Causal Modeling of Climate Activism on Reddit

Abstract

Climate activism is crucial in stimulating collective societal and behavioral change towards sustainable practices through political pressure. Although multiple factors contribute to the participation in activism, their complex relationships and the scarcity of data on their interactions have restricted most prior research to studying them in isolation, thus preventing the development of a quantitative, causal understanding of why people approach activism. In this work, we develop a comprehensive causal model of how and why Reddit users engage with activist communities driving mass climate protests (mainly the 2019 Earth Strike, Fridays for Future, and Extinction Rebellion). Our framework, based on Stochastic Variational Inference applied to Bayesian Networks, learns the causal pathways over multiple time periods. Distinct from previous studies, our approach uses large-scale and fine-grained longitudinal data (2016 to 2022) to jointly model the roles of sociodemographic makeup, experience of extreme weather events, exposure to climate-related news, and social influence through online interactions. We find that among users interested in climate change, participation in online activist communities is indeed influenced by direct interactions with activists and largely by recent exposure to media coverage of climate protests. Among people aware of climate change, left-leaning people from lower socioeconomic backgrounds are particularly represented in online activist groups. Our findings offer empirical validation for theories of media influence and critical mass, and lay the foundations to inform interventions and future studies to foster public participation in collective action.

Figures (9)

• Figure 1: Probabilistic Graphical Model (left) and description of nodes (right). Circles represent stochastic variables; gray nodes are observed and white ones are latent.
• Figure 2: Temporal dynamics of the model. The final time $t_A$ is the activation time, or the end of the observation period for non-activated users. The first three points corresponds to one year, five weeks, and one week before $t_A$.
• Figure 3: Mean values of the estimated parameters, with 95% credible intervals. (a) The coefficients are the causal effects of the parents of $S$, \ref{['eq:sympathy']}. These contribute to the weighted average of the mean of the normal extraction of $S$. (b) The coefficients are the log-odds in \ref{['eq:activation']}. (c) The coefficients are $\log \beta_{p1}$ in \ref{['eq:P_S']}. They represent the interaction term between $D_{sub}$ and $S$. When the coefficient is positive, an increase of $S$ increases the odds of writing in a subreddit with a positive score in that sociodemographic category. (d) The coefficients are $\log \beta_{I1}$ in \ref{['eq:5']}. They represent the interaction term between $P_S$ and $D_{sub}$. When the coefficient is positive, an increase of $P_S \cdot D_{sub}$ increases the odds of interacting with an activist.
• Figure 4: For each value of $\text{var}(S\xspace)$, we compare the distributions of accuracy of 100 samples of the predictive posterior for $A$. The error bars represent one standard deviation of the accuracy. For each $\text{var}(S\xspace)$, the left-hand dot refers to the experiment using all the variables of the model. The other dots refer to the experiment run after the removal of a variable, both long- and short-term.
• Figure A.1: Number of activated users per subreddit in our final user selection.