Data Sources
Navigation
Hub: 00 - Fact-Check MOC · Previous: 01 - Audit Log · Next: 03 - Test Design
The two located repositories, what each publishes, and the coding layer neither one publishes.
Neither repository is mentioned anywhere in the book. The book contains no data availability statement, no methodology appendix, and four URL-like strings across 257 pages, none of them a data link. Both were located from the underlying articles.
Nation study — television
github.com/nationmag/Gaza-Media-Bias. Stored locally at data/nation-tv-dataset/.
Published alongside the October 2024 Johnson and Othman Ali study, which the book cites at Introduction n.6. Eleven CSVs, roughly 68,000 rows. Every row is a 15-second closed-caption clip from the Internet Archive TV News Archive, retrieved through GDELT, carrying an archive.org URL and timestamp.
Unusually transparent in one respect: several files expose the manual coding columns (Victim, ABOUT PALESTINIANS?) that turn raw hits into published counts. That layer is normally invisible and is the hardest part of content analysis to audit.
| File | Book use |
|---|---|
| Sunday Shows Oct 7 2023–Oct 7 2024 | Ch. 6, Fig. 6.1 |
| Emotive words Palestinians_Israelis / Ukrainians – 30 days | Ch. 2, Ch. 7 |
| Health Ministry Data – 100 days | Ch. 2, Fig. 2.4 |
| Children / Journalists / Refugees / Civilians – 100 days | Ch. 7, Fig. 7.1 |
| Gaza & Ukraine GENOCIDE_WAR CRIME – 30 days | Ch. 7 |
These files are partial exports, not the count basis
The emotive-words file covers 14 of the 30 survey days, stopping dead on 2023-10-20. Term counts within that window fall short of published figures and do so non-uniformly. The published counts almost certainly derive from GDELT’s aggregate endpoint, which returns totals the row export never delivered. Recomputing from these CSVs is therefore not a valid verification method.
Intercept study — print
github.com/theintercept/gaza-media-bias. Stored locally at data/intercept-print-dataset/.
From the January 2024 Johnson and Ali analysis of the New York Times, Washington Post, and Los Angeles Times, cited in the book at Ch. 3 n.47. The article calls this the basic dataset and states the full one is available by emailing Othman Ali at [email protected]. Requested 2026-08-01, pending.
| File | Rows |
|---|---|
| Headlines of all relevant articles surveyed | 1,138 (WaPo 484, NYT 424, LAT 230) |
| All sentences containing massacre/horrific/slaughter | 450 |
| Summary statistics of terms mentioned | 4 |
The sentences file carries no victim-attribution column, and the README notes quoted instances were removed from the published statistics but left in the raw file with nothing marking which is which. Both transformations that produce the published figures are therefore unpublished:
| Term | Raw rows | Published (Isr + Pal) |
|---|---|---|
| massacre | 216 | 125 + 2 |
| horrific | 128 | 36 + 4 |
| slaughter | 104 | 60 + 1 |
Between 41 and 69 percent of raw hits were removed by judgments that cannot be inspected. This is the gap the full dataset request is meant to close.
Not located
The book’s own survey spans a full year and eight print outlets including AP, Politico, USA Today, and Axios. Neither repository covers that. Some of the book’s data may be unpublished anywhere.
GDELT — the underlying corpus
Both studies draw on GDELT’s Internet Archive TV Explorer, which is a free public API requiring no key. The exact query strings used by the Nation study are preserved in the CSV headers. This makes the television half of the book’s evidence independently reproducible at zero cost, which is the basis for 03 - Test Design.
Useful modes: timelinevol returns percentage of airtime per time bucket; clipgallery returns clip-level records with snippets, capped at 50 per request. timelinevolraw is not a valid mode and returns an error.