How to estimate anything
The handful of techniques that answer most of the questions on this site.
Nobody knows how many bananas the world eats in a year. The useful skill is not recall but reconstruction: getting to within a factor of two of a number you have never seen, using things you already know. That is what this game trains, and it is a genuinely portable skill — the same moves that place a global banana harvest will place a project estimate, a market size or a repair bill.
Physicists call this Fermi estimation, after Enrico Fermi, who was known for producing startlingly accurate answers to questions nobody had measured. The method is not clever. It is the deliberate refusal to guess a big number directly, and the discipline of breaking it into smaller numbers you can defend.
What follows is the working set. Every one of the techniques below is used somewhere in the several hundred questions on this site, and most questions yield to two or three of them combined.
Six techniques that do most of the work
1 Decompose into factors you can defend
Never estimate a total. Estimate the things it is made of and multiply. A national fast-food chain's annual sales are outlets times sales per outlet per day times 365 — three numbers you can each argue about, producing one you could not have guessed. If a factor feels shaky, break it down again.
2 Anchor on something you actually know
Start from a quantity you hold firmly and adjust. The world population, the distance to the Sun, the size of a familiar country, the price of a coffee — a good anchor plus a ratio beats a direct guess almost every time. The skill is choosing an anchor in the same category as the target.
3 Convert between rates and totals
A huge annual figure becomes judgeable the moment you turn it into a rate. Six hundred million Big Macs a year is meaningless; seventeen a second is something you can weigh. The conversion runs both ways, and doing it is the fastest sanity check available on any large number.
4 Reason in orders of magnitude
For very large or very small quantities, aim at the exponent and stop. Whether there are two trillion galaxies or four is a detail; whether it is trillions rather than millions is the whole question. Being right about the power of ten is usually the useful part, and chasing digits beyond it wastes the effort.
5 Divide by the population
Any global or national total should be converted to a per-person figure before you believe it. Seven hundred cigarettes per person per year, two views of a video per human alive, four hot dogs each — a per-capita number can be checked against your own experience, and an absolute one cannot.
6 Bound the answer from both sides
When you cannot compute a figure, fence it in. Find a value it certainly exceeds and one it certainly cannot, and the answer sits between them. Team sizes are bounded by pitch size, building heights by lift travel times, and record performances by human physiology — a tight bracket is often better than a shaky calculation.
Five traps that produce confident wrong answers
Generalising from a wealthy country
Most of the world does not live the way a European or American reader does. Car ownership, internet access, meat consumption and healthcare spending all look completely different from a rich-country sample, and any global figure estimated from one lands badly low or badly high.
Confusing the mean with the median
Incomes, house prices, playtimes and spending are all right-skewed: a few very large values drag the average above what a typical case looks like. Where both exist, the median describes the middle and the mean describes the tail. Ask which one a statistic is quoting before you use it.
Reading fame as frequency
The examples that come to mind are the memorable ones, not the common ones. Bridges are mostly highway overpasses, not suspension spans; airports are mostly dirt strips, not hubs; buildings in a famous skyline are mostly low-rise. Whenever a count seems too large, the category probably includes a mundane majority.
Ignoring the definition
A city can mean a municipality or a metropolitan area, a count of accounts is not a count of people, and food availability is not food eaten. More wrong answers come from estimating the wrong quantity than from estimating badly, and a question that specifies a boundary is usually warning you that one exists.
Extrapolating linearly through exponential change
Anything growing or shrinking exponentially — electric car sales, model training costs, satellite counts, storage capacity — makes an impression formed five years ago useless. If a quantity doubles every couple of years, an old anchor is not slightly wrong but wrong by a multiple.
Why every number here has a source and a year
An estimate is only as good as the figure it is checked against, which is why every question on this site names where its number came from and what year it refers to. A statistic without a date is not a fact; it is a fact that was true once.
It also means you can disagree with us. If a figure looks wrong, the source is one click away and the correction address is on the about page. Several numbers here have been revised because a reader followed the link and found that the underlying report had moved.
Estimation is a skill you build by being wrong in a measurable way. That is what the daily game is for: five numbers, a score for how close you got, and an explanation of the reasoning you could have used. Do it often enough and the moves above stop being a list and start being a reflex.