How the Element Miss List works

What counts as an answer

The quiz shows an element's symbol and asks for its name. Type a name, or choose "I don't know" — either way, that's one answer to that element. Runs come in 20, 50, or 118 cards, drawn at random, so every element gets asked about equally often no matter which length people pick.

A typed answer is correct when it matches the element's name. We ignore capitalization, extra spaces, and a trailing period; we accept the international spellings (aluminium, caesium, sulphur) and wolfram for tungsten; and we forgive one slipped letter on names of five letters or more — "berylium" counts for beryllium. Two rules protect the data from that forgiveness. First, an answer that is exactly another element's name is always wrong: typing "cerium" for curium is confusion, and confusion is the thing this page measures. Second, a slip that lands one letter from two different elements is wrong too, because we can't know which one was meant. Short names must be exact — one letter off "tin" is a different word, not a typo.

Which answers we analyze

Every run is scored and recorded. To keep replays from stacking the deck, the analysis counts each device's first answer to each element per day (UTC): meet chlorine in two runs the same day and only the first answer counts. Play tomorrow and you count again.

Answers come from two pools: site visitors, and a commissioned panel of US parents of high schoolers (each panelist covers about 40 elements, drawn so all 118 get covered evenly, with a participant id on every answer). The leaderboard's default view sums both pools — same quiz, same grading — and the page header says how many answers came from each. The per-pool views are one click away, and a claim cites the pool that can carry it: a sentence about parents cites the panel (the pool with a defined population), a sentence about overall answer volume cites all answers with its N. Around 100 panel answers per element carries roughly a ten-point margin either way, so treat small gaps as noise.

What we throw out

• Panel runs that miss an attention check. The panel includes two instructions like "type the word iron" — miss one and the whole run is out. (The check questions themselves never count toward any element.)
• Runs typed faster than a person can read: typed answers averaging under 1.5 seconds each.
• Runs that answer on autopilot: one identical typed answer covering 90% or more of 10-plus typed answers.

Keeping scripts out

Anyone can take the quiz without an account, which means anyone could try to feed it junk. The defenses: the server issues every run and caps them per device (6 an hour, 12 a day), each question accepts exactly one answer, more than 25 answers inside 10 seconds gets refused as faster than human, and requests flagged as bots are turned away. The server does the grading — a browser can't claim its own answers were right. Panel runs also require the study's key, so that pool only fills while a commissioned study is running.

Small numbers, plainly marked

Nothing is hidden. The leaderboard shows all 118 elements from the first quiz on, with raw counts and a miss percentage. Below 25 answers the percentage carries a provisional mark — two misses out of three answers is a fact, but 67% would be a statistic pretending to be one. The "most common miss" column only counts wrong answers that are themselves element names; typos and other stray text count as misses but never appear as confusion pairs. The leaderboard's default order is miss rate, not miss count — free gameplay concentrates on the first elements of the table, so raw counts track exposure, and rate is the fair comparison. Rates with fewer than 25 answers sort behind rates that have earned their place.

The lookup-demand table

This one has nothing to do with the quiz. It shows monthly views of each element's Wikipedia article (latest complete month, from the Wikimedia pageviews API) next to monthly search volume where we track the element's name as a keyword. Views and searches are different units, so we never add them together. "Not tracked" means we don't track that keyword — it doesn't mean zero.

When things update

• The live counter and miss rates recompute every few minutes.
• Each August we publish the annual edition — a fresh look at the year's data.
• The lookup-demand snapshot refreshes with each edition.

Privacy

Everything on the Miss List is a total, never a person. Answers are stored against an anonymous device token (panel answers against a study participant id), and no individual result is ever published.

Questions about the data? The full per-element table is downloadable as a formatted Excel workbook (or raw CSV for scripts).