Should I focus on typing speed or accuracy?

These examples use a simple final-character accuracy definition. Actual studies may use minimum string distance to handle inserted, omitted and substituted characters, so the examples should not be mistaken for a reproduction of every site's accuracy metric. The measurement literature explains why scoring must specify the intended and transcribed strings. Soukoreff et al., 2003 Soukoreff et al., 2001

Remaining errors and corrected errors are different

A person who makes many mistakes and fixes all of them can finish with accurate text but spend substantial time correcting it. Someone who leaves mistakes may appear faster while producing less usable output. Final error rate, correction effort and entry rate consequently answer different questions. Soukoreff et al., 2003

The desktop dataset reports both uncorrected errors and correction-related measures. Faster participants generally showed fewer errors, but the study was observational. It does not show that simply forcing everyone to type faster will make them more accurate, or that slowing down by a fixed amount causes a specified improvement. Dhakal et al., 2018

Why there is no universal ideal percentage

The cost of an error depends on the task and what happens afterwards. A transcription benchmark measures supplied text; composing or editing involves additional decisions. Foundational typing and text-entry reviews distinguish these task components and explain why rates from one setting do not transfer directly to all work. Salthouse, 1986 MacKenzie and Soukoreff, 2002

A laboratory study also found large differences between ordinary-text and random-string typing, even among experienced participants. Passage familiarity is part of what a score measures. A personal best on a simple word list is not necessarily evidence of equal performance on punctuation-heavy prose. Feit et al., 2016

Device assistance changes the interpretation

Autocorrect may replace text without a one-to-one sequence of typed characters; prediction can insert a whole word. The mobile study explicitly discusses the resulting limitations of keystroke-level error metrics. Comparing assisted phone entry with unassisted desktop entry as though their accuracy percentages were identical can obscure meaningful differences. Palin et al., 2019

What is a reasonable practice approach?

Use the same task and record rate together with the relevant error measure. Compare blocks of similar unfamiliar passages rather than selecting a single run. This is a measurement recommendation, not a claimed optimal training dose. The evidence on practice history among everyday typists supports task-specific proficiency but does not establish a universal speed-versus-accuracy training ratio. Keith et al., 2007 MacKenzie and Soukoreff, 2002

If a faster run leaves more errors, describe that as a trade-off rather than an unqualified improvement. If final accuracy stays constant while correction effort falls, that is useful information too. The point of multiple measures is to make the change visible, rather than compressing every outcome into one flattering number. Soukoreff et al., 2003 Soukoreff et al., 2001

Improvement should also be described at the level tested. Better typing is a useful skill; the cited keyboard studies do not turn WPM or accuracy into a measure of intelligence, attention disorder or general cognitive enhancement. Their conclusions concern text entry under defined conditions. Dhakal et al., 2018 Feit et al., 2016