10 Scientists Die in a 3 Year Span. Is it Random?

"Nothing in Nature is random. A thing appears random only through the incompleteness of our knowledge." – Baruch Spinoza (Dutch philosopher)

What we know.

The Trump administration is looking into reports that at least 10 American scientists — many of whom were researching UFOs or nuclear power — have either died or mysteriously disappeared since mid-2023.

Authorities have not established any concrete connection among the cases, and there have been no public allegations of foul play but some lawmakers have called for closer scrutiny as high-profile incidents draw attention.

So is there a way to test if this is just random?

Yes, there are several well-established statistical tools for exactly this kind of problem. Here's how analysts would approach it:

1. The Poisson Distribution Test

The Poisson distribution is a way of figuring out how often something should happen. Usually to determine if it's happening too often. Imagine you work at a call center that gets an average of 5 calls per hour. Some hours you'll get 3, some hours you'll get 7. We might call this a normal distribution or normal randomness. The Poisson distribution tells you exactly how likely each of those outcomes is. The idea being that the occurrence of 20 calls, would be so unlikely that now you should investigate to determine the unusual circumstance prompting the calls."

Poisson Distribution is the most direct tool. So in our case the question is: given the base rate of deaths/disappearances among similarly-situated professionals, how likely is it to see 10+ in ~3 years by chance?

• The U.S. employs roughly tens of thousands of scientists with classified clearances.
• We would have to calculate the expected number of deaths and disappearances in that population over that time period using actuarial tables.

The huge challenge: we don't have a clean denominator. How many "scientists with access to classified nuclear or aerospace material" are there? Without that, the calculation is very loose. There are some fallacies that we need to be careful of as well when we look at distributions of events.

2. The Texas Sharpshooter Fallacy

The Texas Sharpshooter Fallacy describes a sharpshooter firing randomly at a barn, we can assume relying on their intuition and skill. Only unlike how sharpshooters usually put up a target in advance the sharpshooter instead goes out and draws the target around the bullet holes afterwards — declaring himself or herself a crack shot.

• With the scientists: it's likely this type of assessment is going on where someone looked at a large pool of deaths and disappearances and selected some that fit a pattern. So the perceived "cluster" may be an artifact of selection, not a real signal.

This is one of the most common errors in conspiracy-adjacent pattern recognition.

3. Did these events happen suspiciously close together?

Scientists have tools to detect when bad things are bunching up or clustering. Like Knox Test, or Scan Statistics. These tests were originally designed to spot disease outbreaks, but they work on any kind of event.

• Knox Test: Measures whether events are closer together in both time and space than expected by chance.
• SaTScan (Kulldorff's scan statistic): A standard public health tool that identifies geographic or temporal clusters without predefining the cluster boundary — avoiding the sharpshooter problem.

Think of it like this: if 10 car accidents happened on the same street corner within a week, that's a very different story from 10 accidents spread randomly across a whole city over three years. These tests measure exactly that — are these events unusually bunched together, or are they actually pretty spread out when you look at the full picture?

If the deaths and disappearances are genuinely packed into a tight time window and happen to people in the same circles, that's a real statistical warning sign. If they're loosely scattered over years with little in common, that's much less alarming. If the deaths cluster significantly in time beyond what chance predicts, this would be a real statistical flag.

4. The "Linkage" Problem

Even a real pattern doesn't tell you why. People often skip the reality that even if the math confirms something unusual is happening, that doesn't tell you it's cause.

Noticing lots of people in one neighborhood getting sick tells you something is going on — but it doesn't tell you if it's the water supply, a shared restaurant, or something else. You'd need to find the thing they all have in common before you can say they're truly connected.

A pattern is a question, not an answer.

The Bottom Line

The pattern of scientist disappearances is intriguing but not yet statistically demonstrable with public data. A proper test would require the FBI or a government body with access to the full population of similarly-cleared scientists to run the numbers — which is exactly why investigators with classified access are better positioned to evaluate this than the public is.

Which brings up a more interesting question about institutional trust. Which is at all time lows. How is it built? And once it's lost how does it get re-built? In this case if the Government comes back with an answer that doesn't satisfy the salacious stories our brains might be making with this data, will it just create even more of a rift. Are government officials just damned if they do. Damned if they don't?

Sources

1. Kulldorff, Martin. "A Spatial Scan Statistic." Communications in Statistics — Theory and Methods, vol. 26, no. 6, 1997, pp. 1481–1496.
2. Knox, E. G. "The Detection of Space-Time Interactions." Applied Statistics, vol. 13, no. 1, 1964, pp. 25–29.
3. Gelman, Andrew, and Deborah Nolan. Teaching Statistics: A Bag of Tricks. Oxford University Press, 2002.
4. Kahneman, Daniel. Thinking, Fast and Slow. Farrar, Straus and Giroux, 2011.