and spatial proximity as instruments for environmental attitudes when predicting behaviours. Estimation implemented with R (estimatr for cluster-robust models; ggplot2, tmap, modelsummary for presentation).

- Environmental attitudes increased after the floods. Strongest effects were observed for the affective dimension and the overall environmental attitude scale: affective t(101)=3.410, P<0.001, β=0.190, 95% CI [0.079, 0.301]; overall scale t(101)=2.947, P<0.01, β=0.169, 95% CI [0.055, 0.283]. The cognitive dimension showed a weaker but significant effect: t(101)=2.018, P<0.05, β=0.123, 95% CI [0.002, 0.244]; the conative dimension was not significant (pre/post comparison). - Spatial proximity mattered. Comparing those living very close to flood areas vs. those not directly affected showed significant differences: conative t(101)=3.206, P<0.01, β=0.360, 95% CI [0.137, 0.583]; affective t(101)=2.967, P<0.01, β=0.333, 95% CI [0.110, 0.556]; overall scale t(101)=2.399, P<0.01, β=0.271, 95% CI [0.047, 0.496]; cognitive t(101)=2.967, P<0.01, β=0.373, 95% CI [0.124, 0.622]. - No direct post-flood increase in reported pro-environmental behaviours after adjusting for attitudes and covariates; environmental attitudes remained the strongest predictor of behaviour. - Moderation: No overall moderating effect of flood exposure on the attitude–behaviour relationship. However, among those living near flood areas, the attitude–behaviour link strengthened for specific behaviours: buying organic food t(101)=2.269, P<0.05, β=0.230, 95% CI [0.029, 0.432]; buying energy-saving electronics t(101)=2.745, P<0.01, β=0.270, 95% CI [0.075, 0.465]. - Causal IV estimates: A significant causal effect of environmental attitudes on select behaviours was identified: reduced meat consumption t(101)=2.199, P<0.05, β=0.468, 95% CI [0.046, 0.889]; energy-saving behaviour t(101)=2.218, P<0.05, β=0.391, 95% CI [0.041, 0.741]. Overall, translation of heightened attitudes into behaviours appeared partial and behaviour-specific.

The findings support experiential learning and psychological distance mechanisms: experiencing a severe, locally salient flood increased environmental attitudes, with effects amplified by geographical proximity to the disaster. Yet, these heightened attitudes did not broadly or directly translate into self-reported behavioural changes once controlling for attitudes and covariates, indicating a persistent attitude–behaviour gap. Moderation analyses show that proximity can strengthen the attitude–behaviour link for some purchasing-related actions (organic food, energy-efficient electronics), suggesting that context and behaviour domain matter. Instrumental variable analyses further indicate a causal influence of environmental attitudes on specific behaviours (less meat consumption, energy saving), although effect sizes are modest. These results imply that relying on the salience of climate impacts alone is insufficient to drive widespread behavioural change; policy instruments and supportive infrastructures are needed to channel increased concern into consistent, sustained pro-environmental actions. Nonetheless, elevated environmental consciousness may bolster support for green policies, which can, in turn, facilitate behaviour change.

This study leverages a natural experiment surrounding the July 2021 floods in Luxembourg and validated multidimensional measures to provide evidence that extreme weather experiences can causally increase environmental attitudes, especially among those in close proximity to the event. However, the translation of attitudes into behaviour is partial and domain-specific, with causal effects detected for reduced meat consumption and energy-saving behaviours but not across all behaviours studied. Contributions include: (1) causal identification of the impact of extreme events on attitudes; (2) subnational analysis of spatial proximity effects; and (3) deployment of validated multi-item scales for attitudes and behaviours. Future research should: (a) examine actual (observed) behaviours alongside self-reports; (b) investigate mechanisms underlying heterogeneous behavioural responses; (c) expand to broader age ranges and diverse populations to assess generalisability; and (d) use larger samples and longitudinal designs to enhance power for causal mediation and domain-specific behavioural effects.

- Behavioural outcomes are self-reported; future studies should incorporate objective behavioural measures. - Heterogeneous and sometimes noncongruent behavioural effects warrant deeper investigation into mechanisms and contexts. - The sample comprises individuals aged 16–29, which may limit generalisability to other age groups. - Larger samples are needed to robustly study causal effects of attitudes on behaviours, as observed relationships appear modest.