Fig 1.
An illustration of the news classification pipeline. With LLMs, we use the VTVNA data to generate “events,” roughly corresponding to the level of granularity of a news segment. From the events we generate “issues” relevant to a year of broadcast news and “topics” covering more than five decades of news broadcasts. We further classify segments not corresponding to an issue into a variety of subtopics; see S1 Table through S4 Table for examples.
Fig 2.
The composition of a typical 30-minute evening news broadcast between 1969 and 2024. Broadcast year is shown on the x-axis, and the proportion of time allotted to each type of content in minutes out of a typical 30-minute broadcast is shown on the y-axis. The content each band represents is indicated by the corresponding color: “other” programming not recorded in the broadcast in gray; commercials in dark purple; “empty” segments with no abstract in lavender; soft news in blue; low-quality non-issue news (see S4 Table) in turquoise; high-quality non-issue news in light green; and issue-based news in yellow. Over this 50-year period, the duration of time allotted to different types of news content changes dramatically, with increasing time spent on less substantive types of programming.
Fig 3.
Changes in coverage of selected news topics.
An illustration of the changing content of the evening news, using six of the 21 “topic” classifications. Broadcast year is shown on the x-axis, and the proportion of total news time (among non-commercial segments with abstracts) spent on these topics is shown on the y-axis. The first column highlights the declining importance of international news and news about civil rights and social movements. The second column shows the increasing salience of sensational and “soft” news topics. The third column provides evidence that topic assignments are meaningful and match expectations about changes in news content: campaign reporting increases dramatically during presidential election years, and technology reporting spikes in the late 1990s, largely covering the “electronic revolution” and concerns about the “Y2K problem.”
Fig 4.
Changes in international coverage.
The proportion of hard news segments mentioning a foreign country, weighted by duration. The red line indicates all foreign countries, and the blue line excludes mentions of Iraq and Afghanistan. Without these two countries, there is a durable drop of roughly 20 percentage points in the proportion of hard news mentioning a foreign country beginning in the early 1990s.
Fig 5.
Coverage of top 10 issues in 1969 and 2024.
Time devoted to coverage of the top 10 political issues during a typical newscast in 1969 vs. 2024. While the top issue in 2024 (the presidential election) received moderately less coverage than the top issue in 1969 (the Vietnam War), coverage of the remaining issues is significantly lower in 2024, ranging from roughly 15 seconds to two minutes, compared to 40 seconds to more than three minutes in 1969. In aggregate, significantly more time was spent covering top political issues in 1969 (19 minutes) than in 2024 (10.5 minutes).