How We Measure Typing Speed

What counts as a word

Typing tests do not count dictionary words. By industry convention, one word is 5 characters, including spaces. The convention dates back to the typewriter era and it is what makes "hi to me" and "elephants" cost the same: both are about two standard words. Every mainstream test uses it, and so do we. See words per minute on Wikipedia for the background.

WPM, the headline number

Our WPM is the modern convention used by most enthusiast typing sites: only characters that belong to correctly typed words earn credit, plus the space you typed after each correct word. Divide by 5, divide by minutes.

wpm = (correctWordChars + correctSpaces) / 5 / minutes

If a timed test ends while you are mid-word, what you have typed of that word counts only if all of it is correct so far. One mistake anywhere in the unfinished word and it earns nothing, the same as a committed wrong word. You never lose credit you already banked.

Raw WPM

Raw WPM answers a different question: how fast are your fingers moving, regardless of correctness? It counts every character you typed, correct, incorrect, or extra.

rawWpm = allTypedChars / 5 / minutes

Raw is always at least as high as WPM. The gap between the two is a direct picture of how much speed your errors are costing you, which you can track over time in your typing stats.

Accuracy

Accuracy is judged at the moment you press a key, and the judgment is permanent. Fixing a mistake with backspace restores your WPM credit for the word, but it never restores accuracy. Backspace itself is not a counted keystroke.

accuracy = 100 * correctKeypresses / (correctKeypresses + incorrectKeypresses)

This is deliberate. If corrections could erase errors, accuracy would always read near 100% and would tell you nothing about your typing. Held at press time, it becomes an honest measure of how often your first instinct is right.

Consistency

Consistency measures how even your speed was from second to second. We build a per-second raw speed series, compute its coefficient of variation (population standard deviation divided by the mean), and map it onto a 0 to 100 scale:

cov         = popStdDev(rawPerSecond) / mean(rawPerSecond)
consistency = 100 * (1 - tanh(cov + cov^3/3 + cov^5/5))

The odd-power series inside the tanh is the Maclaurin expansion of atanh, so for small variation this behaves like 100 x (1 - cov), with a harsher rolloff as your speed gets spikier. A metronome-steady run scores near 100; a burst-and-stall run scores much lower.

The four character classes

Your results screen breaks characters into four buckets:

• Correct: the right character at that position, whether or not the word around it ended up right.
• Incorrect: the wrong character typed at a position.
• Extra: characters typed past the end of a word.
• Missed: expected characters you never typed, usually because you pressed space early and skipped ahead.

These four counts describe characters, not words, so they do not line up one-to-one with WPM. Type teh for the and you get one correct and two incorrect characters, but the word earns no WPM credit at all: a single wrong letter voids the whole word's credit while leaving the letters you did hit counted as hit. Missed characters count in neither WPM nor raw WPM, since nothing was typed, and they void the word's credit too.

For the same reason accuracy and the character counts can disagree, and neither is wrong. Accuracy is a tally of key presses at the moment you made them; the character counts describe the text as it stood when the test ended. Fix a typo before committing the word and the incorrect character disappears from the breakdown, while accuracy keeps the miss - so a run can honestly read 95% accuracy alongside zero incorrect characters.

Edge cases, in plain language

• The clock ends mid-word: an error-free partial word earns credit for the characters you typed, a partial word containing a mistake earns nothing, and the untyped remainder is never counted as missed.
• You skip a word with an early space: the untyped remainder is missed, the word is wrong, but the characters you did type keep their raw WPM and accuracy standing as pressed.
• You type past the end of a word: extras count toward raw WPM, count against accuracy, and the word can no longer be correct.
• You fix an error before committing the word: the word can still earn full WPM credit, but accuracy keeps the original miss. This is why WPM is always less than or equal to raw WPM.
• The space after a fully correct word counts toward WPM; the space after a wrong word earns nothing.

Traditional gross and net WPM

Older tests, and many employment tests, use a different pair of formulas:

grossWpm = allTypedChars / 5 / minutes
netWpm   = max(0, grossWpm - uncorrectedErrors / minutes)

Gross WPM is the same quantity as our raw WPM under a traditional name. Net WPM subtracts one full word for every error left in the final text, floored at 0 because the penalty can otherwise go negative.

For error analysis we follow the taxonomy formalized by Soukoreff and MacKenzie (CHI 2003), which separates corrected errors from errors left in the final text. Their distinction is exactly why accuracy and WPM treat corrections differently.

Conversions

cpm = wpm * 5      // characters per minute
kph = wpm * 300    // keystrokes per hour

CPM is common in Europe, and KPH is the standard for data-entry work, where typical job requirements fall around 8,000 to 12,000 KPH. That range corresponds to roughly 27 to 40 WPM. The number typing test reports both.