Tracking Heated Conversations in a Warming World

Direct link to paper, published September 28, 2021: https://link.springer.com/article/10.1007%2Fs13278-021-00792-6

tags: climate change; affective polarization; social network analysis; stance detection
Image Credit: Canva

Climate change stands as one of the most hotly divisive issues of today. 

In 2020, the Pew Research Center reported that although a rising proportion of American adults view climate change as a major threat, this trend is primarily concentrated among Democrats. Nearly 90% of Democrats consider climate change as a major issue, compared to only 31% of Republicans, resulting in a national average of around 60% of the U.S. population.

While this national average is lower than a previously estimated median of 70% across 14 countries, international patterns likewise bear out substantial variation in public agreement over climate change.

This stands in stark contrast to overwhelming scientific consensus around climate change – estimated in 2016 as around 97% of climate scientists – as being human-caused and in need of urgent and substantial collective response.

What explains these persistent divides?

Previous research in this area has shown that, much like climate change itself, issues in societal discourse around climate change can be complex. Evidence suggests that major factors contributing to these cleavages in public opinion include the political leaders people identify with, the types of news they consume, and especially in recent years, social media.

Social media platforms can be powerful spaces for sharing information and building connections with others. However, these same capacities can also make them an engine for widening social gaps and exacerbating conflicts. Of growing concern have been the ways social media not only separates people into increasingly alienated echo chambers, but also encourages these groups to act in a hostile manner toward each other. This latter phenomenon has been called affective polarization.

In research with IDeaS alumnus Dr. Aman Tyagi and adviser Dr. Kathleen Carley, we took a look at the dynamics of affective polarization in climate change discourse from a social cyber-security approach. Using a combination of machine learning and network science tools, we analyzed the largest-ever recorded dataset of climate change conversations on Twitter, spanning over 200 million messages and a 100-week period from 2017 to 2019. We wanted to know: How affectively polarized were climate change believers and disbelievers over time?

To classify the accounts in our dataset based on their climate change belief or disbelief, we used a state-of-the-art stance detector developed by our lab. The stance detector tracked how different hashtags were used in support or opposition to beliefs in the reality of climate change, and inferred users’ beliefs based on their own patterns of hashtag usage.

We then developed a new measure of affective polarization based on the sentiment expressed by climate change believers and disbelievers toward each other. We obtained sentiment scores on people’s tweets using Netmapper, which can estimate sentiment in over 40 languages. By comparing the relative usage of positive and negative sentiment between groups, we could characterize how hostile these communities were being toward each other over the two-year period. Moreover, we could investigate the possible reasons behind the patterns we observed.

So what did we learn?

First, we learned that disbelievers consistently act in a more hostile manner toward believers than the other way around. We found higher levels of affective polarization among climate change disbelievers than believers in nearly every single week of the two-year period we analyzed.

We had initially expected to see much more variation in the emotional expressions between the groups. But we realized that this finding went hand-in-hand with related work in social psychology that establishes robust differences between political conservatives and liberals in prioritizing group values and maintaining group cohesion. In other words, conflicts between groups may arise unevenly across ideological divides – in this case, these characteristics seemed to belong to climate change disbelievers.

Second, we saw that levels of affective polarization triggered different kinds of conversations among believers and disbelievers. When affective polarization was within regular levels for climate change believers, they were more likely to talk about natural disasters than disbelievers were. But when affective polarization rose to exceptional levels, it was climate change disbelievers instead who spoke more about disasters.

This finding was also intriguing to us, as it suggested another manifestation of social science findings in the online world: that of psychological reactance. Prior work has shown that those who question the reality of climate change are likely to experience intense feelings of opposition when their worldview is threatened. Discussions of natural disaster events often relate to the impacts of climate change. In these contexts, our analysis suggests that climate change disbelievers not only participate in these conversations, but also do so in a more aggressive manner. This antagonizes those who do believe in climate change, while potentially also galvanizing their own skeptical views.

So this begged the question: how do online hostilities shift with actual climate events? To investigate this, we tracked how affective polarization levels in the online conversation varied alongside officially documented long-term climate trends and anomalies. Using data from the National Climatic Data Center (NCDC), we matched our affective polarization measurements with actual changes in global surface temperature and major storms.

Vast social scientific work demonstrates that climate change may fertilize conditions for offline conflicts and violence, particularly through triggering individual-level physiological shifts conducive to aggression, and collective-level competition for resources. Building on our earlier findings, we wanted to know whether online conversations would reflect similar patterns relative to climate events in the offline world.

As before, we observed these effects play out asymmetrically. For climate change believers, levels of affective polarization generally did not increase relative to climate events. Instead, we actually saw a decrease in believers’ hostility toward disbelievers when confronted with more global temperature anomalies. In contrast, disbelievers became significantly more hostile at times with anomalous global temperatures and major storms. Moreover, we saw that the relationship between offline climate events and online hostility was three times stronger among disbelievers than among believers. This indicated that disbelievers’ hostility was much more likely to respond to offline climate events than that of believers.

Taken together, these findings tell us a story of how social media may not only reflect offline divisions, but also potentially lead to their intensification. More precisely, we find evidence that social media opens doors to heightened intergroup conflicts, especially in moments when events in the offline world threaten the worldview of one group or another.

For climate change in particular, our work points to valuable practical directions. Experts and policy-makers may need to pay attention not only to transmitting accurate scientific information about climate change, but also lowering psychological barriers like reactance among climate change skeptics. Given persistent evidence of hostility toward believers, effective communication may also entail changing the minds of disbelievers not just regarding climate change itself, but also about the agents and institutions seeking to respond to it.

Finally, while we have specifically studied climate change in this work, these insights around affective polarization may also apply to many other contentious issues linked to major divides in society. Beyond climate change then, our work thus also demonstrates a powerful way to track these dynamics in online conversations, and better understand and address them.