Opinion editor's note: Star Tribune Opinion publishes a mix of national and local commentaries online and in print each day. To contribute, click here.
•••
Public trust in American media has plummeted since the 1990s. Most of this decline is among conservatives, spurred by Republican charges of liberal bias from avowedly nonpartisan outlets. Such claims are hard to assess fairly: Stories viewed by one party as following the facts are often seen by the other as ideological.
Most public estimates of news sources' partisan leanings rely on subjective ratings. Political scientists seeking an objective approach have used the language in politicians' speeches to set a baseline and compared stories with that. However, most studies in this vein look at the period before 2016 (that is, before Trump); do not discriminate between politics and other topics, and focus on either TV or written journalism, but not both.
In an effort to provide a measure of partisan slant that is comprehensive, impartial and up-to-date, we at The Economist have applied this academic approach to the output in recent years from a wide range of news sources. We find that there is indeed an affinity between the media and the left, because journalists tend to prefer the language used by Democratic lawmakers.
Moreover, this disparity has grown since the start of Donald Trump's presidency. As a result, the number of media sources covering politics in balanced language has dwindled.
The first step in our analysis was compiling a partisan "dictionary." We took all speeches in Congress in 2009-22 and broke them up into two-word phrases. We then filtered this list to terms used by large shares of one party's lawmakers but rarely by the other's. The result was a collection of 428 phrases that reliably distinguish Democratic and Republican speeches, such as "unborn baby" vs. "reproductive care" or "illegal alien" vs. "undocumented immigrant."
Next, we collected 242,000 articles from news websites in 2016-22, and transcripts of 397,000 prime-time TV segments from 2009 to 2022. We calculated an ideological score for each one by comparing the frequencies of terms on our list. For example, a story in which 0.1% of distinct phrases are Republican and 0.05% are Democratic has a conservative slant of 0.05 percentage points, or five per 10,000 phrases.