A census is an attempt to count an entire population at a single reference moment. The questions are the easy part; reaching every household is the work.
The reference date fixes the picture
A census defines one moment and asks who was where at that instant. Births, deaths and moves after that date belong to the next count, not this one.
Without a fixed date, people who move during the collection period would be counted twice or not at all, and the totals would drift depending on when each area was visited.
The rule also settles awkward cases. Students, seasonal workers and people in hospital are assigned to a defined usual residence rather than to wherever they happened to sleep.
Address lists are built before any counting
Before forms go out, the statistical agency assembles a list of every dwelling. This is compiled from property registers, postal records and, where necessary, physical checks of the street.
The quality of that list caps the quality of the count. A dwelling missing from the list receives no form and generates no follow-up, so its residents go unrecorded.
Informal housing, subdivided flats and newly built streets are the persistent problems, because they appear in no register until someone walks the ground and adds them.
Non-response is handled by escalation
Most households respond on their own, increasingly online. The remainder trigger a sequence of reminders, then a visit from a field officer.
Field work is expensive and concentrated in a minority of addresses, so agencies target it using response rates by area rather than spreading effort evenly across the country.
Even after visits, some households are never contacted. Their characteristics are estimated from neighbouring records using published methods, and the extent of that imputation is reported alongside the results.
Coverage is measured separately
A census does not assume it counted everyone. A separate survey of sampled areas is conducted independently and matched against the census records for the same addresses.
The mismatch between the two exercises estimates how many people were missed and how many were counted twice. The published totals are adjusted for both.
Undercount is not random. It concentrates among young adults, recent movers and households with insecure tenure, which is exactly why the adjustment is calculated rather than assumed away.
The results carry money and seats
Census figures determine how electoral boundaries are drawn and how funding formulas distribute money between regions, so an undercounted area loses on both counts.
That consequence explains the intensity of debate over methods. Choices about residence rules and adjustment have distributive effects long after the count itself is finished.
It also explains the confidentiality rules. Individual returns are locked away for decades, because the count only works if people believe their answers will not be used against them.