personal rate λ
0.20 / yr
The Verification Venue · an impossible number that was a category error
The Man Lightning Kept Finding
Roy Cleveland Sullivan (1912–1983), a park ranger at Shenandoah National Park, was struck by lightning seven times between 1942 and 1977 and survived every one — the most documented lightning strikes any person has survived, per Guinness World Records. The number everybody repeats is that the odds against this were one in 1033. That figure is the mistake, not the miracle.
Here is where the story usually goes wrong. Take an average American's chance of being struck — and multiply it by itself seven times, as if the seven strikes were seven independent coin-flips at the national rate. Out drops an astronomically tiny number, and the anecdote gets filed under “impossible.” But a man who spent thirty-five summers on exposed Appalachian ridgelines is not an average American drawing from the national rate, and his strikes are not independent flips. Model it properly and the impossibility evaporates. Drag the one control below and watch it go.
strikes in a 35-year career →7+ tail
P(struck at least once)
99.9%
about 1 in 1.0
P(struck 2+ times)
99.3%
plainly non-trivial
P(struck 7+ times)
55%
not one in 1033
exposure vs. average person
244,000×
expected strikes (λ·T)
7.0
Slide left to an average person and the chance of even one strike falls to background noise — that is where the “impossible” feeling comes from. Slide right and it climbs to near-certainty. The bars are the exact Poisson distribution for the current rate; the yellow ones are the “seven or more” tail.
—
m = λ·T = 0.20 × 35 = 7.0 → P(≥7) = 1 − Σk=0..6 e−m·mk/k! = 55.0%
The model is a Poisson process: rare events arriving at a steady personal rate λ strikes per year, over a career of T = 35 years. The expected number of strikes is m = λT, and the chance of k or more follows straight from it. That is the whole engine — and it is a completely different animal from multiplying a national average by itself seven times.
The famous number, and why it is wrong
An average American's chance of being struck at least once in a 35-year span is about 2.9×10−5. Treat seven strikes as seven independent copies of that and you get:
( 2.9×10−5 )7 ≈ 1 in 6×1031
That is the same species of astronomically tiny number as the widely-repeated “one in 1033” figure (attributed to a university statistician and reprinted everywhere). We recompute our own version above so you can see the machinery — but every such calculation is the wrong model. It assumes an average person's rate and seven independent draws. Sullivan was neither average nor independent. The number isn't his odds; it's a category error.
Why the naive number is off by thirty orders of magnitude
General-population odds are real, but they are averages. The US National Weather Service puts a single year's odds at about 1 in 1,222,000, and an 80-year lifetime at about 1 in 15,300 — and those two figures are consistent with each other (the check below shows it). Crucially, the NWS itself says these vary enormously with personal exposure. They are the odds for a person drawn at random from the whole country, most of whom spend most thunderstorms indoors. They were never Sullivan's odds.
A ranger on high, open ridgelines through thirty-five Blue Ridge thunderstorm seasons has a personal rate λ that is orders of magnitude higher. A crude physical estimate — the lightning-protection collection-area method, a person modelled as a point of height 1.8 m gathering flashes from an area π(3h)² ≈ 92 m² at a Blue Ridge flash density — already lands near 10−4 strikes per year, hundreds of times the national average. Push the exposure slider and the “impossible” probability doesn't inch up; it detonates. By the time the personal rate reaches the level the record itself implies, at-least-one is a certainty and even seven-or-more is a coin flip.
The conditional turn. There is a second reason independence is the wrong model. Every strike Sullivan took is evidence his rate was high. Having already been hit several times, the maximum-likelihood estimate of his rate is simply strikes-so-far ÷ years-so-far — and the chance of the next one stops being astronomical. You don't compute the eighth strike as if the first seven never happened. Condition on what you know, and the tail is ordinary.
The check — every number recomputed in front of you
The probabilities are the exact Poisson formulas below, evaluated at the current rate. Where a figure comes from an outside authority (the NWS base rates, the survival fraction), the green column is the cross-check — not a claim.
| quantity | formula | computed | source figure |
|---|---|---|---|
| NWS lifetime from annual | 1 − e−80λ | 1 in 15,275 | 1 in 15,300 ✓ |
| naive “impossible” | (2.86×10−5)7 | 1 in 6.3×1031 | “1 in 1033” species ✓ |
| P(≥1) at λ=0.2 | 1 − e−m | 99.91% | — |
| P(≥2) at λ=0.2 | 1 − e−m(1+m) | 99.27% | — |
| P(≥7) at λ=0.2 | 1 − Σ0..6 pmf | 55.03% | — |
| survive all seven | 0.907 | 47.8% | NWS ~90%/strike ✓ |
Run it yourself: node research/the-man-lightning-kept-finding/verify.mjs.
The other “impossible”: surviving all seven
Being struck seven times is one thing; walking away seven times sounds like a second miracle. It isn't. The NWS reports that about 90% of lightning-strike victims survive. Most strikes are not the Hollywood direct hit down the crown of the skull. Per the injury-mechanism estimates of Cooper & Holle (carried by the NWS and the National Lightning Safety Council), direct strikes are only a small minority of casualties; most people are hurt by ground current and side flash — the survivable modes, where the current arcs over wet skin or passes through the ground rather than through the heart.
Casualty-distribution estimates, not physical constants — they trace to Cooper & Holle's medical literature and are quoted as ranges. The point is only the ordering: the deadly direct strike is the rare mode.
At the NWS figure of 90% per strike, surviving all seven is 47.8% — better than even. Surviving seven strikes is, under the real physics, roughly a coin flip. Not a miracle: a consequence of most strikes being the modes people live through.
What the record verifies — and what it doesn't
What is solid. Roy Sullivan was a real person (1912–1983). Guinness World Records recognises him as holding the record for the most documented lightning strikes survived — seven, across 1942 to 1977. The park kept documentation, and at least the 1969 strike has physical corroboration: a later superintendent, Robert Jacobsen, described finding the lightning marks on two trees by Sullivan's truck. Colleagues remembered the burns. The record — most-struck documented survivor — is real and Guinness-backed.
What is not independently established. No reliable source cites an eyewitness to the moment of any strike, and no hospital or medical records for the individual strikes are cited anywhere reputable — so the popular claim that the strikes were “medically documented” runs ahead of the record. The strikes were compiled by park superintendent R. Taylor Hoskins, who was never present at any of them; his documentation rests substantially on Sullivan's own detailed accounts, not first-hand verification. The colourful details — hair repeatedly catching fire, his wife struck once while hanging laundry — ride on that same testimonial chain. The record is real; the certainty of each individual strike is not. Reporting the “one in 1033” as his true odds would manufacture a new myth on top of it, which is why the number above is shown only to be dismantled.
Sullivan died in 1983; his death was ruled a suicide, and those who knew him spoke of the loneliness his reputation brought — people kept their distance for fear of the lightning. We note it plainly and without embellishment. If you are struggling, the 988 Suicide & Crisis Lifeline (US) is available 24/7 by call or text.
What's exactly true, what's idealised, and every free choice
Exactly true. The Poisson formulas are exact: P(≥1)=1−e−m, P(≥2)=1−e−m(1+m), P(≥7)=1−Σk=0..6e−mmk/k!, with m=λT. The NWS annual and lifetime base rates are mutually consistent to within rounding. The naive product and the 0.907 survival figure are arithmetic. All of this is recomputed in the verifier.
Idealised. A Poisson process assumes a constant rate and independent increments. Real exposure isn't constant — storms cluster in summer afternoons, and a strike changes behaviour. The model is a deliberately simple, honest replacement for the even more wrong independence-of-seven-national-averages model; its job is to show the impossibility is a modelling artifact, not to nail Sullivan's true rate.
Named free choices. The career length T = 35 years is 1942–1977. The collection-area anchor uses height h = 1.8 m and a Blue Ridge ground-flash density of order 6 flashes/km²/yr (regional maps run roughly 4–8) — approximate, and it only sets a reference tick. The “record's own rate” λ = 7/35 = 0.2/yr is the maximum-likelihood estimate fitted to the outcome — it is not an independent prediction, and the page says so. The honest, model-independent claim is the narrow one: any rate even modestly above the population average makes repeat strikes unremarkable, so the “one in 1033” figure is the wrong model. Getting to seven still requires a personal rate far above even the physical ranger estimate — which is exactly why Sullivan holds a world record and an ordinary ranger does not.
Representative, not universal. The mechanism percentages (ground current ~50–55%, side flash ~30–35%, direct ~3–5%) are casualty-distribution estimates from Cooper & Holle, quoted as ranges; the live NWS page describes them qualitatively. The ~90% survival fraction is an NWS population figure. Individual strike dates (Guinness lists April 1942, July 1969, July 1970, April 1972, 7 Aug 1973, 5 June 1976, 25 June 1977) rest on the park's compiled documentation.