Study Planner

A free, browser-based tool that helps human factors and UX researchers plan two kinds of studies:

• Compare designs: compare two or more designs on measured outcomes such as task time, success, workload, and ratings.
• Find problems: plan how many participants a usability test needs to uncover problems in an early design.

It is built for applied work: small samples, plain-language inputs, results in your measure's own units, and honest uncertainty.

Use it

Once GitHub Pages is enabled, the tool runs at https://fomeroglu.github.io/study-planner/.

To run it locally, open index.html in a browser, or serve the folder:

python -m http.server 8765

then go to http://localhost:8765.

Privacy

Everything runs in your browser. Nothing you enter is uploaded, and there is no analytics or tracking. The only network requests are for the IBM Plex web fonts. The only thing saved in the browser is your light or dark theme choice.

What it does

• Independent variables and conditions, within or between subjects. With one variable, every pair of conditions is compared: with three conditions, after a significant overall test (Fisher's protected LSD), and with four or more, with the Holm adjustment. With two or more variables, you choose specific combinations and main effects, compared with the Holm adjustment.
• Measures: time, count, success / failure, rating scales (with NASA-TLX, SUS and SEQ presets) and other numeric measures, each with a planned test.
• The difference that matters, in your units or as "how many of 10 participants should do better", with your own estimate of variability if you have no data.
• A plan: the recommended number of participants, underpowered secondary measures, your analysis plan, a confidence report for the number you can recruit (with power curves and precision), a counterbalancing order plan, and reporting guidance.
• Every step from inputs to result is shown, with values from your own estimates labeled as such.

Limitations

• Within-subjects plans assume you can match each participant's values across conditions.
• Mixed designs, interaction comparisons and dropout adjustment are not part of this version.

How it calculates

docs/methods.md defines every calculation, with sources. The code implements that document. Reference values and their validation status are in validation/.

The calculations are validated against R. Every reference case in validation/answer-key.md matches R 4.6.1:

• t-test sample sizes and detectable differences: base R's power.t.test with strict = TRUE, which counts both rejection regions, as the pwr package does.
• t-test power and the two-proportion test: the pwr package (1.3-0).
• McNemar's test: 10,000 simulated studies at the planned sample size, analysed with mcnemar.test without continuity correction, the planned analysis.
• Variability from your estimate, precision, the rank-based adjustment, counterbalancing and problem discovery: the arithmetic, computed in R. The problem-discovery cases were also checked by hand against a published source.

One rule is still open: what happens when the chosen power is at or below alpha (methods.md section 4).

Project structure

Path	Contents
index.html, styles.css, app.js	The interface
logic.js	Form state to engine calls, and engine results to display text
engine.js	All statistical calculations (no interface code)
tests/	Automated tests (Node's built-in test runner)
docs/methods.md	Methods and sources
docs/manual-test.md	Click-through checklist for the browser
validation/	Answer key and reference R scripts
design/	Design reference screens

No build step and no dependencies.

Tests

With Node.js 20 or later, from the repository root:

node --test

License

MIT