Skip to content

Media bias detection

Media bias detection involves identifying systematic departures from balanced or neutral reporting in news media. Bias can manifest through editorial choices (what stories to cover), presentation (how stories are framed), and language (word choice and tone).

Dimensions of bias

Gatekeeping bias: Editorial selection of which stories to report. Measured by comparing coverage of political parties, candidates, or events across outlets; tone analysis (negative stories reported more frequently for one party).

Coverage bias: Amount of coverage given to parties or candidates. Quantified by article count, word count, or airtime per outlet and period.

Statement bias: Injection of attitudes or opinions into factual reporting. Detected through linguistic markers of subjectivity, sentiment, and hedging.

Framing bias: How an issue is presented—which aspects are highlighted, which downplayed. Frame analysis involves identifying generic frames (e.g., conflict, morality) and issue-specific frames in news text.

News slant: Ideological positioning of outlet language. Measured by comparing phrase frequency to known partisan speech (Congressional records); audience co-citation patterns; shared audience analysis on social media.

Detection methods

  • Linguistic analysis: LIWC lexicons, hedging cues, sentiment and subjectivity markers, rhetorical patterns
  • Content comparison: Comparing coverage of the same event across outlets; identifying divergent framing
  • Audience analysis: Inferring outlet slant from follower ideology on Twitter, Facebook; polarization measurement
  • Graph-based: Co-citation with partisan actors; network positions of outlets in media ecosystems
  • Multimodal: Including visual bias (photograph selection, camera angles) alongside text

Key papers