EXPERIMENTS · BUILT
Which earthquakes did the news under-cover?
USGS lists every quake whatever the news does. Matching a year of them to coverage shows which got less than quakes of their size.
- Question
- Which earthquakes got more or less coverage than quakes of their size?
- Method
- Every magnitude-6+ quake in the USGS catalogue for the year, matched to the record by headline (earthquake words, the quake's place, no conflicting magnitude) the day it struck and the next; coverage is the distinct outlets whose headlines name it. A two-part model predicts coverage from magnitude, PAGER alert, felt reports, tsunami flag and how busy the news day was; a quake outside its 90% bootstrap band is above or below expectation.
- Result
- 19 of 138 quakes had any coverage in the record. 15 got less than quakes like them, 5 more. The matcher's hand-checked precision is 0.947 and recall 0.818.
- Decision
- Published as a replication under a public rule, not a discovery: disaster coverage is known to follow size and news-day crowding. The matcher is re-checked by hand on new quakes; openFDA recalls and FRED series follow with the same spine.
How it was measured
The news chooses what to cover; an event register does not. The US Geological Survey lists every earthquake, with its magnitude and a PAGER alert (its estimate of human impact: green, yellow, orange, red). This study takes every quake of magnitude 6.0 or more since 2025-09-27 and asks how many outlets in Argus's record carried it.
- Matching. An article counts when its headline, the day of the quake or the next, uses earthquake words, names the quake's place (the town or region USGS gives), and states no magnitude more than half a point from the quake's. Features, appeals and anniversaries are left out. When a headline fits several quakes in one country, it goes to the largest.
- Coverage. The distinct outlets whose own headlines name the quake. A story that mixes two quakes counts only the outlets that named this one.
- Expectation. A two-part model fitted on all the quakes: the chance of any coverage, then how much, from magnitude, alert level, felt reports, the tsunami flag, and how busy the news was that day (the score it took to reach the day's top 25). Refitting it 200 times on resampled quakes gives each quake a 90% band.
The numbers
19 / 138quakes of magnitude 6.0+ with any coverage in the record
15below the band: less coverage than quakes like them
5above the band
0.947matcher precision, 18 of 19 matches checked by hand
0.818matcher recall, on quakes the record did cover
By alert level
| PAGER alert | Quakes | With coverage |
|---|---|---|
| green | 121 | 9 |
| yellow | 7 | 3 |
| orange | 5 | 4 |
| red | 4 | 3 |
| null | 1 | 0 |
Less coverage than expected · outlets, with the 90% band for quakes like it
| Date | Magnitude | Where (USGS) | Alert | Outlets | Expected |
|---|---|---|---|---|---|
| 2025-12-06 | 7 | 2025 Hubbard Glacier Earthquake | green | 0 | 16.0 (2.1–91.7) |
| 2026-03-24 | 7.5 | 158 km W of Neiafu, Tonga | green | 0 | 9.7 (1.5–33.4) |
| 2026-02-22 | 7.1 | 57 km NNW of Kota Belud, Malaysia | green | 0 | 7.7 (1.0–23.8) |
| 2026-01-02 | 6.5 | 0 km W of San Marcos, Mexico | yellow | 0 | 6.6 (0.5–43.7) |
| 2025-10-10 | 7.6 | Drake Passage | green | 3 | 21.3 (3.7–302.1) |
| 2025-10-10 | 6.7 | 12 km SE of Santiago, Philippines | green | 0 | 4.3 (0.6–20.5) |
| 2026-03-30 | 7.3 | 51 km ENE of Luganville, Vanuatu | green | 0 | 3.7 (0.5–10.7) |
| 2025-12-27 | 6.6 | 29 km ESE of Yilan, Taiwan | green | 0 | 3.2 (0.2–23.9) |
| 2026-08-15 | 6.9 | 9 km WNW of Pematangsiantar, Indonesia | green | 0 | 2.1 (0.5–4.4) |
| 2026-05-25 | 6.9 | 34 km ENE of Calama, Chile | green | 0 | 1.9 (0.4–5.3) |
| 2026-06-16 | 6.7 | 40 km SE of Palu, Indonesia | yellow | 0 | 1.7 (0.1–5.6) |
| 2025-10-07 | 6.6 | 17 km W of Lae, Papua New Guinea | green | 0 | 1.5 (0.1–6.3) |
| 2026-06-08 | 6.5 | 18 km SW of Balangonan, Philippines | green | 0 | 1.2 (0.2–3.7) |
| 2025-10-27 | 6.5 | 164 km E of Beauséjour, Guadeloupe | green | 0 | 1.2 (0.0–7.2) |
| 2026-05-15 | 6.7 | 41 km ESE of Ōfunato, Japan | green | 0 | 1.1 (0.2–2.8) |
More coverage than expected
| Date | Magnitude | Where (USGS) | Alert | Outlets | Expected | A headline that named it |
|---|---|---|---|---|---|---|
| 2025-12-12 | 6.7 | 117 km NE of Kuji, Japan | green | 59 | 1.2 (0.0–5.1) | Japan hit by 6.7 magnitude earthquake, tsunami warning issued | World News |
| 2026-06-16 | 6.3 | 262 km SSE of Dunhuang, China | orange | 14 | 0.4 (0.0–3.9) | One dead, four injured after magnitude 6.3 earthquake in China's Qinghai |
| 2026-08-20 | 6.7 | 35 km ENE of Tambo, Peru | green | 16 | 1.3 (0.4–2.5) | Magnitude 6.7 earthquake shakes Peru's southern Andes, USGS says |
| 2026-04-20 | 7.4 | 102 km ENE of Miyako, Japan | green | 133 | 27.2 (4.3–82.3) | Waves hit Japan after tsunami warning following 7.5-magnitude earthquake |
| 2026-06-15 | 6.2 | 76 km ESE of Pondaguitan, Philippines | green | 1 | 0.1 (0.0–0.4) | Powerful quake lifts seabed by up to two metres in Philippines |
Who checked, and limits
- The matcher was checked by Claude, an AI model: every match for whether its headline is about that quake, and every unmatched quake of magnitude 6.5+ for whether the record carried it under other words. 4 unmatched quakes had coverage of their own; two of those are the smaller quake of a same-country pair, whose stories go to the larger by rule. The rule that leaves out features was written after reading the first matches, so the precision above is on the set that shaped it; new quakes are the real check.
- Coverage is read on the backfilled year: one clustering a day from Google News searches, GDELT and Reddit, a capped sample of each day's top stories in English, not everything published. "No coverage" means none in this record.
- A quake USGS names only by region ("northern Mid-Atlantic Ridge") or by an event name gives a headline little to match; such quakes are rarely news, but a real one would be missed.
- Coverage of disasters is known to follow their size and to fall on crowded news days. This is a replication under a public rule, with its misses listed, not a discovery.
Claims
What this study said in advance, recorded when first seen and scored on its due date.
| Claim | Registered | Due | Outcome |
|---|---|---|---|
| The earthquake matcher's hand-checked precision is still at least 0.9 after the next three months of quakes are checked. validation.precision >= 0.9 | 2026-09-29 | 2026-12-31 | pending |