Stand-Ins for the Living

Before the dummy, there was the elevator shaft. In the late 1930s, at Wayne State University in Detroit — the beating heart of the American car industry — researchers confronted a problem that sounds absurd until you realize no one had ever seriously tried to solve it: nobody knew what a human body could withstand. Cars were killing people in growing numbers, and the prevailing wisdom, stated on the record by manufacturers well into the 1950s, was that a crash simply could not be made survivable. The forces were too great. Nothing could be done.

The Wayne State researchers set out to find whether that was true, and the only instrument available to them was the human body itself. There was no reliable data on how a skull fractured, how a chest crushed, how much deceleration a spine could take. Biomechanics barely existed as a science. So they improvised, with a grimness that is hard to read comfortably today: they dropped steel ball bearings onto cadaver skulls to measure fracture thresholds. They fitted bodies with crude accelerometers, strapped them into cars, and staged head-on collisions and rollovers. And, famously, they dumped cadavers down an unused elevator shaft onto steel plates to learn what a body does when it stops very suddenly.

It was macabre. It was also the beginning of the single most successful public-health engineering project of the twentieth century — and it started with the dead standing in for the living, because the living could not volunteer for what needed to be learned.

What the dead taught us

The moral arithmetic of that work was eventually calculated, and it is staggering. In 1995, Albert King — a Wayne State professor who had been doing cadaver research since 1966 — published a paper in the Journal of Trauma with the deliberately provocative title "Humanitarian Benefits of Cadaver Research on Injury Prevention." King's conclusion was that the safety designs validated through cadaver testing, counting only changes made through 1987, were saving roughly 8,500 lives every year. Broken down per cadaver used, the numbers become almost impossible to hold in your head.

More than sixty lives, every year, for each body that consented to be studied. There is no other line of research in the history of the automobile — arguably in the history of medicine — with a return like that.

But cadaver work was never clean, in either sense. The bodies that were available skewed heavily toward older men who had died non-violent deaths; they were not a cross-section of the people actually dying on the roads. Child cadavers were, for obvious legal and moral reasons, effectively impossible to obtain, which left an enormous blind spot exactly where it hurt most. Every specimen was different from every other, and any given part could be used only once, which made repeatable, comparable data maddeningly hard to produce. And as crash testing became routine, suitable donors grew scarce. The dead could teach the fundamentals of tolerance, but they could not be a factory for standardized answers.

The Colonel and the professor

If the dead could establish where the human body breaks, only the living could reveal how it behaves right up to that edge — and a small number of extraordinary people volunteered to be the instruments. The most famous was Colonel John Paul Stapp, a U.S. Air Force flight surgeon who, in 1954, strapped himself to a rocket sled, accelerated past 620 miles per hour — faster than a .45 bullet — and was brought to a halt in 1.4 seconds. He did it to prove that the human body could survive far greater deceleration than the aviation establishment believed, and in doing so he redrew the map of human tolerance that all crash safety would be built on. He briefly held the title of fastest man on earth. He also detached both retinas.

Less famous, and in some ways more remarkable, was Lawrence Patrick, a Wayne State professor who decided that the most ethical test subject for the survivable range was himself. Over his career Patrick endured some four hundred rides on a rocket sled. He let himself be struck in the chest with heavy metal pendulums. He was hit in the face by pneumatically driven hammers and sprayed with shattering glass to simulate a window imploding. He described the cumulative result, with magnificent understatement, as leaving him "a little sore." The data his battered body produced became the mathematical baseline against which later research was measured.

The dead defined the limit. The living mapped the approach to it. Neither could do the other's job — and neither could do the job that was coming next.

The menagerie

By the mid-1950s, researchers needed something cadavers could not provide: living tissue that would respond to injury the way a crash victim's would, in tests too dangerous for any human volunteer. The answer, for roughly three decades, was animals, and this chapter is the hardest to read. Chimpanzees were sent down rocket sleds. A bear was strapped to an impact swing. The writer Mary Roach documented anesthetized pigs — chosen because their internal organ layout resembles a human's and because they can be seated upright — crashed into deep-dish steering wheels at around 10 miles per hour. In 1978, the University of Michigan used baboons.

Grim as it was, the work answered questions nothing else could, and one of those answers saved an enormous number of lives. By 1964, impalement on the rigid steering column had been implicated in over a million deaths — a staggering share of all traffic fatalities. Understanding how to stop the column from spearing the driver required living subjects; the result, General Motors' collapsible steering column, cut the risk of steering-wheel death roughly in half. Animal testing drew fierce and sustained protest — far more than cadaver work ever had, because the dead felt nothing and the animals plainly did — and it wound down as the data was exhausted and better tools arrived. General Motors ended live-animal crash testing in 1993; the rest of the industry followed. The tool that replaced the menagerie had, by then, already been evolving for forty years.

Sam, and the family he started

The crash test dummy was born not in a car but in an ejection seat. In 1949, an inventor named Samuel Alderson built "Sierra Sam," a 95th-percentile male figure — taller and heavier than 95 percent of men — to test aircraft ejection systems and pilot harnesses at forces no human could survive, on rocket sleds running past 600 miles per hour. Alderson followed Sam with a series called the VIP-50, built specifically for General Motors and Ford, and a rival firm answered with a dummy it called "Sierra Stan." The idea — a durable, repeatable, instrumented human simulacrum — was exactly what the car industry needed, and through the 1950s and '60s the dummy migrated from the runway to the road.

General Motors, finding neither early dummy good enough, combined the best features of both and, in 1971, produced Hybrid I: a 50th-percentile male, an "average man" in height, weight and proportion. Crucially, GM shared the design with its competitors through the Society of Automotive Engineers, making it an industry standard rather than a proprietary edge. Hybrid II followed in 1972, with better shoulders, spine and knees — and it made history as the first dummy repeatable and reproducible enough to be written into a federal safety standard, the benchmark for testing lap-and-shoulder belts. Then, in 1976, came the one you have seen a thousand times: Hybrid III.

Hybrid III is the familiar figure, and he has become the patriarch of an entire household. In his standard form he stands five foot nine and weighs 170 pounds, and he occupies the driver's seat in the offset crash tests you watch on the news. Around him GM and its successors built a family meant to stand in for the range of real bodies on the road: a 95th-percentile "big brother" at six foot two and 220 pounds; a 5th-percentile female counterpart; and three child dummies sized to a ten-, six-, and three-year-old. Every part of a given dummy type is interchangeable with every other, so a single dummy can be crashed again and again, its damaged pieces simply swapped out — and a well-kept dummy can stay in service for up to thirty years. They are not mass-produced: only a few hundred are made in a typical year, most of them by a single Detroit-area company, Humanetics, which has quietly come to dominate the world's supply. A basic, uninstrumented Hybrid III body already costs well over $100,000; fitted with its full complement of sensors, the price climbs to roughly $170,000 to $200,000. That is the entry-level model.

Inside a tenth of a second

Here is where the reassurance lives, if you want to feel safe in your car: consider how much is measured, and how quickly, every time one of these dummies is put to work. The crash itself is almost unimaginably brief — a frontal impact is essentially over in 100 to 150 milliseconds, a tenth to a seventh of a second, faster than a blink. But in that instant the dummy is a full-body instrument. A Hybrid III carries as many as fifty-eight separate data channels running from its skull to its ankles — accelerometers in the head and chest, load cells measuring force in the neck, spine, femurs and pelvis, string potentiometers reading exactly how far the ribs compress — and together they record something on the order of thirty to thirty-five thousand individual measurements in that single, sub-second event.

The precision is enforced by ritual. Before every test the dummy is calibrated: its head is detached and dropped from sixteen inches to confirm the head instruments read true; the neck is set swinging and stopped short to check its flexure; the chamois-leather skin is inspected; the knees are struck with a metal probe; and the chest is slammed with a heavy pendulum to make certain the ribs bend and flex the way a person's would. Calibration dots are fixed to the head so that high-speed film — shot at a thousand frames a second or more — can be matched frame by frame to the sensor data. The measurements are written to a recorder buried in the dummy's chest and downloaded to a computer the instant the test ends. The crash lasts a heartbeat. Preparing for it takes hours or days; making sense of it takes longer still.

From numbers to a promise

Thirty-five thousand numbers are not, by themselves, safety. The genius of the system is how those numbers are turned into a judgment about a human body. Engineers translate the raw sensor data into injury criteria — validated formulas, calibrated against all that cadaver and volunteer research, that convert force and motion into the probability of a specific injury. The Head Injury Criterion distills the head's acceleration into a concussion-and-fracture risk — and it is no modern invention but a direct descendant of the Wayne State Tolerance Curve, the injury threshold those very ball-bearing-and-cadaver experiments produced back in the 1930s. The grim work in the elevator shaft, in other words, is quite literally the math still deciding how hard your airbag fires. Chest deflection — how far the ribcage is crushed — turns out to be the single best predictor of rib fracture, better than acceleration alone. Femur load predicts leg fractures; neck loads predict the injuries that paralyze. Each figure becomes a percentage: the odds that a real person in that seat, in that crash, would suffer a serious injury.

Those percentages are the whole point. They are what a car has to satisfy to pass a federal safety standard, and what earns it stars in the crash-test ratings you check before you buy. Every mainstream car you can walk into a showroom and drive away has been through this — instrumented proxies subjected to frontal, side, rear, rollover and small-overlap impacts, tens of thousands of measurements per run, each one weighed against a tolerance established, ultimately, by a body that took the blow first. The chain runs in a straight line: learn how an injury happens, measure how much force causes it, build a tool that can feel that force, validate the tool against real human tissue, and turn the result into a rule every carmaker must meet. When you buckle your seatbelt, you are trusting the output of that chain.

THOR, the biofidelic supercomputer

The dummy's own evolution has not stopped, and its successor is a machine of almost absurd sophistication. THOR — the Test device for Human Occupant Restraint, developed by the U.S. safety regulator over many years and millions of dollars with Humanetics — is built to behave far more like a living body than Hybrid III ever could. It has a genuinely human-like, flexible spine and pelvis, a ribcage that deforms the way ours does, and a face studded with sensors precise enough to read a facial impact. Its chest is measured not at one point but at several, by instruments called IR-TRACCs that track deflection in three dimensions; its neck moves in all six degrees of freedom; and it carries more than 150 data channels, roughly three times a Hybrid III's. Europe's testing program has already begun putting THOR in the driver's seat in place of the older dummy.

That capability is expensive in every sense. A single THOR is assembled from more than thirty thousand parts — steel "bones," rubber, foam, and precision electronics — and takes about five months to build. Its skull, ribs and pelvis are close replicas of the real thing. And it costs, depending on configuration and instrumentation, anywhere from $400,000 to well over $1 million — more than three new supercars, for one faceless figure that exists only to be destroyed and rebuilt.

And THOR is only the frontal-impact specialist. A whole family of dummies exists because a crash can come from any direction, and each type answers a different question no other can.

Different bodies for different collisions — because a side impact, a rear-end hit and a rollover injure people in completely different ways.

The working family of crash test dummies

Dummy	Debut	Stands in for	What it measures
Sierra Sam	1949	95th-percentile male	Ejection seats, aviation — the first of them all
Hybrid II	1972	50th-percentile male	First dummy written into a federal standard; belts
Hybrid III family	1976	50th & 95th male, 5th female, children 3/6/10	Frontal impact — the global workhorse
THOR-50M / 5F	2013	mid-size male, small female	Advanced frontal; 150+ channels, human-like torso
WorldSID	2000s	50th-percentile male	Side impact — ribs, spine, organ compression
BioRID	2000s	average adult	Rear impact & whiplash; 24 vertebra simulators
CRABI	1990s	infants 6/12/18 months	Child seats and airbag risk to the very young

For all the attention frontal impacts command, they are not the only way a body is broken, and several of these dummies exist because of injuries the head-on test cannot see. Side impacts are especially lethal, because only inches of door stand between an occupant and the striking vehicle — which is why WorldSID is instrumented down its ribs, spine, and internal organs. And the single most common crash injury of all is whiplash, the vicious snap of the neck in a rear-end hit, which is why BioRID's spine is built from twenty-four separate segments standing in for the vertebrae, chasing a wound that rarely appears on a fatality chart yet costs billions and ends careers. There are even dummies that never ride inside the car at all: instrumented headforms and legforms, fired at a vehicle's hood and bumper, standing in for the pedestrian it strikes.

Beyond the automobile

The dummy's descendants no longer work only in cars. The same science now stands in for soldiers: the U.S. Army's WIAMan — the Warrior Injury Assessment Manikin — was built specifically to measure the upward blast of a roadside bomb detonating beneath a vehicle, a punishment no automotive dummy was ever designed to feel. Its cousins turn up in football's concussion research, in aviation, and in spaceflight. The faceless proxy that began as a stand-in for a crashing driver has quietly become the backbone of injury science across nearly every field where a human body meets sudden, violent force.

Increasingly, the newest "dummy" is not a physical object at all. Computational human body models — digital cadavers such as Toyota's THUMS — can be crashed ten thousand times before breakfast, for a fraction of the cost of one physical test, and Volvo used exactly this approach in 2002 to create "Linda," a virtual pregnant occupant modeled in her 36th week to study forces on the womb and fetus that no physical dummy could ethically address. Cheaper, faster, endlessly repeatable, unburdened by the weight of the real thing. Which raises the obvious question — and points straight at the one blind spot the whole apparatus carried for half a century.

The 110-pound problem

For all its sophistication, the standard dummy carried a bias built into its bones. Hybrid III was, definitionally, an average male. When a "female" dummy finally entered U.S. testing in 2003, it was not a new design at all; it was the male dummy scaled down to represent roughly the smallest five percent of women, based on body data from the mid-1970s, and — as safety engineers now bluntly admit — it lacked sensors in the very places a woman's body is injured differently from a man's. It is routinely tested in the passenger seat or the back, and only rarely in the driver's seat, even though most licensed drivers are women. For decades, in other words, the "default human" tested by the entire car industry was a man, and everyone else was an approximation of him.

The consequences show up in the crash data, and they are not subtle. Women make up 62 percent of new-car buyers in the United States, yet compared with men in comparable crashes they are markedly worse off.

Regulators had been asking for a genuinely female dummy since 1980. The answer took more than forty years. In May 2023, at Sweden's national road-and-transport research institute, a team led by Dr. Astrid Linder unveiled the world's first crash test dummy actually modeled on a female body rather than a shrunken male one — built to represent women at the 50th and 25th percentiles, and developed specifically because the data showed women suffer whiplash at far higher rates. Humanetics, meanwhile, has spent more than a decade developing a fully sensored female THOR that captures injury where women actually sustain it; it costs around $1 million, roughly twice the dummy in use now, and cost is precisely the objection some American automakers raise. In late 2025, U.S. regulators released design details for their own advanced small-female dummy, shaped and instrumented around real female anatomy. As of this writing it is not yet in routine use — and that it still is not required tells its own story: rewriting a federal crash-test standard is glacially slow, a fully sensored female dummy costs about twice its predecessor, and it has taken acts of Congress to force the question. Nor is sex the only gap. The elderly, whose bones fracture at lower loads, and the obese, whose mass moves differently in an impact, were long represented by that same average male; specialized dummies for both now exist but are far from standard, as does a physical pregnant dummy — because a belt routed wrong across a pregnant abdomen is its own distinct danger that, for decades, no test could see. The default human is finally being redrawn — half a century late, and still against resistance. And the absence runs deeper than the dummy: the women who were missing from the test lab were, for most of a century, missing from the rooms where the car itself was designed — even as some of them were inventing its safety anyway.

The impossible subject

There was one subject the cadaver era could never study, and it was the one that mattered most: children. For reasons that need no explaining, child cadavers were effectively unavailable — leaving the deadliest blind spot precisely where a parent's fear lives. The gap was closed the hard way. Some of it came from the animal research of the earlier era; the rest came from child-sized dummies — the Hybrid III youngsters and the infant CRABI, small enough to represent a six-month-old — and, increasingly, from computational models of a child's body. The price of having been blind was paid in the 1990s, when the airbags meant to save lives began killing small children, flung into a cushion that deploys as fast as 200 miles per hour, because the systems had been designed and certified around grown men and no one had adequately reckoned with a body a third that size. The fix — dual-stage "smart" airbags that sense an occupant's size and weight before they fire — is a monument to a lesson learned in the cruelest way there is.

The willing dead, still

For all the sophistication of THOR and the digital humans, one uncomfortable fact remains, mostly against the industry's wish: real bodies are still needed. A 2024 review in the journal Traffic Injury Prevention, written by an international group of impact-biomechanics researchers, laid out both the ethics and the necessity of what the field delicately calls Post-Mortem Human Subjects. Its central finding is the uncomfortable one: cadaver testing keeps identifying injuries that even the best dummies and computer models fail to predict. The airbag is the paper's key example. Airbags are estimated to have saved 28,244 lives in the United States through 2009 — and to have been implicated in some 320 deaths, many of them children and small adults killed by the very device meant to save them. It was cadaver testing that revealed those lethal scenarios, which preliminary work with dummies and simulations had missed entirely. Real tissue tears in ways a model does not yet know to imitate.

Which is why the practice continues, mostly in universities funded by regulators and, quietly, by automakers who prefer not to be named — because, as one former Ford safety chief put it, human modeling simply "hasn't reached that state yet." And it is conducted, now, under a real ethical architecture that the early elevator-shaft era never had. The framework traces from the Declaration of Helsinki through modern international guidelines: informed consent from the donor or next of kin, mandatory review by an independent ethics board, and a duty to publish so the knowledge gives back. In the labs, the care is visible and deliberate — bodies are swaddled, head and all, in stockings before a test, both for scientific reasons and, as the researchers put it, out of respect. The dead are not disposable here. They are, precisely, donors — people who gave their bodies so that strangers might survive a Tuesday-morning commute.

Where do we go from here?

The trajectory is clear: toward a digital human faithful enough to retire the physical one — a "human body model" that finally reaches full biofidelity, capturing not just where a body breaks but how each tissue tears, for every age, every sex, every size. When that arrives, the physical dummy will become a calibration tool and the cadaver a last resort rather than a necessity. In the nearer term, the belated female, child, elderly and obese dummies are closing the demographic blind spots that skewed a century of data, and the sensor counts keep climbing — from Hybrid III's fifty-eight channels to THOR's hundred and fifty and beyond, each new channel another place we can now see an injury coming. And the industry's center of gravity is shifting from surviving the crash to preventing it altogether — the logic that leads, ultimately, to the autonomous car.

But there is a humility at the bottom of this story that no amount of computing power erases. For nearly a hundred years, the only way to learn how to protect a living human body has been to study one that was no longer living, or willing to be hurt, or unable to refuse. The seatbelt across your chest, the airbag in your wheel, the glass that crazes instead of spearing you — each was underwritten by a body that took the blow first, and then measured, tens of thousands of times over, by a faceless proxy built to be a person. The least the rest of us can do is know it — and, perhaps, feel a little safer for it.