Eleven dead researchers, a supposed pattern: What statistics say

Contents

For several months, lists and charts have been circulating on social media platforms, compiling deaths and missing persons reports from the past three years: Eleven scientists in the USA – all with connections to security-relevant or politically sensitive topics – are said to have died or disappeared under sometimes unexplained circumstances. Behind the lists are real biographies: a nuclear physicist at MIT, an astrophysicist with a long-standing connection to NASA's Jet Propulsion Laboratory (JPL), a project manager who went missing after a hike, an anti-gravity researcher from Alabama, several aerospace and defense engineers, and administrative employees.

Four of the eleven cases are clearly classified as homicides or suicides, others as natural deaths or missing persons cases. The question is whether they add up to more than the sum of individual fates. Also striking is what does not connect the cases: they are spread over several years, different states, and diverse institutions. The causes of death range from heart attack to suicide to missing persons cases after hikes. A coordinated series typically leaves a recognizable pattern in location, time, and method – which is completely absent here.

The circulating graphics heavily use logos: JPL, Los Alamos National Laboratory (LANL), Kansas City National Security Campus (KCNSC), Air Force Labs. The connections between the individuals and institutions are real – however, these are also large US employers in the research sector.

Institution	Employees (rounded)
NASA JPL (Pasadena)	approx. 5,500
Los Alamos National Laboratory	approx. 16,000–18,000
Kansas City National Security Campus	approx. 7,000

If one includes Air Force research facilities and involved companies, one quickly reaches 30,000 to 50,000 employees in the environment from which the eleven cases originate.

Mortality in working age: What CDC data says

The US health authority CDC reports annual mortality rates for the age group 20 to 64 years of roughly 0.3 to 0.6 percent, depending on the age band. However, research jobs in security-relevant areas are rarely “nine-to-five”: high technical demands, tight budgets, sometimes strict secrecy, political expectations – all contribute to chronic stress. Occupational health studies show that occupational chronic stress measurably increases the risk of cardiovascular diseases, that high demands with low control over working conditions correlate with increased suicidal ideation, and psychological stress and addiction problems indirectly affect the risk of accidents and suicide.

The following sample calculation uses the average of 0.4 percent – this is a rather conservative estimate for this group, as heart attacks and suicides are statistically more likely to occur in high-stress professions than on average. For an institution the size of JPL with around 5,500 employees, the picture is as follows:

Cause of Death	Rate per 100,000/year	Expected cases/year for 5,500	Expected cases in 3 years
All causes of death	approx. 400	approx. 22	approx. 65
Cardiovascular	approx. 100	approx. 5–6	approx. 15–18
Accidents total	approx. 60	approx. 3	approx. 9
Firearms total	approx. 12	approx. 0.7	approx. 2
of which: Homicide/Manslaughter	approx. 5	approx. 0.3	approx. 1
of which: Suicide by firearm	approx. 7	approx. 0.4	approx. 1–1.5

These are not abstract numbers – behind each of these cases is a person. Statistically, they are nevertheless to be expected: Even with an employer of this size, one murder and one suicide would be expected within three years.

Why clusters are not automatically suspicious

The excitement surrounding the “missing scientists” is fueled by an intuitive impression: several deaths, similar professional fields, short period – this cannot be a coincidence! But that is exactly what it is – with a high probability.

In a population of 50,000 employees in security-relevant research, with a mortality rate of 0.4 percent per year, around 200 deaths per year are to be expected – thus around 600 over three years. These 600 cases are not evenly distributed across all institutions, professional groups, and causes of death. They inevitably form clusters: sometimes several people die of heart attacks in a particular lab, sometimes suicides increase in a certain segment, sometimes people with similar research topics are coincidentally affected. This looks like a pattern – but it is not necessarily one. A real pattern cannot be derived from retrospectively compiled lists, but only proven through investigations: through concrete connections between perpetrators, victims, and motives. These are not available here.

The model that describes this behavior is called Poisson distribution, named after the French mathematician Siméon Denis Poisson. The core message: Even with completely random, independent events, clusters arise. They are not an indication of a hidden cause, but a mathematical inevitability.

The mathematics behind it

The Poisson distribution describes the probability that an event occurs exactly k times in a period, when it is expected to occur on average λ times:

λ is the expected value – in the case of 50,000 employees, it is 600 deaths over three years. k is the actually observed number – i.e., about the eleven cases from the circulating lists. e^⁻λ is, simply put, a normalization factor that ensures all probabilities add up to 1.

Two properties are crucial for our case: First, the expected value and variance are identical – both equal to λ. The larger the population, the greater the natural range of fluctuation. Second, the standard deviation is √λ – so for λ = 600, it is around 24.5. Between 575 and 625 deaths in three years would therefore be unremarkable. A possible pattern would only be signaled by the Poisson distribution if the observed number exceeds the expected value by several standard deviations – i.e., roughly above 670. Eleven cases are far below that.

What this means for individual causes of death

The model becomes particularly insightful when applied to individual causes of death. For an institution the size of JPL, the expected value over three years is around one murder (λ = 1, standard deviation √1 ≈ 1) and around one to two suicides (λ = 1.2, standard deviation √1.2 ≈ 1.1). Zero or two cases per category would therefore be equally unremarkable – that is the natural fluctuation. Statistically significant, in the sense of a probability below two percent, would only be from four cases per category in three years.

Why clusters are unavoidable

Scaled up to the entire community of 50,000 employees, around 600 deaths are expected over three years, including statistically about 15 homicides and 40 suicides, spread across dozens of institutions and research fields. In this quantity, clusters inevitably arise: groups of cases that coincidentally share similar characteristics – the same employer, the same research topic, the same cause of death.

The human brain is designed to recognize precisely such clusters and interpret them as significant, even when they have arisen purely by chance. Anyone who, in retrospect, selects from hundreds of deaths precisely those that appear particularly dubious or thematically related will always find a seeming pattern. The circulating lists contain eleven cases, less than two percent of the expected deaths in this group. The question is therefore not why so many researchers have died – but why precisely these eleven are perceived as a pattern.

Selection Bias: When a list promises more than it likely can deliver

The circulating graphics and lists are not a random sample of all deaths in these institutions, but the result of a retrospective selection. Cases with spectacular circumstances such as murder, suicide, missing persons, or unclear causes are preferentially listed, along with symbolic employers like JPL, Los Alamos, or Air Force Labs, and with topics that already ignite the imagination: UFOs and UAPs, fusion energy, AI, anti-gravity.

More common cases – such as cancer in retirement or heart attacks beyond the age of 60 – are not counted, although statistically they account for the majority of expected deaths. In statistics, this is a classic example of selection bias: the sample is deliberately or unconsciously constructed to maximize the impression of a pattern. Statements about probabilities based on such a sample are misleading.

A look at China: a separate story

Online, the US cases are often linked with five Chinese AI researchers who died early in recent years – as if it were a connected global series. Public reports cite causes including heart attack, acute illness, and altitude sickness. However, the connection is constructed: the Chinese and US cases share neither employers nor research fields nor demonstrable operational connections.

In themselves, the Chinese cases are also not statistically unusual. China's AI industry employs tens of thousands of researchers, often under extreme working conditions. Individual early deaths in this context are more an expression of a broader problem – high work stress and inadequate health and safety protection – than an indication of targeted action. Some even link the cases to Liu Cixin's science fiction classic “The Three-Body Problem” – in the novel, a highly developed alien civilization selectively eliminates top Earth scientists. But even for the deaths among Chinese researchers, the narrative's drama is convincing – the data situation, based on current knowledge, does not support it.

Likely tragic individual cases rather than a pattern

The available data, the structure of the affected institutions, and statements from investigative authorities collectively point to random clusters in a large population – not a coordinated series of killings. The relevant work environment encompasses tens of thousands, and in a broader definition, hundreds of thousands of employees in security-relevant areas. For this group, hundreds of deaths are expected over a few years – unfortunately, including statistically, murders, suicides, and spectacular accidents. The eleven US cases form a small subset of this population, selected based on conspicuous criteria. Where investigative authorities have closed cases, personal or random backgrounds have emerged – no indication of an operational link.

This does not diminish the tragedy of the individual fates. However, for assessing possible patterns, it is crucial to distinguish between emotionally understandable concern and statistically robust statements.

Important: This article does not claim that there are no connections; this cannot be proven or ruled out based on the available data. It only shows: What appears to be a sinister pattern can also be explained by known statistical laws.

(vza)