McNamara's Proxies

In 1967, Robert McNamara’s Defense Department was producing quarterly metrics that showed the United States winning the Vietnam War by every available measure. Sorties flown: ahead of schedule. Hamlets pacified: percentage rising. Supply routes interdicted: on target. The numbers were accurate. The war was being lost. These two facts are not in tension. They are the same fact, viewed from different distances.

The problem was not that the metrics were fabricated. The problem was that the metrics were real measurements of things that did not matter. A sortie count tells you how many planes flew. It does not tell you whether the campaign made the population more sympathetic to the government or less. Hamlets pacified tells you how many villages were processed through a checklist. It does not tell you whether the checklist had anything to do with the war. Each number was clean, verifiable, defensible, and almost perfectly uncorrelated with the thing the war was supposedly about.¹

McNamara was not stupid. He was a systems analyst from Ford Motor Company, where quantitative management had worked, because cars are closed systems and wars are not. The technique transferred. The domain didn’t. And because the technique produced legible numbers at every intermediate step, no one inside the system had standing to argue that the aggregate was failing. Each metric was locally correct. The system was globally incoherent. And the distance between the two was invisible from inside, because from inside there were only the metrics, and the metrics looked fine.

I keep thinking about this not as a case study in institutional failure but as something closer to a feeling I recognize.

Proxies are not bad. I want to be careful about this because the lazy version of the observation is that measurement corrupts the thing being measured, and I don’t think that’s true, or at least it’s not true in the way that matters here. If I can’t bench press because of an injury, my dumbbell press is a perfectly good proxy. It measures a real thing that transfers to the thing I care about. The proxy is connected to the reality. Most proxies are fine. Some are excellent.

The problem is not that proxies exist. The problem is what determines which proxies you pay attention to. And the answer, in most cases, is not how well the proxy tracks the underlying reality. The answer is how close the proxy is to your face.

When you are young, almost nothing you do is the thing itself. Everything is a proxy for a proxy for a proxy. The SAT score points to the GPA which points to the university which points to the internship which points to the job which points to the salary which points to the life. Each layer adds distance from anything you can touch. Some of these layers are fine. The university really does teach you things. The internship really does build skills. The proxy chain is not fraudulent. It is just long, and noisy, and the noise has a specific structure: the proxies nearest to you are the loudest, regardless of whether they are the most informative.

Your roommate’s GPA is visible right now. The internship deadline is in three weeks. The person sitting next to you in lecture is doing something you can see and compare yourself to, involuntarily, continuously. Your own sense of what you actually want from your life is none of these things. It is not visible. It is not graded. It cannot be compared. Which means it loses every attentional competition to the thing that can be.

But the problem is worse than attention. The problem is that proxies are path-dependent. Committing to a proxy is not a measurement you take and put down. It is a fork in a graph. You walk down the branch. The further you walk, the more expensive it becomes to return to the fork and take the other path. Three semesters into pre-med is not the same as three semesters into anything else, because three semesters into pre-med means three semesters of organic chemistry and clinical volunteering and MCAT prep that do not transfer, and the path you did not take has been getting more expensive to reach with every term you spent on this one. You can feel the edges disappearing behind you. Each proxy you commit to forecloses alternatives, not absolutely, not irrevocably, but at a cost that increases with every step, and the cost is measured in the one thing you cannot get back, which is time.

This is what produces the anxiety. Not that the nearest signal drowns out the most important one, though it does. But that choosing the nearest signal is not a reversible decision. It is a traversal. You are exploring a graph in which backtracking is expensive and you cannot see more than one or two nodes ahead, and every step forward is simultaneously an investment in this path and a disinvestment in every path you are not on. The feeling is not confusion. The feeling is the specific dread of someone who suspects they may have missed a fork three nodes back and is calculating, in real time, whether the cost of returning is already too high.

There is something I notice about the people I know who seem to have built lives they actually inhabit rather than lives that look correct from outside, and the thing I notice is not that they are smarter or more disciplined. It is that at some point, usually without being able to articulate it, they found a way to make the longer proxy louder than the local one. Not by ignoring the local proxies. You can’t. They are right there. But by getting close enough to something they actually cared about that the signal from that thing could compete with the noise.

This is not a formula. I don’t think there is a formula. I think what happens is more like this: you stumble into something, or someone asks you a question at the right time, or you spend a summer doing something that was not on the approved list and it turns out the thing you were doing had a signal you could hear, and for the first time the comparison to your roommate faded, not because you decided to stop comparing but because the new signal was interesting enough that you forgot. That forgetting is the whole thing. You cannot will yourself to stop hearing the local proxy. You can only get close enough to something real that the local proxy stops being the loudest thing in the room.

If anyone gave themselves five years to do anything, they could probably do it. Five years of effort pointed at a single thing is enough to become genuinely good at almost anything. I believe this. The issue is not ability or discipline. The issue is that five years is a commitment to one branch of the graph, which means five years of watching other branches become unreachable, five years of not being able to answer the question at the dinner table in thirty seconds, five years of partial deafness to the local proxies that are rebuilt every semester. And every semester you spend on the five-year path is a semester that makes the twelve-week paths harder to return to if you’re wrong. The twelve-week cycle is not the enemy exactly. It is just so much less expensive to be wrong about.

I think about McNamara’s quarterly reports not as a cautionary tale but as a description of a feeling. The feeling of being inside a system where every local measurement says you are doing well and something at a different altitude, something you cannot quite name and certainly cannot graph, says otherwise. Not that you are failing. Just that the measurements and the meaning have become decoupled, quietly, in a way that will only be visible later, and the only evidence you have right now is a faint dissonance you are not sure you trust, because the numbers are right there, and the numbers look fine.

Most people I know who are anxious are not anxious because they are failing. They are anxious because they are succeeding at the local metrics and cannot tell whether the local metrics are connected to anything, and they can feel the graph narrowing behind them. The anxiety is not about the outcome. It is about the irreversibility. Every semester that the metrics look fine is also a semester further down a branch you chose before you understood the graph, and the question that produces the anxiety is not “am I doing well” but “am I doing well at the right thing, and can I still get back if I’m not.”

I don’t have a resolution for this. I don’t think there is one. The proxies are real and some of them are good and the problem is not structural but navigational, which means there is no policy fix, no system redesign, no clever reframing that makes the graph less expensive to traverse. The only thing I have ever seen work is proximity to something that matters enough to make the cost of the branch feel worth it. And that is not advice. That is just an observation about what seems to happen, sometimes, to some people, if they are lucky or stubborn or both.

McNamara retired in 1968. He spent the next thirty years not talking about Vietnam. In 2003, in a documentary, he finally admitted the metrics had measured nothing that mattered. He was eighty-five. It had taken him thirty-six years to hear the signal through the noise, and by then the noise was all anyone remembered.²