UC Davis

In spite of climate change being considered an existential threat to the human species, this issue receives relatively little attention in U.S. news media. In this manuscript I provide an overview of attention to the issue of climate change in media communications and address shortcomings in methods for measuring media attention to this issue. Results indicate that attention is lower than previously reported and increasing at a slower rate than previously reported.

This manuscript shows that the standard text classification method in climate change communications literature overestimates news media attention to climate change by a factor of two to three. A Support Vector Machine (eSVM) model enriched with features from an experimental Latent Dirichlet Allocation (LDA) topic model was trained on pre-labeled data ($N \approx 50,000$). This model produced substantially higher climate change story classification accuracy (F1 scores) compared to the industry standard (Boolean classification) and showed better performance than other text classification alternatives.

Applying an the eSVM text classification model on a novel database of 1.1 million news stories distributed on the front page of the New York Times (1996-2023) and via Twitter by a diverse set of news content creators (2007-2023), this manuscript provides a comprehensive analysis of climate change attention across different domains and platforms. Results from machine learning classification (eSVM) indicate that news media attention to climate change is increasing over time but at a far slower pace than previous literature suggests.

In this manuscript I show that the inflation of climate change attention in past literature is due to the diffusion or permeation of climate change as a relevant consideration in other policy topics. Using an experimental LDA topic model, I analyze the network of associations of climate change with other topic considerations, including energy, agriculture, health, the economy, and others. The experimental "guided" LDA was trained using prior information about the structure of policy topics in news media: the model was fit to produce a topic structure that closely mirrors the Comparative Agendas Project’s topic coding schema at the four-digit level by selecting highly informative keywords from pre-labeled data (tf-idf) as central to each topic. Results indicate that the issue of climate change is becoming a more relevant consideration in a greater number of policy topics over time; however, this finding holds only for "prestige" and "niche" news sources and not for "new" media sources.