Data Stories · Media & Disaster

When the Earthquake Became Everyone’s Story

The 1906 San Francisco earthquake did more than destroy a city. It entered thousands of distant newspapers, where catastrophe became spectacle, relief became local, and the same stories traveled from press to press.

Who this data represents Digitized English-language U.S. newspaper pages available in the American Stories archive for April 1–May 2, 1906. Coverage reflects what survived, was digitized, and could be segmented from OCR—not all newspapers, readers, or Americans. Frame scores are model-assisted descriptive measurements of visible language, not facts about author intent.

24,014

earthquake-relevant articles identified

13.5%

of all digitized articles were earthquake-related at the peak

21,673

distinct story clusters after grouping probable reprints

501

newspapers carried relevant coverage

At 5:12 on Wednesday morning, April 18, the ground beneath San Francisco broke. Telegraph wires, wire services, trains, and newspaper exchanges carried the rupture outward. But news did not arrive everywhere at once. A disaster that happened in seconds took days to occupy the national press.

That delay is visible because this archive contains a denominator: not just earthquake stories, but every detected article printed by the newspapers available on each date. The share devoted to San Francisco rises sharply after April 18 and peaks five days later. On April 23, roughly one article in seven and a half in the digitized sample concerned the disaster.

Figure 1 · A national story, five days in the making

Earthquake-relevant articles as a share of all detected articles, Apr 1–May 2, 1906

Attention peaked after the shock. The vertical line marks April 18. The denominator changes with the newspapers available each day, so the figure plots a share rather than raw volume. April 23 reached 13.5%.

“Relevant” means an article scored at least 60 on a separate event-relevance screen. Broad retrieval came first; model measurement came second. This keeps the denominator independent of the framing analysis.

The first draft was catastrophe

Attention is only the first layer. To measure what the articles emphasized, each relevant text was scored on eleven event-specific constructs: suffering, solidarity, government order, social disorder, economic reconstruction, religion, spectacle, race and class, technology, local connection, and optimism about rebuilding.

The first three days are dominated by two frames: human suffering and spectacle. Both score near 50 on a 0–100 salience scale. Then the register changes. As trains and telegraphs carry more detail—and communities begin raising money—spectacle falls while civic solidarity becomes the strongest frame in the relief period.

Figure 2 · From spectacle to solidarity

Daily mean frame salience, one vote per distinct story cluster; select frames to compare

The emotional center moves. Suffering and spectacle crest during Apr 18–20; solidarity rises during Apr 21–25; reconstruction and optimism become more prominent after Apr 26. Cluster weighting prevents a widely reprinted story from receiving a new vote every time it appears.

The disaster became less spectacular as it became more social.Frame shift, shock → relief → recovery

Distance changed the story

California papers and papers elsewhere did not tell the same version. Across the full post-earthquake period, nearby coverage scores 12.8 points lower on spectacle and 9.1 points higher on civic solidarity. It is also more locally personal: names, family connections, eyewitnesses, and hometown donations carry more weight.

Distant newspapers, by contrast, lean harder into dramatic destruction, government order, property loss, and technology. This is not proof that distance causes sensationalism—the California sample is smaller and composition differs—but the contrast is large, consistent, and substantively revealing.

Figure 3 · Near the disaster, solidarity; far away, spectacle

Mean salience by publication geography, post-event articles; sorted by absolute gap

CaliforniaOutside California

Geography changes emphasis. Each row connects the California and outside-California article-weighted means. These are descriptive differences, not causal estimates. 2,352 articles (9.8%) have no resolved publication state and are grouped with "outside California" by default; publication place is otherwise recovered from newspaper metadata.

What the printing network amplified

Early twentieth-century newspapers copied. Wire stories, exchanged columns, and near-verbatim reprints made a successful account travel. We identified 3,513 articles as probable reprints and grouped the corpus into 21,673 clusters. Comparing an article-weighted mean with a one-cluster-one-vote mean shows what replication added to the record.

The answer is reassuringly modest. Reprinting raises government-order and spectacle scores a little and suppresses local connection and solidarity a little, but no shift exceeds a single point on the 100-point scale. The central frame story is not an artifact of counting syndicated copy many times.

Figure 4 · Reprinting nudged the story; it did not rewrite it

Article-weighted mean minus cluster-weighted mean, salience points

Replication mostly cancels out. Positive values indicate frames amplified by repeated publication; negative values indicate frames more common among distinct stories than among printed articles.

That leaves a broader lesson about disaster news. The national press did not merely transmit an event from a ruined city to distant readers. It transformed the event as it traveled. At first the earthquake was an astonishing spectacle. Then it became a human emergency. Finally, through donation drives, hometown names, business reports, and confident claims of rebuilding, it became thousands of local stories about what communities do after catastrophe.