Wendover’s Five Rules of Risk Begin With Two Wrong Numbers

A video warning viewers not to trust their instincts about danger should at least get its own mathematics right.

Wendover Productions’ The Five Rules of Risk makes a memorable opening comparison: an American licensed driver faces a 1-in-600 annual chance of dying while driving, while a mountain biker faces a 1-in-30,000 chance of dying on the trails. The contrast supposedly exposes how badly people misunderstand danger.

There is a problem. Neither figure matches the underlying research cited in the video. Both probabilities are inflated by approximately a factor of ten, and the driving statistic counts people who were not driving.

The broader conclusion—that road travel carries greater annual mortality risk than mountain biking—survives. But the numbers used to establish it are wrong, and the video subsequently turns several tentative research findings into confident rules about human psychology.

That is not a small editorial problem. The numbers are the foundation of the argument.

The Driving and Mountain-Biking Comparison Is Numerically Wrong

The video cites a Statista series measuring motor-vehicle fatalities per 100,000 licensed American drivers.

For 2017, the underlying rate was 16.63 fatalities per 100,000 licensed drivers. That converts to approximately one fatality per 6,013 licensed drivers annually, not one in 600.

There is a second problem with the statistic: its numerator includes everyone killed in motor-vehicle traffic crashes, not just drivers.

The National Highway Traffic Safety Administration's 2017 figures distinguish:

• 18,819 drivers killed.
• 6,237 passengers killed.
• 5,226 motorcyclists killed.
• 6,075 pedestrians killed.
• 806 pedalcyclists killed.

There were 225.346 million licensed drivers.

Using driver fatalities alone produces approximately 8.35 deaths per 100,000 licensed drivers, or one in 12,000 annually. That is the more appropriate comparison when discussing the probability of a licensed driver being killed while driving.

The mountain-biking figure is similarly misrepresented.

The 2019 scientific review cited by Wendover reports an annual mortality rate of 0.0035 deaths per 1,000 people at risk in the Austrian Alps. That equals 3.5 deaths per million participants, or approximately one in 286,000.

A separate Greater Vancouver estimate was 0.002 deaths per 1,000 people, equivalent to one in 500,000.

Neither supports the video's one-in-30,000 figure.

The Austrian estimate itself is approximate: researchers used 25 deaths across 2013–2017 and an estimated 1.424 million mountain bikers in 2014. The review also reports that, in its broader Austrian fatality records from 2006–2017, 52.7% of deaths involved cardiovascular events and 33.8% involved accidents.

These figures are not simply measurements of crash risk.

The comparison also combines American driving statistics with mountain-biking estimates from Austria and Canada. It compares annual person-level mortality, not risk per hour, mile, or individual outing.

Interestingly, the relative comparison is not entirely destroyed by the mathematical errors. Using all road fatalities per licensed driver produces a ratio of approximately 48 to one against the Austrian mountain-biking estimate. Using driver fatalities alone produces approximately 24 to one.

Driving still appears substantially more dangerous by this crude annual measure.

But a video that tells viewers to evaluate danger scientifically should not present a tenfold error as a statistical fact.

The NHTSA's 2018 crash report and its Fatality Analysis Reporting System provide the figures needed to check the calculation. The mountain-biking estimates come from Stoop and colleagues' 2019 review.

Three of the Five Rules Come From One Exploratory Model

The first three rules are presented as established findings about human psychology:

1. Voluntary risks are more acceptable than involuntary risks.
2. Acceptance decreases as the number of people exposed increases.
3. Disease provides a psychological yardstick for acceptable danger.

These are not three independent discoveries.

Their principal historical source is Chauncey Starr's 1969 paper, Social Benefit versus Technological Risk, published in Science.

Starr was attempting to develop a model for understanding how society balances technological risks against benefits. His approach estimated fatalities per person-hour of exposure and assigned monetary values to the benefits of different activities.

For voluntary activities, he used average participant spending as a proxy for personal benefit. For involuntary activities, he used their contribution to annual income.

This is an extremely consequential assumption.

Spending money on an activity does not necessarily establish how much someone values its benefits. Nor does a contribution to national income establish that everyone exposed to an associated danger accepts it.

Starr acknowledged that the monetary approach was uncertain and overly simplistic.

His conclusion was explicitly exploratory. He described the evidence as indications that people accepted voluntary risks roughly 1,000 times greater than involuntary risks.

Wendover converts that tentative result into a much more confident statement: research has shown that people are approximately 1,000 times more tolerant of voluntary danger.

The distinction matters.

Paul Slovic and Baruch Fischhoff examined Starr's revealed-preference approach in their own work on acceptable risk. They identified serious limitations:

• It assumes historical behaviour predicts present-day preferences.
• It can preserve existing economic and social arrangements rather than question them.
• It assumes people make reasonably rational marketplace decisions and enjoy meaningful freedom of choice.
• It can underestimate hazards whose consequences develop slowly, including carcinogens.
• Measuring the benefits and risks consistently is difficult.

These are not criticisms invented by AngryPages. They appear in the risk-research literature that Wendover itself references.

The underlying voluntary-versus-involuntary distinction remains useful. The problem is presenting an approximate historical model as though it produced a universal psychological constant.

The Prevalence Rule Is Explicitly a Hypothesis

Starr's second rule is particularly revealing.

He compared historical participation and mortality trends in motor vehicles, commercial aviation, and general aviation. He observed that activities involving more participants tended to operate at lower accepted risk levels.

He then proposed that acceptable risk might be inversely related to participation.

His actual wording was that the similarity of the trends might provide the basis for another hypothesis.

Wendover removes that qualification and presents the relationship as a rule.

The video then extends the argument to space travel, commercial aviation, vaccines, and experimental medicines.

The vaccine comparison is especially weak.

A vaccine administered to millions of healthy people is not directly comparable to an experimental treatment given to a small group of seriously ill patients. Consent, expected benefit, disease severity, treatment stage, and available alternatives all differ.

The video offers no matched evidence establishing that population size itself explains the difference in public tolerance.

It has taken an observation about three transportation categories and extended it across fundamentally different medical decisions.

That is an interesting hypothesis. It is not a demonstrated universal law.

The Vietnam Example Does Not Establish Public Acceptance

The disease-yardstick argument receives its most striking illustration through the Vietnam War.

Wendover explains that the average hourly mortality risk for American soldiers fighting in Vietnam was approximately equal to the average hourly mortality risk from disease in the general population.

The video then concludes that the public more or less accepted the soldiers' exposure because it fell near this familiar benchmark.

The original Starr paper contains important qualifications that change the interpretation.

Starr estimated that approximately 500,000 men were exposed to the Vietnam War's mortality risk annually, with 10,000 deaths per year. He then assigned the entire American population an intangible benefit from the war's assumed $30 billion annual expenditure.

That is a modelling assumption about societal benefit, not a survey of public attitudes.

More importantly, Starr explicitly distinguished society's perspective from the perspective of those directly exposed.

For Americans aged 20–30, he calculated that the Vietnam mortality risk was approximately ten times normal mortality from accidents or disease.

He wrote that the population as a whole and those directly exposed saw the matter from different perspectives.

The distinction is fundamental.

The video acknowledges the age difference but then dismisses its significance by returning to the supposed public disease benchmark.

Yet Starr's original argument is more qualified. His disease yardstick concerns the acceptability of voluntary activities, while he treats war as an extreme example of involuntary risk, where decision-makers are separated from those bearing the danger.

The video moves between these categories without explaining the difference.

A statistical relationship between national expenditure, population-wide benefits, and military fatalities cannot establish that the public psychologically accepted the risks faced by individual soldiers.

The original source never demonstrated that conclusion.

Nuclear Power Was Not Feared Simply Because It Was New

The fourth rule states that novelty increases perceived risk and that greater familiarity reduces fear.

Wendover illustrates this through a 1970s study in which college students ranked nuclear power as the most dangerous of 30 technologies, while experts ranked it twentieth.

The comparison is real. The explanation is incomplete.

The associated psychometric research examined substantially more than familiarity.

In a 1978 study, 76 members of the League of Women Voters rated 30 technologies across nine qualitative dimensions, including:

• Voluntariness.
• Familiarity.
• Controllability.
• Potential for catastrophic accidents.
• Immediacy of consequences.
• Whether the risk was known to the public and scientists.
• Dread.
• Lethality.

Nuclear power scored extremely high on several characteristics associated with perceived danger. It was rated higher on dread than any of the other 29 technologies.

The researchers also identified its perceived lack of control, involuntary exposure, and potential for catastrophic harm.

This matters because the video reduces a complicated perception problem to a story about unfamiliarity.

The same research explained that X-rays gained acceptance through familiarity, clear benefits, and trust in the practitioners. It also cautioned that nuclear power's perceived potential for catastrophic accidents might remain difficult to change.

The distinction between experts and the public is also more complicated than the video suggests.

In their 1982 paper, Lay Foibles and Expert Fables in Judgments About Risk, Fischhoff, Slovic, and Lichtenstein found that expert assessments were more closely associated with annual fatality estimates.

But they also explained that experts and ordinary people sometimes meant different things by risk.

Experts frequently interpreted risk in terms of annual deaths. Ordinary people also considered catastrophic potential and other qualitative characteristics.

Those are not necessarily identical questions.

A person concerned about an accident capable of killing thousands at once is not simply making the same calculation as an engineer estimating average annual fatalities.

There is another problem with the video's claim that nuclear risk perception has greatly declined over four decades.

Paul Slovic's 2012 research, The Perception Gap: Radiation and Risk, reported that surveys continued to show high public perceptions of risk from nuclear power and nuclear waste.

It also identified trust in the institutions managing the technology and the perceived direct benefits of that technology as important influences.

The video provides no longitudinal survey establishing the substantial decline it describes.

Familiarity matters. But familiarity alone does not explain the nuclear debate, and education does not automatically resolve questions of trust, catastrophic potential, or involuntary exposure.

The Numbers-Are-Numbing Rule Has a Replication Problem

The fifth rule argues that people become less emotionally sensitive as casualty numbers increase.

The video uses this to explain why enormous disasters, pandemics, and genocides may receive surprisingly little additional public concern compared with smaller terrorist attacks.

There is genuine experimental research behind the concept, but the video extends it beyond what its principal study establishes.

In 1997, David Fetherstonhaugh, Paul Slovic, Stephen Johnson, and James Friedrich published Insensitivity to the Value of Human Life: A Study of Psychophysical Numbing.

One experiment involved 54 undergraduate volunteers imagining that they were government officials deciding how to allocate funding.

They evaluated hypothetical humanitarian programs designed to provide clean water to Rwandan refugees.

Both programs could save 4,500 lives. One operated in a camp containing 11,000 refugees, while the other operated in a camp containing 250,000.

Participants tended to rate the intervention in the smaller camp as more beneficial.

A second experiment involved 162 university students evaluating similar hypothetical interventions.

These were interesting results, but they were not observations of actual public reactions to genocide, terrorism, pandemics, or natural disasters.

The experiment measured judgments about hypothetical aid programs.

It did not establish that statistics are generally ineffective at communicating mass tragedies, nor that people consistently care little more about thousands of deaths than hundreds.

A subsequent development makes the broad rule even less secure.

In 2021, Ignazio Ziano and colleagues published a preregistered replication and extension in the Journal of Experimental Social Psychology.

Across five data collections involving 4,799 participants in the United States and United Kingdom, the researchers found mixed results.

Two replications supported psychophysical numbing.

Three replications of the first original experiment instead produced the opposite result: participants preferred directing aid toward the larger refugee camp.

The researchers concluded that the phenomenon depends substantially on the experimental approach used.

This later study was published after Wendover's May 2020 video, so it is an update rather than a criticism of the original publication date. But it changes how confidently the rule should be presented today.

The effect is worth studying. It is not a universal law of human compassion.

The Most Dangerous Oversimplification Is the Final Message

The video ends with a sweeping warning:

"We just can't trust ourselves."

That conclusion is too broad.

Human risk judgments are imperfect. People can overestimate unfamiliar hazards, underestimate familiar dangers, and allow perceived control to influence their decisions.

But risk perception also incorporates considerations that cannot be dismissed as simple mistakes.

Who bears the danger? Did they consent? Can they escape it? Could a single accident cause catastrophic harm? Do they trust the institution responsible for protecting them?

Those questions matter when deciding whether a technology, policy, or industrial activity is acceptable.

A fatality rate does not settle every question of public safety. Nor does a disagreement between public opinion and expert estimates automatically prove that the public is irrational.

The researchers Wendover cites understood these distinctions. Their work repeatedly discusses the limitations of statistical models, the importance of qualitative risk characteristics, and the possibility that experts themselves overlook important considerations.

The video gives viewers the rules but leaves much of that intellectual caution behind.

The Verdict

The Five Rules of Risk is built around a legitimate subject, but its presentation is more certain than its evidence.

Its opening driving and mountain-biking probabilities are incorrectly converted. Its first three rules largely derive from one exploratory 1969 risk-benefit model, whose assumptions were subsequently criticised by leading researchers. Its Vietnam example converts a societal benefit assumption into an assertion about public acceptance. Its nuclear-power example reduces multiple dimensions of perceived risk to novelty. And its final psychological rule has produced mixed replication results.

The danger is not simply that viewers might remember two incorrect numbers.

It is that they might learn to dismiss their own concerns whenever those concerns disagree with an expert ranking, without understanding what the ranking actually measures.

I agree that intuition is an unreliable instrument for estimating danger. But so is an argument built on mismatched denominators, historical assumptions, and experimental results presented without their limitations.

Wendover tells viewers not to trust their instincts.

Before asking them to do that, it should have demonstrated that its own calculations and conclusions deserved their trust.

References

1. The Five Rules of Risk — Wendover Productions — YouTube, May 20, 2020. Original video, its five claims, numerical comparisons, sponsorship disclosure, and cited bibliography.
2. 2018 Data: Summary of Motor Vehicle Crashes — National Highway Traffic Safety Administration, November 2020. Provides the 2017 rate of 16.63 fatalities per 100,000 licensed drivers and the underlying driver population.
3. FARS Encyclopedia — NHTSA. Separates 2017 traffic fatalities by person type, establishing that the total includes passengers, motorcyclists, pedestrians, and cyclists, not just drivers.
4. Mortality in Different Mountain Sports Activities Primarily Practiced in the Summer Season — International Journal of Environmental Research and Public Health, 2019. Provides the Austrian and Vancouver mountain-biking mortality estimates, their approximate population denominators, and the distribution of causes of death.
5. Social Benefit versus Technological Risk — Chauncey Starr, Science, September 19, 1969. Original source for voluntary and involuntary risk, the disease benchmark, the participation hypothesis, and the Vietnam comparison. Full text is also reproduced in this document.
6. How Safe Is Safe Enough? Determinants of Perceived and Acceptable Risk — Paul Slovic and Baruch Fischhoff, working draft, 1979. Examines Starr's methodological assumptions and documents the multiple qualitative dimensions involved in nuclear risk perception.
7. Lay Foibles and Expert Fables in Judgments About Risk — Baruch Fischhoff, Paul Slovic, and Sarah Lichtenstein, The American Statistician, August 1982. Explains why expert and public risk assessments can differ because they measure different aspects of danger, and why experts are fallible.
8. The Perception Gap: Radiation and Risk — Paul Slovic, Bulletin of the Atomic Scientists, 2012. Documents the continuing public-expert divide over nuclear risk and the importance of trust and perceived benefits.
9. Insensitivity to the Value of Human Life: A Study of Psychophysical Numbing — David Fetherstonhaugh, Paul Slovic, Stephen M. Johnson, and James Friedrich, Journal of Risk and Uncertainty, 1997. Original experimental research using hypothetical humanitarian aid decisions involving Rwandan refugees.
10. Numbing or Sensitization? Replications and Extensions of Fetherstonhaugh et al. (1997) — Ignazio Ziano and colleagues, Journal of Experimental Social Psychology, 2021. Reports five replication and extension studies involving 4,799 participants, finding both numbing and effects in the opposite direction.

Clarify the opening statistical correctionAdd a concise executive summary