What your estimates have in common#
Back of the Envelope · Week 8
Errors that differ partly cancel#
Ask ten people and the errors point in different directions — for reasons particular to each of them. Put the estimates together and those differences work against each other.
That is the whole mechanism. It only works when the errors actually differ.
Round A · Round B#
Round A — Hobart Airport, August 2025. The question, and nothing else.
Round B — Gold Coast Airport, August 2025. Everyone starts from the same real published figure: Gold Coast, August 2021 — 12,485 passengers.
Written, folded, handed in. No conferring, either round.
12,485 → ?#
Gold Coast Airport passengers, August 2021 → August 2025.
Reason from the reference however you like.
12,485 → 505,540#
Roughly a factor of forty. August 2021 fell in a period of domestic border closures — a real, correctly sourced figure, and the wrong one to reason from.
Scatter, or cluster#
(schematic — shape only, not measured data)
Independent errors — spread around the truth
Shared input — clustered tight, off target
Five polls, unusually close, all wrong the same way#
May 2019: every published poll had Labor winning the Australian federal election. The Coalition won.
An inquiry found this met the criteria of a polling failure, not a miss — the polls erred in the same direction, at a similar level, and the error was in the polls themselves.
“…no direct evidence of herding…”
The inquiry into the 2019 polls, on five separate organisations with five separate methods.
More data does not dilute a shared mistake#
It reproduces it — a shared input passes through every estimate at full strength, no matter how many you collect.
The check is procedural, not statistical: did they start from the same figure, the same dataset, the same training, the same room?
And the room you are in#
Your own estimates this semester were made after the same lectures, the same worked examples, the same reference points.
Where they agree, some of that is real — and some of it may be the course. Worth examining, not something anyone can settle for you.