Units of analysis in task-analytic research.

Haring et al. (1988) looked at how researchers count correct steps in task analysis.

They said simple percent-correct hides useful detail.

The paper drew a grid that marks competent performance on every single step.

The grid meets four big rules: it shows level, form, latency, and stability of each step.

Old percent scores miss these rules, so two kids with 80% can look the same while their patterns differ.

Ferron et al. (2017) and Wolfe et al. (2026) later tested masked visual analysis to cut false alarms in single-case graphs.

They extend the 1988 call for tighter rules by giving one clear decision safeguard.

Lanovaz et al. (2021) went further and let a computer read the graphs.

Machine scores beat human eyes, showing the grid idea can live inside software too.

Hinson (1988) wrote the same year and also slammed weak comparison studies.

Both papers push analysts to pick sharper tools, just for different parts of an experiment.

Next time you break a skill into steps, skip the one overall percent.

Draw a quick grid: rows are steps, columns are trials, cells get a check only for fast, stable, correct responses.

You will see exactly where the learner stalls and when mastery really happens.