USA Times Analysis: CNN’s Presidential Coverage Sentiment

Live, auto-updating. This dataset is still growing — we are continuously collecting
headlines back to January 2016 for this publication. Scores reflect data collected so far and
will shift as more headlines are added. Every number is reproducible from the linked source data.

• Universe: every headline published by CNN (cnn.com) that contains the sitting president’s
  name (case-insensitive, whole word — “Trump” for 2016–2021 and 2025–present, “Biden” for 2021–2025), collected via Google
  News search restricted to this domain and to the correct name for each date range, date-sliced to avoid missing coverage during high-volume periods.
• Scoring model: VADER (Valence Aware Dictionary and sEntiment Reasoner) — a published,
  peer-reviewed lexicon-and-rule-based sentiment model (Hutto & Gilbert, 2014), run identically and
  deterministically across every publication, administration, and headline, with no manual adjustment per outlet.
• Classification thresholds: compound score ≥ 0.05 = positive;
  ≤ -0.05 = negative; between = neutral — VADER’s standard published defaults.
• Time windows: last 7 / 30 / 365 days are computed relative to the real current date, not
  the dataset’s latest headline — “no data yet” means collection hasn’t reached that period, not that no coverage occurred.
• What is scored: the headline text only, exactly as published — not the article body, not social framing.
• Reproducibility: every scored headline, its source URL, publish date, subject, and individual score
  is listed below, so any result can be independently checked against the live article.

This measures headline sentiment, not factual accuracy or overall editorial bias.
A negative score means the language of a headline skews negative in tone — not that the reporting is inaccurate, nor
that the outlet is biased against that president. A publication can score negatively while reporting completely accurately on
genuinely negative news, and can score positively while reporting accurately on genuinely positive news. The by-administration
table above is a comparison tool, not a verdict on its own — the underlying news cycles of each presidency were not identical.

Known limitations:

• VADER can misread sarcasm, quoted speech attributed to the president himself, and domain-specific political language.
• Headline-only scoring misses tone set by article context.
• Sample size varies by period, outlet, and administration; small-sample scores carry wider uncertainty than the number alone shows.
• The “contains the president’s name” filter can pull in incidental mentions, not just president-focused stories.
• This dataset is still growing (see banner above).