SPARKLING is a groundbreaking project designed to deepen our understanding of climate change awareness by investigating the causal impacts of natural disasters. Utilizing innovative methodologies and spatiotemporal data, the project will develop a framework to estimate causal effects, leveraging social media to construct an awareness indicator that analyzes sentiment changes over time and space. The project will examine why populations less affected by natural disasters often show greater concern for climate issues, using advanced statistical models and social listening techniques to explore the underlying causal mechanisms. It will also investigate how media narratives influence public perceptions of natural disasters as related to climate change. SPARKLING will significantly impact both academia and the public by providing new methods to address complex data and enhance policy communication, leading to more effective climate policies. Results will be shared through participation in public events like Bright Night and a final study day, promoting engagement and collaboration among stakeholders. This initiative is part of a broader research agenda to maximize public policy action by understanding causal narratives in climate change, setting the stage for further exploration in an upcoming ERC Project.
Understanding how perceived climate risk shapes citizens’ willingness to engage in mitigation is a defining challenge of the twenty‑first century. While the frequency of extreme weather events is rising, many communities newly exposed to such hazards have limited experience from which to form risk perceptions. We study whether direct exposure to extreme climatic events—specifically heatwaves—alters public beliefs about climate change and attitudes toward more ambitious, including self‑protective, mitigation actions. Prior research has documented correlations between disasters and pro‑environmental attitudes, yet few studies provide clear causal identification.
Leveraging the spatial nature of both treatment (continous heat exposure) and outcomes (attitudinal measures), we develop a Bayesian latent‑factor framework that imputes the counterfactual attitude matrix by exploiting shared spatial and cross‑item structure. This approach allows us to isolate treatment effects while fully accounting for latent heterogeneity and spatial spillovers. Additionally, we provide insights about the link between our proposal, balancing weights and outcome regression methods for estimating causal effects.
Our empirical application examines the effects of persistent heatwaves in 2018, 2019 and 2020 in Germany. Combining high‑resolution temperature data with the ARIADNE climate‑attitude survey (Hertie School), we compare directly and peripherally affected areas with unaffected counties to estimate the causal impact of heat exposure on support for environmental policies. The study demonstrates how Bayesian spatial counterfactual imputation can uncover causal effects in settings where random assignment is impossible, and offers fresh evidence on the relationship between climate hazards can mobilize public support for mitigation. Beyond its substantive relevance for climate‑risk perception, the approach is transferable to any setting where geographically referenced interventions interact with spatially correlated attitudes or behaviors.
w/ Leo Vanciu (Harvard College), Veronica Ballerini (Harvard School of Public Health) and Falco J. Bargagli Stoffi (UCLA)
Gun violence in the United States has become an increasingly pressing issue in recent years. The upward trend is alarming, and policymakers are focusing their efforts on reducing the death toll. In this study, we examine gun violence in a changing world. Our primary focus is the relationship between heatwaves—direct consequences of climate change in intensity and frequency—and gun violence. We investigate these relationships within a potential outcomes framework, defining causal effects for areas that have experienced a heatwave and those potentially subject to spillover effects.
We estimate the causal effect by introducing a spatially augmented version of the synthetic control method, leveraging the spatial information in the data to improve the interpretability of our estimates and reduce their variability. We employ a Bayesian regression approach to penalize the selection of more distant control units, within a semiparametric framework that balances unobserved spatial confounding.
Our findings contribute substantively by investigating the role of environmental factors in gun violence and methodologically by naturally extending the synthetic control method to spatial data.