We Mapped 1,904 Disappearances. The Densest Point Is A Visitor-Centre Car Park.
Nineteen months, sixty-three separate records requests, two blinded statisticians. Against park area the clusters reproduce the published atlas almost exactly. Against visitor-hours they dissolve, except at one place where nobody has ever gone missing.
A standing series in which this newspaper investigates a widely held claim and reports that it could not substantiate it.
FERN RIDGE, Idaho — Seventy-one of the 1,904 incidents in this newspaper's file share a location to four decimal places. It is a strip of asphalt about 180 metres long: the overflow row of the visitor-centre car park at Fern Ridge.
Nobody has ever gone missing there.
It is where the ranger was standing.
What we had to build first
There is no central record of people who vanish inside the national park system. That is the most quoted fact in the whole subject, it is true, and we tested it separately — the answer turns out to be an IT estate rather than a refusal.
So we asked the units. Sixty-three requests, nineteen months, three appeals. Fifty-nine units released incident records. Four sent annual counts only, and we excluded them rather than guess at what was inside.
That leaves 1,904 incidents between 1985 and 2026, each with a date, a unit, a coordinate and a disposition.
What we promised before we looked
On 11 March 2025, before a single record arrived, we posted the plan: a kernel-density estimate and a Ripley's K test, run twice.
Once against park area. A cluster is then a place with more incidents per hectare than chance would put there. That is how the published maps do it.
Once against visitor-hours. A cluster is then a place with more incidents per hour of human presence than chance would put there.
Dr. Ingeborg Falkenrath-Musa in Uppsala and Dr. Osric Banda-Whitlow in Auckland ran it separately, neither told what the points were. The closest comparable exercise in our own vault, PX-2041, made the same move on a different subject and found the same kind of answer.
Against hectares, the clusters are real
They reproduce. Seven of the eleven clusters named in Harlan Trethewey-Marsh's atlas come out of our pipeline within two kilometres of where he puts them, and two more come out a little further off.
Granite. Water. The edges of boulder fields. All of it holds.
Against visitor-hours, eight of nine dissolve
The second denominator came from 211 infrared trail counters and the backcountry permit system: 4.6 billion recorded visitor-hours across the 63 units, with fourteen counter-years thrown out for calibration drift.
Both statisticians return the same result. Every cluster but one falls inside the null envelope.
The granite units carry 4.1 times the foot traffic per hectare of the non-granite ones. The domes are where the photographs are. The photographs are where the people are. The people are where the incidents are.
The one that survived
The surviving hot spot sits 6.4 standard deviations above the envelope, and it is a car park.
In 2003 the service issued a training deck for the incident form then coming into use. Slide 14 instructs the person entering a record to fill the location field with the coordinate of the reporting party — the ranger, the desk, the trailhead station — because that is the position dispatch can confirm at the time of the call.
At Fern Ridge the desk is inside the visitor centre. The overflow row is where a ranger stands with a satellite handset, because it is the only spot on that side of the building with sky.
So: seventy-one husbands, wives, sons and hiking partners failed to come back somewhere across 340,000 hectares, and the file records the place where somebody said so out loud.
What we could not settle
Mr. Trethewey-Marsh's reply runs below and his objection is the strongest anyone made to us. Counters are bolted to trails. The people who vanish are disproportionately the people who left the trail, and a person who leaves the trail stops being counted.
We ran a sensitivity check he had not asked for, inflating off-trail hours by factors of two, five and ten. Two clusters return at fivefold. We cannot say whether fivefold is right.
He is also correct about the footnote. He found the car-park artefact himself, in his third edition in 2011, and removed the incidents from every map he published afterwards. We cited a training slide and a document number and not his page 344, and that was a discourtesy we are correcting here rather than quietly in the file.
What this desk takes from it
The atlas was not invented. Its arithmetic is sound. It divides by the ground, and the ground does not go missing. People do, and people are not spread evenly across it.
Nineteen months and £24,980 in fees and copying charges to establish that a map of vanishings is mostly a map of visitors, and that its densest square kilometre is a map of a filing convention.
The 1,904 are still 1,904. Nothing in this story brings back one of them, and this desk would rather that had been the finding.
The six o'clock edition
Three stories a morning, in your inbox before the coffee is made.
Sources & Method
We posted the analysis plan on 11 March 2025, before the first records release arrived, and did not amend it. Two spatial statisticians who do not know each other ran it separately, blind to the subject, and were told at the outset that both reports would be published whatever they said. The point file, both denominators, the code and both reports are published with this story so that anybody who prefers a different denominator can rerun it themselves.
Who we spoke to
- Dr. Ingeborg Falkenrath-Musa, Spatial statistician, Uppsala. Commissioned to run the pre-registered plan on the point file and both denominators, without being told what the points were March–July 2026 Flagged the Fern Ridge square as a probable entry artefact in her interim note, three weeks before she learned what a visitor centre was doing in her data.
- Dr. Osric Banda-Whitlow, Spatial statistician, Auckland; second blind analysis. Given the identical file and plan, with no sight of the first report and no contact with its author April–July 2026 Asked us, at the end, whether the subject was hospital admissions. It was the only question he put to us in four months.
- Incident records from 63 national park units, Unit-level incident files, 1985–2026. Requested unit by unit under freedom of information; 59 units released records, three appeals were required, four units supplied annual counts only and were excluded January 2025 – July 2026 1,904 usable incidents. The whole file, including the four exclusions and the reason for each, is published with this story.
- Trail-counter and backcountry permit data, 211 infrared counters and the permit system. Obtained from the units and from one university transport group that had already cleaned four years of it; 14 counter-years discarded for calibration drift 2025–2026 4.6 billion visitor-hours across the 63 units. Counters were placed for trail management, not for us, and that is the weakness of this story.
- Harlan Trethewey-Marsh, Compiler of the cluster atlas, six editions since 2014; retired sheriff's deputy. Sent the full draft, the point file, both statisticians' reports and our code on 29 July, with no deadline imposed July–August 2026 Replied at length. Printed entire under Right of Reply, including the part where he is right and we were discourteous.
Documents
- PX-2041 — 214 reader sighting reports plotted against published low-level route charts accepted
What we could not confirm
- The denominator, which is the whole story. Infrared counters count people on trails. The people who fail to come back are disproportionately the people who left the trail, and a person who leaves the trail stops being counted. Off-trail hours come from backcountry permits, which do not capture day use at all. We ran a sensitivity check: inflate off-trail hours fivefold and two of the dissolved clusters return. We have no principled way to say whether fivefold is right, and neither does anybody we asked.
- How complete 1,904 is. Four units gave us annual totals we could not use. Incidents thin sharply before 1998 in every unit that keeps electronic records, which the agency's own retention schedule predicts and which means the early years of our map are a map of what survived, not of what happened.
- Whether the reporting-party convention was applied uniformly at Fern Ridge. We have the 2003 training slide and we have 71 records sharing a coordinate. We do not have the keystrokes. Three of the 71 involve a person last seen in the car park itself, and those three may be exactly what they appear to be.
How Others Covered This
The same events, as reported elsewhere on the same day. We list what each outlet had that we did not, as well as what we had that they did not — including where we come off worse. Why we print this.
-
The Meridian Telegraph1,904 Gone, And The Knots On The Map Are Real
Took our point file off our own preprint and remapped it against park area, which is the thing the preprint was written to test.
Had that we did not
The point file, plotted more clearly than we plotted it. Their cartographer is better than ours and we have said so to her.
Left out
That the same file, divided by where people actually walk, dissolves eight of the nine knots — which is paragraph two of the document they took the file from.
-
ApertureThe Geography Of Not Coming Back
An essay on why granite, and on what a landscape has to look like before people will believe it takes them.
Had that we did not
The observation that the domes, the photographs and the crowds are all the same places — which is our whole finding, arrived at without arithmetic.
Left out
The 211 counters. The essay treats the denominator as unobtainable. It had been obtained, by us, and offered to them in March.
-
They Buriedthis newspaperWe Mapped 1,904 Disappearances. The Densest Point Is A Visitor-Centre Car Park.
Assembled the file the agency does not hold, pre-registered the test before the data arrived, and ran it against two denominators with both analysts blind.
Had that we did not
Both reports entire, the code, the point file, the sensitivity check nobody asked us for, and the one cluster that survived.
Left out
Our headline sells a car park as the finding. The finding is that we spent nineteen months telling 1,904 families' worst days apart from a filing habit, and we put the habit in the display type. — V. Ashcombe-Doyle, standards editor
Right of Reply
They Buried contacted Harlan Trethewey-Marsh, compiler of the cluster atlas on 29 July 2026 by post and email, with the full draft, the point file, both statisticians' reports and our code, and no deadline. Replied 14 August 2026, by letter, asking that it run entire. Printed unedited, including the paragraph about our footnote.
You have done a great deal of careful work and you have divided by the wrong thing, and I do not think you understand yet how much of your story rests on that one decision.
I will start with what is true. My atlas divides by ground. You are right that it does, I have always said that it does, and if you divide by ground you will of course find that the places with the most people in them have the most incidents in them. I have never claimed otherwise and I would be a fool to. The atlas is a finding aid. It says: here are the squares, go and look at them. It does not say the granite ate anybody.
Now the denominator. Your counters are bolted to trails. Every one of the 211. The whole subject of this atlas is people who are not on trails — who stepped off to find a place to relieve themselves, who cut a switchback, who went round a snowfield and came down the wrong drainage. The moment a man leaves the tread he leaves your denominator, and every hour he then spends is an hour your model believes did not happen. You have not measured exposure. You have measured the exposure of the people who were fine.
You ran a sensitivity check, which is decent of you, and it brings two of my squares back at fivefold. You then write that you cannot say whether fivefold is right. I agree with you. That is the honest position and I would ask your readers to notice that it is also the position that leaves the question open, and that your headline does not.
About Fern Ridge. The reporting-party convention is not your discovery. It is in the third edition of the atlas, 2011, footnote 9 on page 344, where I set out that the car park square was an artefact of the location field and removed the 44 incidents I then knew about from every subsequent map. Your newspaper cites a training slide and a document number. It does not cite the footnote. I do not much mind for myself, but it means a reader of your piece will believe I built a monument on a car park, and I did not; I found it, I said so, and I took it out fifteen years ago.
The last thing is not about statistics. I have taken telephone calls from families for nineteen years. Somebody rings because the search was called off on day nine and nobody will ring them back, and they have found a map with their son's square on it and they want to know what it means. What it means is that I will read the incident file with them, and most of the time the answer is weather, or a drainage that funnels, or a man who was sixty-eight and told nobody he had a heart condition. A null result is a thing a newspaper can print on a Tuesday. Those families have to live in it. If you are going to take the map away — and you may be right to — then I would ask you to be as careful about who answers the telephone afterwards as you have been about your envelope.
Published unedited under our right-of-reply guarantee.
How was this story?
We publish the result, whatever it is. Reader verdicts appear on the front page and in our newsroom metrics.
Readers' Letters 0
Printed at once under the name you give and read by the desk afterwards; anything unfit is removed, with a note saying so, and nothing else is ever deleted — only corrected. Letters that changed something in the story carry a mark saying so, and there are 5 of those across the archive.