The app is MCPower without code: the same native engine as the Python and R packages, behind a point-and-click interface. It ships in two forms with an identical interface — the web app runs in your browser with nothing to install, and the desktop app is a download (install guide).

The window splits in two: configuration on the left, results on the right (a narrow window collapses them into one pane with a toggle). A tour, top to bottom:

Pick a family

The ribbon in the header selects the analysis family — ANOVA, Regression (continuous or binary outcome), or Mixed effects. Each family keeps its own configuration, so switching back and forth loses nothing. Next to the ribbon sit Settings (appearance, run configuration, and the scenario definitions), History (a searchable list of past runs — click one to replay it), and Tutorial, which opens these pages.

Describe the model

The left panel is the model description, as collapsible cards in workflow order:

  • Upload data (Regression and Mixed effects only) — hand the simulation a pilot CSV instead of describing every predictor from scratch.
  • Model — the formula box, then one card per predictor: its type (continuous, binary, or factor) and the standardised effect size to detect. The panel adapts to the family: Regression adds a Continuous/Binary outcome toggle (Binary adds a baseline-probability field), Mixed effects adds the cluster card (cluster name, ICC, number of clusters), and ANOVA swaps the formula for structured factor and covariate rows.
  • Correlations (not ANOVA) — optional: pairwise correlations among continuous predictors.
  • Run — target power, α, which coefficients to test and the multiple-testing correction, plus the inputs for the two run modes: Find power takes a fixed sample size n; Find sample simulates a from/to grid and reports the n where the fitted power curve reaches the target (how required N is estimated).

Run it

The action bar above the results pane summarises the current target, α, and sample-size range, and holds the two run buttons — Find power and Find sample — alongside the Scenarios toggle (re-run the same design under the optimistic / realistic / doomer assumption sets; see scenario analysis) and a status badge (Ready to run, Running…, Last run done). While a run is in flight the buttons are replaced by Cancel.

Read the results

Before the first run, the results pane shows a get-started checklist and a short guide to the active family. Each run then lands in its own tab with four views:

  • Summary — the power table and plot. With Scenarios on, one chip per scenario plus an ⧉ Overlay chip shows all scenarios side by side on a 3-column grid. For a Find sample run the table's Required N is the model-based estimate read off the fitted power curve, with a 95% CI column in single-scenario runs — see how required N is estimated for the ≤ / ≥ / appr. markers and the non-monotone warning.
  • Joint dist — how many of your tested effects reach significance together in the same simulated study. When ≥ 2 effects are tested, two charts appear: At least k (P(≥ k significant) vs N) and Exactly k (P(exactly k significant) vs N, including k = 0).
  • Script — the equivalent script for the analysis you just configured, ready to copy when you outgrow the GUI. A Python | R toggle next to the Copy button switches the output language; the choice is remembered globally and persists across sessions.
  • Export — pick a plot block from the selector and save it as a PNG or SVG file. Saved files use the print style (white background, black axes, colourblind-safe palette) — the same as save_plot() in Python and R. The on-screen charts keep the app's live colour theme.

Export

The Export tab saves a run's charts as image files. Pick which chart from the Chart selector — the power-by-effect plot, the power curve, any per-scenario panel, or the overlay — choose PNG or SVG, and click Save; PNG adds a Scale field (1–4×) for higher-resolution output. Saved files use the print theme (white background, colour-blind-safe palette), independent of the app's on-screen colours, matching save_plot() in Python and R.

When something goes wrong

The app never fails silently:

  • A run that fails shows an error card in the results area with the engine's message and a Copy details button — so you can read (and copy) exactly why it stopped. Dismiss it to return to your previous results, which are kept. (This replaces the old, dead-end "see console" status badge.)
  • A formula or model problem is flagged inline, under the formula box, the moment it can't be turned into a valid model — fix it there and the run buttons re-enable.
  • A background hiccup (for example, settings or history that couldn't be saved) raises a brief toast in the corner with a Details option, rather than losing your change without a word.

The panel guides

One page per panel, in reading order:

  1. Upload data — driving the simulation from a CSV.
  2. Regression — OLS and logistic power, panel by panel.
  3. Mixed models — clustered designs in the GUI.
  4. ANOVA — factorial designs and post-hoc power.