The Index Is an Argument

A statistic arrives in public as a fact. Unemployment is this. Growth was that. The number is reported without the definitions underneath it, and the definitions are where the argument actually happened.

Every official series rests on decisions that could have gone another way. Who counts as looking for work. Whether a category is a household or a person. What is treated as investment rather than as consumption. Each is defensible and each excludes something.

The agencies are candid about this. The methodologies are published in full, revisions are documented, and the reasoning is available to anyone who wants it. Opacity is not the problem.

The problem is that the reasoning is published where nobody reads it and the number is published where everybody does.

This asymmetry has a predictable effect. A definitional choice made for sound technical reasons becomes, over a decade, the thing people mean by the word. The measure stops standing for the concept and starts replacing it.

The design of a survey is the clearest case. The categories it offers are the categories that can subsequently be counted, compared and legislated for. Anything outside them is not rare in the data. It is absent from it.

None of that is an argument against measurement, which is the usual bad-faith reading. A government that measured nothing would be worse in every respect, and the alternative to a contestable number is not a truer one but an anecdote.

It is an argument for treating the definitions as political objects rather than as technical furniture. They are chosen, they have distributive consequences, and they are almost never revisited once a series depends on continuity.

Continuity is the trap. A series is valuable because it is comparable over time, so changing a definition destroys the thing that made it useful. Where a measure is written into law, the incentive to keep measuring what was measured becomes an obligation.

So the definitions harden. They were set by people addressing the conditions in front of them, and they persist because the cost of changing them is borne now while the benefit is diffuse and arrives later.

The modest fix is publication practice rather than methodology. Require the headline release, which already appears in the public record, to state in plain language and above the fold the two or three choices that most affect the number, and what it would read if they had gone the other way.

That is not a caveat, and it should not be filed as one. It is the actual content of the measurement, and putting it where the number is would let people argue about the thing that was decided rather than about the thing that was reported.

Revisions make the point more sharply than the construction does.

A series is republished as better data arrives, so the number everyone argued about last quarter is quietly replaced by a different number nobody argues about at all.

The original figure has already done its work by then, moving markets and shaping coverage, and the correction reaches a fraction of the audience.

That asymmetry is a property of publishing rather than of statistics, and it is why the first estimate matters far more than its accuracy warrants.

The definitional choices are the most consequential and the least visible part of any official series.

What counts as employment, as a household, or as a price change is settled by a methodology document almost nobody outside the agency reads.

Changing one of those definitions moves the headline number without anything in the world having changed, and the change is announced in a technical note.