How wrong can you be, and still be useful?#
Back of the Envelope · Week 1
25,417,978#
Eight digits. Count them.
The 2021 Census count of Australian residents present on Census night — ABS, 2021 Census overcount and undercount.
“These projections are not predictions or forecasts.”
The ABS, on the page that also shows you a population clock advancing person by person.
Four questions, one number#
- Precision — how finely it is written
- Accuracy — how close it sits to the thing
- Uncertainty — the spread it is really known within
- Sufficiency — whether it is good enough for this decision
Only the last one is yours to settle. Settle it first.
What the count missed#
- Census count — 25,417,978
- PES estimate of who should have been counted — 25,608,022
- Net undercount — 190,044 persons, or 0.7%
The lowest net undercount recorded for an Australian census. It was 1.0% in 2016.
Better than you guessed, probably.
17.4%#
2021 Census net undercount for Aboriginal and Torres Strait Islander people.
Which population is the decision about?#
Two teaching exercises — not descriptions of any real allocation process:
A. A national planning task sized in millions: do a few tenths of a percent change the decision?
B. A service for a community of a few thousand, counted with substantially higher uncertainty: does the same relative error change it?
Before you leave#
This number is precise enough to decide ______, and not precise enough to decide ______, because ______.
And the habit that starts today: commit to a range before you look it up.