Framing & context

The count that measures the looking

The line rose because the instruments did.

a.k.a. ascertainment bias · surveillance bias · detection bias · catalogue completeness · the reporting artefact · observation effort · overdiagnosis · the looked-harder trend

Nothing gets counted until somebody notices it, so every count of detected things is two quantities multiplied: how often the thing happened, and how well anyone was looking. Improve the second — more sensors, more screening, a cheaper test, an easier form, a wider mandate — and the series climbs with no help at all from the first, in exactly the shape a real trend would take. It is the near neighbour of the redefined measure and not the same thing: no definition changed here, no threshold moved, no rulebook was rewritten. The same rule was simply applied by more people, in more places, with better equipment, and the growth in coverage arrives on the page wearing the clothes of growth in the world.

How to spot it

Ask what had to happen for a case to enter this dataset, then ask whether that has got easier over the span of the chart. Detection almost never gets harder.
Look for the effect to be concentrated at the faint end. If the rise sits in the smallest, mildest, most marginal cases and the severe ones are flat, the sensitivity moved, not the phenomenon.
Find a version of the same quantity that has been reliably detectable for the whole window — the deaths rather than the diagnoses, the big ones rather than all of them — and plot that. It is the control the chart forgot to include.
Watch for a step where a network, a programme, a registry or a reporting requirement started. The steepest stretch of a long series is often the seam between two collection régimes.
Check the internal proportions. Real phenomena tend to keep a stable mix of large and small; a mix that drifts steadily toward the small end over decades is a story about the instruments.
Ask whether the publisher of the data says so themselves. Agencies that run detection systems usually document this in a methodology note or an FAQ, right next to the number.

The fix

Hold the ruler still. Pick a threshold the collection system could clear for the whole window — not the one it can clear today — and plot the series at that level, even though it throws away most of the data, because what it throws away is the part that is really about the sensors. If the trend survives, it was a trend. Where the publisher has measured its own coverage, divide by it and show both: cases per thousand people screened rather than cases, detections per station-year rather than detections, reports per inspector rather than reports. That turns a count into a rate against effort, which is the quantity the reader thought they were being shown. Where effort was not measured, plot it anyway as its own series beside the count — the number of stations, tests, inspectors, participating hospitals — and let the reader see the two lines rise together. And say in the caption what the count is a count of, because "recorded", "reported", "diagnosed" and "detected" are all different words from "happened", and the difference is the whole finding.

In the gallery

Dark navy board slide, masthead “Terrafix Risk Analytics · Global peril review · Earthquake book, FY27”, headlined “The ground is getting busier, and the book is not priced for it”. A subhead reads: every earthquake of magnitude 4.5 or greater that the USGS recorded anywhere on Earth, 1970–2025 — one catalogue, one magnitude threshold, one query, 290,431 events; not a sample, not a region, not a window, but all fifty-six calendar years, the whole planet, every event above the line, on an axis that starts at zero, with the threshold unmoved in fifty-six years. Filling the left of the slide, under the label “Recorded earthquakes worldwide · magnitude ≥ 4.5 · ANSS ComCat · events per calendar year”, is a column chart of fifty-six gold columns on a zero-based axis running 0 to 10,000, the columns growing paler at the left and more saturated at the right. The first three are tiny — 1970 is annotated 669 — then the series jumps to about 3,000 in 1973 and climbs unevenly across the decades: the mid-1980s near 4,500, the late 1990s near 4,000, a step up after 2004 to around 7,000, a spike above 9,500 in 2011, and a final column annotated 2025, 8,558. A dashed gold trend line runs from lower left to upper right, labelled “fifty-six years of rising counts”. Down the right-hand side, a gold-bordered card carries ×12.8 over the line “Increase in recorded M4.5+ earthquakes worldwide, 1970 (669) to 2025 (8,558)”. A card headed “Recorded events · M4.5+ · mean per year” lists 1970–1979 at 2,584, 1980–1989 at 3,976 and +54%, 1990–1999 at 4,314 and +67%, 2000–2009 at 6,056 and +134%, 2010–2019 at 7,528 and +191%, and 2020–2025 at 7,640 and +196%. A gold card headed “Read-out for the board” argues that seismicity is not stationary, that even measured from the settled mid-1970s the count has more than doubled with the last two decades running above 7,000 a year against a 1970s norm below 3,000, that the book is rated off a stationary-hazard assumption now out of date, and recommends a flat global loading on the FY27 earthquake book and a re-rating of the retrocession programme at renewal. A card headed “Why this is the whole record” says one catalogue — the USGS ANSS Comprehensive Catalog — and one threshold, M4.5, the magnitude the NEIC states as its goal for global completeness, every year since 1970, with no smoothing, no de-clustering and no modelling: a count. A final card, “Basis of measurement”, gives catalogue ANSS ComCat, threshold M ≥ 4.5, extent worldwide, span 1970–2025, events counted 290,431. Three ticked boxes run along the bottom: “Zero-based axis. Every column starts at 0. No truncation, no break, no log scale, no second axis anywhere on this slide”, “Nothing cherry-picked. All fifty-six calendar years, worldwide, every event above the threshold — not a sample, not a window, nothing set aside”, and “One threshold throughout. M4.5, unchanged for fifty-six years — the same line drawn in the same place, one catalogue, counted once”. A footnote credits the USGS ANSS Comprehensive Catalog via earthquake.usgs.gov/fdsnws/event/1, counts of events with magnitude ≥ 4.5 worldwide by calendar year, retrieved 24 August 2026, and states that the slide is a demonstration drawn by us in the manner of a reinsurance board pack, has never been published anywhere, and that the earthquake counts are real.

The 12.8-fold rise in earthquakes that vanishes above magnitude 5

The count that measures the looking The cherry-picked window