Where and how risk actually concentrates across geography, activity, and time and where the data itself falls short.
Steven Speilberg once said his greatest regret was framing sharks as deadly. This analysis trades the Jaws for a data analysis of 6,890 recorded incidents spanning three centuries.
Counts are on the rise, but largely in part by reporting coverage and beach use, not per-capita risk. Death by shark attack is down for the count.
Each claim is paired with the caveat it deserves. None of this is risk in the P(attack | hour in water) sense, GSAF has no exposure denominator.
Despite an uptick in raw attack counts (which rise with reporting and beach use), the share of attacks ending in death declines across the decades. This is consistent with faster medical response and beach patrols, not any change in shark behavior.
Surfers are bitten more often but survive most encounters (boards as flotation, proximity to shore, frequent mistaken-identity bites). Swimmers are more often in deeper water without flotation. Divers not considered in dataframe.
The USA, Australia, and South Africa recorded highest number of incidents. All countries with coastal recreation, substantial reporting infrastructure, and longest GSAF coverage. High counts ≠ high per-swimmer risk.
In the logistic regression, these three categories carry the largest positive coefficients on fatality. Species is also vulnerable to reporter bias; 45% missing, and witnesses aren't ichthyologists.
Almost certainly exposure-driven (surfing, spearfishing, commercial fishing demographics) rather than a shark-side effect. Nothing in this analysis or other literature suggest a gendered preference.
Two lines tell competing stories. One is mostly reporting infrastructure catching up. The other survives that bias.
Beach population, reporting coverage, and GSAF's collection practices grew over this window. Treat as a lower bound on recorded activity, not a risk curve.
Better reporting should raise the share of non-fatal attacks captured. A falling rate across decades is therefore hard to explain as pure reporting bias.
The conclusion isn't that sharks changed. It's that medical response, evacuation, and beach patrols did. — Finding i.
Within-attack fatality rate by activity. Note the inversion: surfers top the count column but bottom the fatality column.
Left bar: share of attacks recorded during that activity. Right bar: share of those attacks that were fatal.
Counts, not risk. Every country on this list is one with extensive coastal recreation and long-running reporting — the absence of a country tells you more about its reporting coverage than its sharks.
USA and Australia alone account for roughly 58% of all recorded incidents. South Africa is a distant third.
Row-normalized: given an attack during a given activity, how often was each species identified? Unknowns are included because pretending otherwise would mislead — 45% of the species field is missing.
Subject to heavy misidentification. Treat as suggestive, not definitive.
Fatality ~ activity + species + country. Rare species (< 30 attacks) dropped to avoid perfect separation. Sensitivity check restricted to unprovoked attacks holds the same direction on the main coefficients.
Positive bars raise fatality odds; negative bars lower them. All relative to the reference category in each group.
The regression is associational, not causal. Activity, species, and country are entangled: surfers in California meet different sharks than spearfishers in Papua New Guinea. Methodology note