How Countries Count Their People — Census, Sampling and the Numbers Behind the Numbers

From Tukuk-OS Wiki, the free knowledge base.

Every ranking you have ever argued about online — population, GDP, literacy — ultimately traces back to someone knocking on a door, or deciding not to. The census is one of humanity's oldest technologies of government, and one of the few whose errors are inherited by every statistic built on top.

The Oldest Government Software

Counting subjects is ancient: the Roman census registered citizens and their property for tax and military duty; a Han-dynasty record from the year 2 CE counts some 57 million people, one of the most accurate pre-modern censuses anywhere; William the Conqueror's Domesday Book of 1086 was essentially a property census with legal teeth. Modern decennial censuses — the US every ten years since 1790, most states each decade or in registers since the 1800s — added the statistical machinery we recognise: definitions, schedules, enumerators, and published tables. The key invention was treating the count as a system: undercounts, coverage ratios and post-enumeration surveys are studied like engineering tolerances.

Headcount Versus Reality

A census enumerates everyone on a chosen night, but "everyone" is contested politics: who counts as usually resident? Are students counted at home or at college? Do undocumented residents get a question? Are citizens asked about ethnicity or language, and who decides the categories? The US citizenship-question fight, Europe's register-based counts (the German Zensus rebuilt a register instead of mailing forms), and China's household-registration system all show that the method is the policy. Then there is pure logistics: remote islands, nomadic peoples, high-rise buildings with locked doors, disaster weeks. Every country publishes a different estimate of how many people it missed, and those error bars quietly propagate into every per-capita statistic that follows — including the country comparison pages at tukuk.org/statistics, where the underlying figures come from UN and World Bank compilations with their own correction models.

Counting Less to Know More

Because full enumeration is expensive and slow, official statistics increasingly leans on sampling: interview a few thousand households a month and you get unemployment in near-real time (the US Current Population Survey, the UK Labour Force Survey). Sampling raises the same privacy question you meet in any modern dataset — how do you learn about millions without reading over individual shoulders? The statistical answer is the same trick used in privacy-preserving analytics, explained in plain terms in counting people without reading over shoulders. The 20th-century insight — that a carefully drawn fraction can estimate the whole with measurable confidence — is why censuses can be complemented, but never replaced: only a full count can show where the rare people are, district by district, for the hospital, the school and the bus route.

Counting Money Too

Population was the first big number; GDP is the second. National accounting, formalised after Simon Kuznets's 1930s work for the US government, defines what counts as production — and its choices are visible: ignoring unpaid housework, treating owner-occupied housing as if you rent it from yourself, struggling with digital free services entirely. When two countries' GDP "can't be compared," it is usually because one counts the informal market by survey and the other guesses. The honest habit, before quoting any ranking, is to open the sources tab: which census year, which revision, which estimate method. Free data endpoints for exactly that verification are collected in free world data APIs with no key and the broader open data guide. Numbers are not the enemy of arguments — unexamined numbers are.

See also

• When Averages Lie: Mean, Median, and the Story They Hide
• Weather vs Climate — The Real Difference, and Why the Confusion Costs Us
• How Music Charts Work — Streams, Sales and What Number One Really Means