Method and Domain

The four publications this study examines, The Guardian, The New York Times, The Washington Post, and AP News, were chosen due to their popularity and prestige in the U.S. All four outlets also hold either a middle or left-leaning bias rating from Ad Fontes Media. This study examined four articles from each source, so 16 total. The domain of this study spans from late March to late May, 2026. In late March, the Iran war was about one month old and Trump’s voter approval ratings were starting to reflect Americans' growing displeasure with the conflict. Throughout April and May, new coverage showed newer polls with increasingly pessimistic numbers for Trump. Although the news continues to cover Trump’s voter approval ratings, the domain of this study ends in late May, when coding began. 
 
Six categories were used to code each article: negative sentiment, attribution of responsibility, political partisanship, political foresight, comparisons, and finally, whether the Epstein files or Jeffrey Epstein are mentioned. The first four of these categories have been adapted from previous scholarship on news framing of Trump and Trump-related matters (Kang & Yang, 2022; Famulari, 2020; Entman, 1993). The final two were developed after beginning coding and noticing framing patterns. 
 
Negative sentiment is broken into two parts: negative words used to describe Trump’s polling data and negative words used to describe Trump himself, his parties, or his actions as president. This category was adapted from Kang and Yang (2022)’s use of the negative sentiment news frame. 
 
Attribution of responsibility identifies a particular cause (such as the economy) for Trump’s voter approval ratings. This category was adapted from Famulari (2020)’s attribution of responsibility news frame. 
 
Political partisanship frames Trump’s voter approval ratings as a partisan issue by acknowledging that his approval ratings are different among Democrats and Republicans. Political foresight is characterized by looking ahead to the midterm elections and how Trump’s approval ratings might affect them. Also adapted from Kang and Yang, these two politics categories were modeled after their use of the politics frame. 
 
Comparison framing puts Trump’s current approval ratings into context by bringing up approval ratings from earlier this term, his first term, or any previous president’s approval ratings. This category was added after several articles framed Trump’s current approval ratings not in a vacuum, but along with a reminder or reminders of what Trump’s approval ratings and those of previous presidents have looked like in the past. 
 
The mentions of Epstein category was added after article after article neglected to mention or allude to the Epstein files or Jeffrey Epstein at all. This seems like a significant omission considering every single article includes some attribution of responsibility, listing at least one potential cause for Trump’s low voter approval ratings. 

Each article was coded for its text, not its visuals or headline. Coding for all categories was calculated by prevalence judged by mentions per article. For instance, if an article used negative descriptors for Trump’s polling numbers, the negative sentiment frame was considered present. The more negative descriptors used, the more prevalent the negative sentiment frame was considered.