The State Learns to Count

Most states, for most of history, did not know how many people they governed. Counting is expensive, intrusive and resisted, and no administration undertakes it out of curiosity. Censuses were conducted for two purposes: taxation and conscription. Every other use a census now has (planning, representation, entitlement, research) is a later application of a machine built to extract.

That origin explains the resistance, which is otherwise puzzling to anybody who thinks of a census as routine. Being counted meant paying and serving. Undercounting was therefore rational, and it was widespread: households concealed young men, enumerators were paid by the day and worked from incomplete lists, and in some places had incentives of their own. Historians treat the totals as a lower bound and reconstruct from parish registers, tax rolls and household surveys, and the corrections are frequently large enough to change the conclusion.

The word itself preserves the history. Statistics arrives in English in the late eighteenth century from the German Statistik, meaning the systematic description of a polity: its territory, population, revenue and armed forces. The mathematics came later and took the name with it. Knowing the etymology is a useful corrective, because it makes explicit what is easy to forget: the numbers were collected by somebody with an interest in the answer.

The part that matters most is the categories, because a census does not only measure a society. It partly creates the one a state can act on. Funding formulas, electoral boundaries, anti-discrimination law, health planning and entitlement all run on categories, and a group that does not appear on the form is invisible to every one of them. That is why census disputes are so much fiercer than a measurement argument warrants. They are arguments about who exists administratively.

Colonial censuses are the clearest case anywhere of an instrument creating what it claims to record. Administrations in India and across Africa took fluid, overlapping, situational identities and required people to pick one exclusive box. Then they attached quotas, reserved positions and rights to the boxes. Within a generation the categories had organisations, grievances and a politics; administrators pointed to the hardened divisions as evidence of how divided the society had always been. The measuring instrument had manufactured its own finding and then cited it.

This leaves modern census debates in a genuinely difficult position rather than a simple one. Adding an ethnic category makes a group visible to policy, which is how it gets served. It also gives that group a formal existence it did not previously have, which is how divisions harden. Both effects are well evidenced and they point in opposite directions. What the history rules out is the comfortable claim that the form is a neutral instrument recording what is already there. It never has been, and the people designing it have always known.