Accuracy vs Speed: The Arithmetic

The intuition that fails here is that a typo costs one keystroke. It costs four separate things.

First, noticing. Errors are not detected instantly; you carry on for some number of characters before something registers as wrong. Call it several characters typed past the mistake.

Second, the backspaces. You have to delete not just the wrong character but everything typed after it, which is why noticing late is expensive twice over.

Third, the retype. Every character removed has to be produced again, at full cost.

Fourth, the restart. After a correction, the flow of reading ahead has broken and has to re-establish, which shows up as a slower few words rather than as a discrete pause.

Add those together and a single error costs somewhere in the region of a dozen keystroke-equivalents rather than one. That multiplier is the whole argument.

Take a typist producing 300 characters per minute of correct output — sixty words a minute by the five-character convention — with a 4% error rate. That is 12 errors in those 300 characters, which is unremarkable and would strike most people as reasonably accurate.

At a dozen keystroke-equivalents per error, those 12 errors consume the equivalent of about 144 keystrokes. In other words, roughly a third of that minute's total keystroke budget went to producing and repairing mistakes rather than producing text.

Now change one thing at a time. Increase raw finger speed by 10% and leave the error rate alone: more characters, but proportionally more errors, so the net gain is close to the full 10% and no better. Instead leave finger speed alone and halve the error rate to 2%: six errors instead of twelve, saving about 72 keystroke-equivalents out of a budget of roughly 444. That is a gain of around 16%.

Halving your error rate is hard. Increasing your raw speed by 10% is also hard. The point is that the harder-sounding of the two returns more, and it returns it without requiring your fingers to move any faster at all.

The multiplier is not constant. It grows with speed, for a reason that is easy to miss.

Noticing an error takes roughly a fixed amount of time, not a fixed number of characters. A faster typist covers more ground during that interval, so they type further past the mistake before catching it, so they have more to delete and more to retype. The same error costs a fast typist more than a slow one.

This is why error rates that were tolerable at forty words a minute become the binding constraint at eighty, and it is a large part of why so many people plateau exactly there. The technique that got them to eighty — pushing pace and correcting as they go — becomes self-limiting at the point where corrections consume the gains, which is one of the specific plateaus described in why typing speed plateaus.

There is a second-order effect worth naming, because it explains why some typists are much faster than their raw speed suggests.

An error you never make costs nothing. An error you catch immediately costs a few keystrokes. An error you catch five words later costs dozens. The distribution of your detection latency therefore matters almost as much as your error count — two typists with identical error rates can differ substantially in output if one notices immediately and the other does not.

Detection is trainable, and it is a different skill from accuracy. It is what the error recovery drill works on: not avoiding mistakes, but shortening the gap between making one and knowing. To put a number on your own overhead rather than an estimate, take the no-backspace test: with correction removed, the shortfall against your usual score is exactly what repairing was costing you.

It does not mean typing as slowly as possible. There is a pace at which your error rate stops improving because the errors that remain are not caused by haste, and going below it costs speed for nothing. The target is the fastest pace at which you are accurate, not the slowest pace you can manage.

It also does not mean accuracy practice replaces speed practice permanently. Both are needed, in an order: accuracy first because it raises the ceiling, then speed because it fills the space underneath. Doing them in the other order means building speed on top of a movement pattern that will have to be rebuilt.

And it does not mean that every error is equally expensive. A mistyped character in a first draft you will revise costs almost nothing. The same character in a search query, a password field or a command line costs a great deal more, which is why the argument matters most for the people whose typing has consequences.

The practical version is a rule about which number you watch. If you are tracking words per minute alone, you are tracking the output of the process and not the variable that most controls it.

Watch the gap between your gross and net figures instead. Gross is what your fingers produced; net subtracts the errors. A large and stable gap means the balance point described above, and it means the available gain is in accuracy rather than pace. How both figures are computed, and why different sites compute them differently, is set out in how WPM is calculated.

Then practise accordingly: a period at a deliberately reduced pace with a clean run as the only success condition, using the zero-error drill or the lesson on deliberately slowing down, followed by letting the pace return on its own. It does return, and it returns higher than where it started.