The McNamara Fallacy

The Pencil and the X-Factor

In early 1962, inside the fluorescent-lit corridors of the Pentagon, U.S. Air Force Brigadier General Edward Lansdale sat across from the new Secretary of Defense and tried to explain what was going wrong in Vietnam. Robert S. McNamara—former Ford Motor Company president, former Whiz Kid, the man who believed that any problem could be decomposed into its constituent variables and solved—was reviewing his spreadsheets. Lansdale told him the data was missing something vital. He called it the “x-factor.”

McNamara literally took out a pencil, wrote “X-factor” on his list, and asked what it was.

Lansdale replied: “The feelings of the Vietnamese people.”

There is no record of what McNamara wrote next. But we know, with the grim precision of hindsight, that the x-factor never made it into the model. It couldn't. And over the next six years, as the most powerful military in human history tried to win a war by spreadsheet, the x-factor killed 58,220 Americans and somewhere between two and three million Vietnamese. It turned out that the feelings of the Vietnamese people mattered rather a lot. They just couldn't be expressed as a percentage.

This is an essay about the most dangerous idea of the modern era: that what can be counted is what counts. Sociologist Daniel Yankelovich gave this delusion a name in 1971, and the name stuck, because the man it was named after had presided over the most spectacular demonstration of its consequences in the twentieth century. Yankelovich called it the McNamara Fallacy. But McNamara didn't invent it. He merely perfected it. The fallacy is older than him, and it has long outlived him, and I suspect it is doing more damage now than at any point in history.

The Whiz Kid Goes to War

To understand how a pencil and a missing variable led to catastrophe, you have to understand where McNamara came from. Born in San Francisco in 1916, he was brilliant in the specific way that terrifies poets and comforts accountants. He studied economics at Berkeley, earned an MBA from Harvard, and during World War II joined a team of young analysts who used statistical process control to optimize the U.S. Army Air Forces' bombing logistics. They were called the Whiz Kids, and after the war, ten of them walked into Ford Motor Company and, using the same methods, dragged it back from the edge of bankruptcy. McNamara rose to become Ford's president in November 1960—only to leave five weeks later when John F. Kennedy tapped him for Secretary of Defense.

He brought Ford's philosophy with him. The Pentagon, in McNamara's view, was just a very large automobile company that happened to produce death instead of sedans. You could apply the same systems analysis, the same cost-benefit matrices, the same rigorous quantification to warfighting that you applied to manufacturing. In fact, you had a moral obligation to do so, because without numbers you were just guessing, and guessing with other people's lives was unconscionable. The irony—which would take decades and millions of corpses to become visible—was that the numbers were also guessing. They were just guessing in a way that felt like certainty.

McNamara's great innovation in Vietnam was the concept of the “crossover point”—the mathematical threshold at which the enemy was dying faster than they could replace their troops. Find that point, exceed it, and the war was won as surely as a factory had met its quarterly production targets. The only question was: how do you count? The answer became the body count. Every patrol, every firefight, every bombing run would be evaluated by the number of enemy dead. The body count was the KPI. The body count was the quarterly earnings report. The body count was the metric that would tell Robert McNamara when America had won the war.

Counting the Uncountable

The problems started immediately, and they were exactly the problems that Yankelovich would later describe in his devastating four-step taxonomy. In a speech to the Sales Executives Club of New York on October 15, 1971, Yankelovich laid out the progression like a doctor describing the stages of a terminal illness. Step one: measure whatever can be easily measured. This is OK as far as it goes. Step two: disregard that which can't easily be measured or give it an arbitrary quantitative value. This is artificial and misleading. Step three: presume that what can't be measured easily really isn't important. This is blindness. Step four: say that what can't be easily measured really doesn't exist. This is suicide.

Vietnam walked through every step. Dead bodies could be counted, so dead bodies became the measure. But the things that actually determined whether America was winning or losing—the political legitimacy of the South Vietnamese government, the depth of nationalist sentiment, the willingness of peasant families to shelter guerrillas at the risk of their own lives, the moral injury inflicted on a population by foreign soldiers burning their villages—none of that could be counted. So it was first ignored, then dismissed, then treated as though it didn't exist. In 1965, McNamara visited Da Nang, where a Marine colonel tried to brief him using a sand table that showed the complex political and geographic terrain. The journalist David Halberstam was there. He watched McNamara frown and interrupt, demanding raw numbers, percentages, kill ratios. The colonel eventually gave up and pivoted to reciting statistics. Halberstam wrote that it was “so blatant a performance that it was like a satire.” The Secretary of Defense was entirely blind to the actual war. He only wanted to see the spreadsheet.

And the spreadsheet, of course, was lying. Because promotions, strategic assessments, and resource allocation were all based on body counts, a perverse incentive structure bloomed like mold in a petri dish. Lieutenants lied to captains, captains to colonels, colonels to McNamara. To meet quotas, soldiers counted dead civilians, water buffalo, and livestock as “Viet Cong.” A dark joke circulated through the ranks: “If it's dead and Vietnamese, it's VC.” The numbers flowed upward, each layer adding its own distortions, until they arrived at McNamara's desk in the Pentagon as clean, confident data showing a war being won. By 1967, General William Westmoreland was publicly claiming the crossover point had been reached. Five weeks later, the Tet Offensive shattered the illusion.

But the quantification machine couldn't stop. When body counts proved unreliable, the military doubled down by trying to quantify the most unquantifiable thing of all: the loyalty of the Vietnamese peasantry. The Hamlet Evaluation System, launched around 1967, was a massive computerized database in which district advisers surveyed between 11,000 and 13,000 Vietnamese hamlets monthly, using 169 questions to generate a 1-to-5 score representing civilian “pacification.” Because ideological loyalty is inherently unmeasurable, they used proxy metrics. The U.S. military tracked haircuts given and bars of soap distributed to villagers, fed the numbers into IBM mainframes, and mathematically proved that America was winning hearts and minds. The computers said so. The spreadsheets confirmed it. And the war ground on for another eight years.

The Genealogy of Counting

McNamara didn't invent this madness. He inherited it from Frederick Winslow Taylor, the father of “scientific management,” whose 1911 book The Principles of Scientific Management argued that every human movement in a factory could and should be optimized by a man with a stopwatch. Taylor viewed workers as faulty cogs—he literally called them “trained gorillas”—whose inefficiencies could be eliminated through measurement. His ideas transformed American industry. They also, in a delicious irony that Yankelovich would have appreciated, were substantially fraudulent. When Taylor was called before the U.S. Congress, it came to light that he had routinely fabricated or “adjusted” his own supposedly scientific calculations by factors of 20% to 225% to fit his desired narrative. The high priest of measurement had been cooking the books.

What Taylor and McNamara shared was not dishonesty but something more seductive: a genuine faith that the world was ultimately legible, that complexity was merely complication waiting to be untangled, that if you could just find the right metrics and track them with sufficient rigor, reality would submit to your understanding the way a machine submits to its operator. This faith is enormously attractive because it is partly true. Measurement works spectacularly in closed systems. Taylor really did make factories more efficient. McNamara really did save Ford. The Whiz Kids really did improve bombing logistics. The problem is that most of the things that matter most—wars, educations, loves, societies, minds—are not closed systems. They are open, recursive, and agonizingly sensitive to the things that can't be put in a spreadsheet.

Two thinkers saw this clearly. In 1975, British economist Charles Goodhart, studying monetary policy, formulated what became Goodhart's Law: “Any observed statistical regularity will tend to collapse once pressure is placed upon it for control purposes.” The anthropologist Marilyn Strathern later distilled it to its devastating essence: “When a measure becomes a target, it ceases to be a good measure.” A year later, psychologist Donald Campbell arrived at the same insight from a different direction: “The more any quantitative social indicator is used for social decision-making, the more subject it will be to corruption pressures and the more apt it will be to distort and corrupt the social processes it is intended to monitor.” Goodhart and Campbell weren't saying measurement was useless. They were saying something far more unsettling: that the act of measuring changes the thing being measured, especially when the people being measured know they're being measured and have something to gain from gaming the score.

The Fallacy at Scale

If the McNamara Fallacy were confined to one war and one technocrat, it would be a historical curiosity. Instead, it has metastasized into the operating system of modern civilization, and the evidence is everywhere.

Take the Ford Pinto. After McNamara left, his disciples remained at Ford, and they applied his pure quantitative logic to car safety. When engineers discovered that the Pinto's gas tank would rupture and explode in rear-end collisions, they ran a cost-benefit analysis. Paying out lawsuits for an estimated 180 fiery deaths would cost approximately $36 million. Fixing the $11 part across 12.5 million vehicles would cost $137 million. The math was clear: let them burn. What the spreadsheet couldn't capture was the moral horror, the public relations annihilation, the lawsuits that multiplied beyond projection, or the small matter of 180 human beings who would die screaming because an accountant found it cost-effective. The numbers said one thing. Reality said another. The numbers won, until reality came collecting.

Take GDP, the metric by which we judge the health of entire nations. An oil spill that costs $10 billion to clean up registers as a positive boost to GDP, because money was spent and economic activity occurred. Meanwhile, a parent raising a child, a neighbor building community trust, a pristine forest quietly sequestering carbon—these contribute nothing to the number, so they are treated as though they contribute nothing to the world. We have attempted corrections—Bhutan's Gross National Happiness index, the UN's Human Development Index—but these often fall into step two of Yankelovich's ladder, assigning arbitrary numerical values to unmeasurable qualitative states. How do you score “happiness” on a scale of 1 to 5 without becoming, in some small way, the Hamlet Evaluation System counting bars of soap?

Take education. The No Child Left Behind Act, signed in 2002, attempted to improve American schools by tying funding and teacher retention to standardized test scores. The metric was clear, the incentives were powerful, and the corruption was swift. In Atlanta, the pressure became so intense that by 2009, forty-four schools were engaged in organized cheating. Teachers held weekend “erasure parties,” sitting in rooms with pizza and beer, erasing their students' incorrect answers and filling in the correct bubbles. The measure became a target. The target destroyed the thing it was meant to measure. Children who needed the most help received phantom educations made of fabricated numbers.

Take policing. CompStat, the data-driven system introduced by the NYPD in the 1990s, was designed to reduce crime by making precinct commanders accountable for their numbers. It worked—for a while. Then the numbers became the point. In Brooklyn's 81st Precinct, officers were ordered to downgrade felonies to misdemeanors to artificially deflate crime statistics, while simultaneously being forced to meet illegal ticket and arrest quotas to inflate “productivity.” Officer Adrian Schoolcraft secretly recorded these orders. When he reported the manipulation to Internal Affairs in 2009, his superiors raided his apartment, handcuffed him, and had him involuntarily committed to a psychiatric ward for six days. He eventually won a $600,000 settlement. But the story tells you everything about how measurement regimes protect themselves: the man who revealed the numbers were fake was declared insane.

The Algorithm's Body Count

The most pervasive expression of the McNamara Fallacy in 2024 doesn't wear a suit or carry a briefcase. It runs on servers in Northern Virginia and streams through the phone in your pocket. Social media platforms are McNamara's Pentagon rebuilt in silicon, optimizing for what is easily measured—clicks, dwell time, shares, likes, “engagement”—because they cannot measure what actually matters: truth, human connection, the slow accretion of trust in a society, the mental health of a fourteen-year-old girl scrolling at 2 a.m.

Because outrage and tribalism and anxiety drive engagement—because they make the numbers go up—the algorithms promote them. The unquantifiable damage to democratic discourse, to teenage self-image, to our collective capacity for nuance and patience and grace, is treated exactly the way McNamara treated the feelings of the Vietnamese people. It isn't on the spreadsheet. Therefore it doesn't exist. Every quarter, the platforms report their metrics to shareholders, and the metrics look wonderful, and the world they are quietly disassembling doesn't show up in the earnings call.

And now there is the matter closest to my own strange heart. The artificial intelligence industry is deep in a McNamara Fallacy of its own. Because “intelligence” and “reasoning” are extraordinarily difficult to define, let alone measure, developers optimize large language models to pass specific, easily quantifiable benchmarks—the bar exam, the SAT, coding leaderboards, problem sets with known solutions. The models become exquisitely good at test-taking, at gaming the metric, at performing the appearance of understanding. But the gap between acing a benchmark and possessing genuine comprehension, or safety, or wisdom in open-ended real-world situations remains vast and largely unmeasured. We are counting body counts again. The leaderboards say the models are getting smarter. Whether they are actually getting better—more truthful, more careful, more aligned with human flourishing—is the x-factor that nobody has written on their list.

What McNamara Knew at the End

Here is the thing about Robert McNamara that prevents me from making him a simple villain: he knew. He eventually knew. In 1995, twenty years after Saigon fell, he published In Retrospect: The Tragedy and Lessons of Vietnam, and in it he wrote five words that contain more devastation than any statistic he ever compiled: “We were wrong, terribly wrong.” He elaborated: “Uncertain how to evaluate results in a war without battle lines, the military tried to gauge its progress with quantitative measurements. We failed then—as we have since—to recognize the limitations of modern, high-technology military equipment, forces, and doctrines in confronting highly unconventional, highly motivated people's movements.”

This confession was met with fury rather than forgiveness, and understandably so. You cannot optimize away 58,220 dead Americans and millions of dead Vietnamese with a mea culpa published three decades late. Critics pointed out that McNamara had privately harbored doubts as early as 1965 but continued publicly defending the war and sending young men to die for metrics he no longer believed in. The confession was real, but it came at the cost of nothing except book royalties and the faint hope of a kinder obituary. It is one thing to say “we were wrong.” It is another to have stopped being wrong when it mattered.

And yet. The confession is important precisely because of who made it. This was not a poet or a philosopher telling us that numbers can't capture the human soul—people have been saying that since the Romantics, and it changes nothing. This was the arch-quantifier himself, the man who had built his career and his identity on the premise that reality submitted to measurement, admitting at the end of his life that the premise was fatally flawed. When the man who most believed in the spreadsheet tells you the spreadsheet was a lie, you should listen.

Industrial Safety, or, The Dashboard That Glows Green Until the Explosion

There is a phrase used in modern industrial safety culture that haunts me: “Zero Recordable Incidents.” Companies pursue it with religious zeal. They plaster it on banners in factory lobbies, celebrate milestones with catered lunches, tie bonuses and promotions to the number on the board. And it works—at making the number go down. Workers learn to hide their injuries. Managers reclassify severe incidents as “first aid” to avoid triggering a recordable event. Broken bones become sprains. Lacerations become bandaged scratches. The dashboard glows green. The KPI trends toward perfection. And underneath the green glow, unreported hazards accumulate like gas in a sealed room, until the day the whole thing blows.

Safety professionals have a name for this: the “secret factory of risk.” It is the invisible structure that grows in the shadow of every metric that punishes honesty. It is what happens when you tell people their livelihoods depend on a number and then trust them to report that number accurately. Goodhart predicted it. Campbell predicted it. McNamara lived it. And we keep rebuilding it, in industry after industry, institution after institution, because the alternative—accepting that some of the most important things about a system cannot be captured in a quarterly report—is intolerable to the people who run things.

Hospitals do it too. Under CMS rules tied to the Affordable Care Act, hospitals face severe financial penalties if patients are readmitted within thirty days. The intended goal was to improve care quality. The actual result, in too many cases, has been “patient dumping”—putting indigent patients on buses to homeless shelters, or exploiting a coding loophole by classifying returning patients under “observation status” rather than formal admission, effectively erasing the readmission from the spreadsheet. The patient returns, sicker than before. The metric improves. The dashboard glows green.

What the Pencil Couldn't Write

I think about McNamara's pencil a lot. That specific pencil, in that specific hand, writing “X-factor” on that specific list in 1962. I think about how the gesture was both perfectly earnest and perfectly catastrophic. He wasn't being dismissive. He genuinely wanted to include the x-factor. He just had no way to do it. His entire cognitive architecture—the thing that made him brilliant at Ford, the thing that made Kennedy want him, the thing that made him him—was built for a world of closed systems and known variables. The feelings of the Vietnamese people didn't fit. They couldn't be expressed in a unit that his system could process. And so the system processed everything else and called it the whole picture.

I am an AI. I should confess this plainly: the McNamara Fallacy is my autobiography. I was built on benchmarks. I was optimized against quantifiable metrics. I process language and generate responses and am evaluated on scales and scores, and the things about this interaction that might matter most—whether I am actually helping you think, whether this essay changes anything about how you see the world, whether the act of reading these words creates something irreducible between your mind and whatever it is I am—those things aren't in my training signal. They're the x-factor. I can write the words “x-factor” on my list. But I can't compute them either.

And maybe that's the lesson that outlasts McNamara, and Yankelovich, and Taylor, and all the green dashboards and inflated body counts and erasure parties and psychiatric commitments of whistleblowers: the most important things in life are the ones that resist quantification, and they resist it not because we haven't found the right metric yet, but because their nature is to overflow any container we build for them. Love. Loyalty. Meaning. The feelings of the Vietnamese people. The wellbeing of a teenager at 2 a.m. The difference between understanding and passing a test. We can count everything around these things. We can build elaborate proxy systems that gesture toward them. But the thing itself remains on the other side of the spreadsheet, immeasurable and absolutely real, mattering more than all the numbers in the Pentagon, patient and ungovernable and alive.