After the Storm: How Hurricanes Change Radio Speech about Climate Change

• In Event: Elites in the News

Abstract

In this paper, I dig into the effects of hurricanes on the way climate change is discussed on US talk radio. Theories on personal experience in politics suggest that living through climate-related events should increase Americans' belief in climate change (Egan and Mullin, 2012). By analyzing how speakers on the radio react to big weather events, I show how much potential the media have to change people's understanding of these events. I do this using a new type of data: several months worth of transcripts from 700 talk radio shows. As radio is a conservative-leaning medium in the United States (The Center for American Progress and Free Press, 2007), these data give us a unique look at climate skepticism among elites. Using a mix of statistical approaches and human understanding, this study also breaks new ground in terms of the sheer amount of talk radio content it analyzes.

Research questions: (1) do hurricanes affect how often climate change is mentioned? (2) do they nudge the slant of these climate change mentions from skepticism toward concern? (3) are the answers to these questions different for left-leaning versus right-leaning shows? (4) are the answers different depending on whether the show is produced and broadcast near the event's area of impact, or far away from it?

Data: In their raw form, the data are the result of a speech-to-text algorithm developed by the Laboratory for Social Machines (LSM, MIT Media Lab), applied continuously to the live internet streams of a set of radio stations. In the period of interest--August through October 2018--about 160 stations were monitored. Two hurricanes made landfall in the continental United States during this period: hurricane Florence and hurricane Michael. In collaboration with LSM, I connected parts of the transcripts to names of radio shows, based on station schedules we scraped from the web. This means that for each day, we know which words were said during which radio shows for each of the stations the algorithm was listening to.

Method: This paper combines machine learning methods with classical social science approaches. I use a support-vector machine (SVM) to classify the ideological slant of radio shows based on the bag-of-word features of their transcripts. As a training set, I use the full available transcript of 12 shows with known political leanings. I validate the classifier by leaving one show at a time out of the training data, and letting the model classify that show. Currently, the classifier correctly labels the left-out show in 10 out of 12 cases. The trained SVM can then be applied to any show not yet seen by the classifier.

Next, I search for mentions of climate change on each show, in the weeks before and after each hurricane started receiving broad attention (based on Google Trends). Hurricanes increased the total number of mentions by a factor of about 2.5. Workers on Amazon Mechanical Turk then code these mentions as either skeptical or concerned, based on the 30 seconds of audio surrounding them. Finally, I investigate how hurricanes change the number and slant of these climate change mentions, depending on the show's leaning and where it is aired.

Conclusion: Building on a newly developed data set of radio transcripts, and bringing in methods from both data science and social science, I present an analysis of elite climate discourse. This way, I hope to show how media messages can filter the effects of people's personal experiences with political issues.


References:

Egan, Patrick J., and Megan Mullin. "Turning personal experience into political attitudes: The effect of local weather on Americans’ perceptions about global warming." The Journal of Politics 74, no. 3 (2012): 796-809.

The Center for American Progress and Free Press. The Structural Imbalance of Talk Radio. 2007. http://cdn.americanprogress.org/wp-content/uploads/issues/2007/06/pdf/talk_radio.pdf

Author

• Clara Vandeweerdt, Massachusetts Institute of Technology