Sanity-Checking a Number Before You Say It

You've just multiplied out a decomposition tree and gotten "the average American household spends $14,000 a year on gasoline." Before you say it, is that number obviously wrong?

Pause and actually check before reading on — the whole point of this page is that the check takes ten seconds and catches the kind of error that sinks an otherwise good answer.

Why the multiplication isn't the finish line

A decomposition tree gives you a number, but a chain of five multiplied estimates has five separate places to drop a decimal, mix up a unit, or misremember an anchor. The multiplication is mechanical and easy to trust; the inputs are where errors live. A sanity check is a second, independent glance at the output that doesn't retrace the same chain of reasoning — because if you just redo the same multiplication, you'll make the same mistake again.

Three checks catch almost everything:

1. Order-of-magnitude bound. Is the number plausible against something you know cold? A household spending $14,000/year on gas means over $1,000/month — more than many households' entire grocery bill. That alone should trigger a re-check.
2. Compare to a known anchor. US household income is roughly $75,000/year median. Spending nearly 20% of that on gasoline alone, before food, rent, or anything else, is implausible on its face.
3. Re-derive via a different route. If gas costs $3.50/gallon and a household buys, say, 800 gallons a year (roughly 12,000 miles at 25 mpg for two cars combined ÷ shared driving — call it 640 gallons), that's $3.50 \times 640 \approx$ $2{,}240/year, not $14,000. The two routes disagree by a factor of 6 — so somewhere the first chain dropped a zero or mixed up per-car and per-household figures.

Worked example: catching the error

Trace the original $14,000 chain to find the bug. Suppose it came from: households own 2 cars on average × 12,000 miles/car/year × 1 gallon per 25 miles × $3.50/gallon:

$2 \times 12{,}000 \times \frac{1}{25} \times 3.50 = 2 \times 480 \times 3.50 = 960 \times 3.50 = 3{,}360$

That's $3,360, not $14,000 — so the "$14,000" version of this claim had an arithmetic slip somewhere upstream (a common one: multiplying by 25 instead of dividing by it, since $2 \times 12{,}000 \times 25 \times 3.50 \div$ nothing would land near that inflated range). The corrected number, $3,360/year, comfortably clears both sanity checks: it's a small, believable slice of a $75,000 household income (about 4.5%), and it roughly matches the independent $2,240–$3,360 range from the second-route estimate above. That's the number you'd actually say out loud.

A second example: catching a unit slip

Estimating total annual coffee spend in a city of 1 million people, you compute: 1,000,000 people × 0.3 (fraction who buy coffee daily) × 365 days × $4/cup = $438,000,000,000. Sanity-check it: $438 billion for one city's coffee spend is bigger than the GDP of many small countries — an instant red flag. Re-deriving: $1{,}000{,}000 \times 0.3 \times 365 \times 4$. Compute step by step: $1{,}000{,}000 \times 0.3 = 300{,}000$ daily buyers; $300{,}000 \times 365 = 109{,}500{,}000$ cups a year; $\times 4 =$ $438{,}000{,}000. The chain was arithmetically correct — the earlier "$438,000,000,000" simply had three extra zeros from a slip in stating it. $438 million for a city of a million people ($438/person/year, about $1.20/day) is completely plausible and passes the anchor check cleanly.

The generalizable skill

Sanity-checking is the closing move on every Fermi and market-sizing question: it's the difference between "I multiplied some numbers" and "I have a number I'd stake my credibility on." The reusable habit is to always have at least one independent check ready — a bound from common knowledge, a comparison to a trusted anchor figure, or a second derivation route — and to run it before you speak, not after someone else catches the error for you.