McNamara Fallacy

There is a dangerous slippery slope with four steps: it starts at "measure what is easy to measure" and slides to "what cannot be measured does not exist." Its name is the McNamara fallacy, after Robert McNamara, US Secretary of Defense during the Vietnam War.

Vietnam was a counterinsurgency war. No front line, no way to measure progress in kilometres advanced. McNamara brought a whole systems analysis apparatus over from Ford, and he needed a progress measure that could go into a report. What he picked was the body count: enemy dead, the kill ratio between the two sides. Measurable, reportable, a number every day, and it could be drawn as a rising curve.

Then the law unfolds as standard. Officers at every level of the front found that promotion and assessment turned on that number, so the number got padded as it moved up the chain. The assessment pressure also induced indiscriminate killing, and Campbell (the social scientist behind the corruption law in B02) in 1979 put body count and the My Lai massacre side by side as examples. Meanwhile the things that actually decided the course of the war, the political mood of the villages and where the people's loyalties lay, went by the name x-factor in systems analysis jargon: because they could not be quantified, they were simply discarded. The two forces stacked on top of each other: the number got padded, so the report looked good; the x-factor got dropped, so the report could not show the real risk. The reports said "we are winning," until the unmeasurable part of reality came to collect.

One thing matters here: McNamara was not a cartoon villain, and that is precisely what makes the fallacy frightening. He came from Ford's "Whiz Kids" statistical management team, and bringing systems analysis into the Pentagon was, in management history, an innovation. So the fallacy is not a matter of motive or competence: the tool was not the problem. The problem was mistaking the boundary of the tool for the boundary of reality. The four-step slope describes how that single leap gets completed in four moves, each of which looks reasonable at the time.

The four steps are worth going through word by word. Notice how gentle the gradient is at each one.

Step one: measure whatever can be easily measured. Nothing wrong so far, and the original says this is OK as far as it goes. Step two: disregard what cannot be easily measured, or give it an arbitrary quantitative value. The original verdict is artificial and misleading. Step three: presume that what cannot be measured easily really is not important. The verdict is blindness. Step four: say that what cannot be easily measured really does not exist. The verdict is suicide.

The stairs are built very shallow. Between step one and step two lies only one act of "set that aside for now"; between step three and step four lies only one budget meeting. Any two adjacent levels look like reasonable simplification, and walking the whole flight is cognitive self-termination. The dynamics are not mysterious either. Measurable dimensions naturally grow into reports, weekly updates and leaderboards, and every time an unmeasurable dimension gets omitted its standing drops a notch. The four steps just put marks on a gradual process. The word the author picked for the last step is suicide rather than error, and the choice is exact: his point is not that someone miscalculated, it is that a decision system has cut out one of its own organs of perception.

One more detail that gets overlooked: the passage was not spoken in a military setting. It was born at a marketing conference, in front of an audience of sales executives. The author, Yankelovich, worked in public opinion and consumer research, and the people he was warning were colleagues who compressed consumers' lifestyles into a few easily measured numbers. Moving the fallacy from the battlefield into the conference room needs no translation. The conference room is where it was named.

Vietnam shows up twice in the blue branch, and the two should be kept apart. Here it is the measurement end: body count defined what counted as progress. B06 (rewarding A while hoping for B) is the incentive end: Kerr points out that the rotation and assessment system actually rewarded individual risk avoidance rather than winning the war. One war, with a ruler measuring the wrong thing at one end and a reward hung on the wrong behaviour at the other, and the two threads amplify each other.

Within the family, the McNamara fallacy occupies a distinctive position. Goodhart and Campbell describe how a targeted measure goes false; this describes how an unmeasured dimension goes missing. One is the ruler being bent, the other is the world outside the ruler being written off. The two are welded together by the same mechanism: the multitask theorem of B09 (Holmström and Milgrom proved that as long as a person's effort trades off between tasks, paying heavily for the measurable task is equivalent to subsidizing withdrawal from the unmeasurable one) shows that step four needs nobody to announce it, and the incentive structure executes it by itself. As for step two's "give it an arbitrary quantitative value," the red branch has a dedicated study of it: R03 on commensuration (forcing incomparable things into one ruler) is exactly about that compulsory pricing, and once an arbitrary weight is written into a formula, it becomes a fait accompli nobody questions again.

The fallacy later acquired its own clinical department: O'Mahony wrote "Medicine and the McNamara Fallacy" in 2017 for the journal of the Royal College of Physicians of Edinburgh, wiring the four-step slope into the quantification culture of medicine. The modern vitality of the fallacy is not in war history. It is in every trade that "lets the data talk."

The boundary should be stated too: the four-step fallacy is not against measurement, and the original explicitly waves step one through. What it opposes is the leap that treats the boundary of measurement as the boundary of existence. Measurability is a property of instruments, importance is a value judgment, and the two are independent. The cost of confusing them runs from Saigon to the ward to the leaderboard, and the bill is still being written. That is the whole proposition of this section.

One open question: in the eval report in front of you, which dimensions are sitting at step two, assigned an arbitrary value that nobody has questioned? Listing them changes the conclusion more than squeezing one more point out of an existing score. And a harder follow-up: once they are listed, who speaks for them at the next meeting?

The one-line takeaway: measurable is not the same as important, and unmeasurable is certainly not the same as nonexistent; the report decides what an organization can think about before it decides what the organization does.