MCPower vs G*Power, pwr, simr, superpower

Most power calculators are either formula-based (fast but limited to the designs their formulas cover) or simulation-based for one narrow family. MCPower is simulation-based across OLS, logistic regression, mixed-effects models, and ANOVA in one tool — so you do not have to switch calculators when your design has a binary outcome, clustered observations, or an interaction.

Feature comparison

"partial" means the tool covers a restricted version of the feature (e.g. a single-family formula, Bonferroni only, or a web form with limited model types).

Tool-by-tool notes

G*Power is the long-standing reference for formula-based power. It covers dozens of statistical tests very quickly, and its GUI is widely taught. It does not do simulation, so it cannot handle correlated predictors, non-standard distributions, or custom interaction structures — and mixed-effects designs are out of scope entirely.

pwr (R package) offers the same closed-form approach in code: fast, exact for the supported tests (t-tests, chi-squared, one-way ANOVA, correlation, linear models), and zero configuration. The right choice for those tests. It stops at simple designs.

superpower (R package) is simulation-based for factorial ANOVA designs and handles multiple-comparison corrections well within that family. It does not cover regression or mixed-effects models.

simr (R package) is the closest relative for mixed-effects power: genuinely simulation-based, driven by a fitted lmer / glmer model object. Its strength is that the simulation parameters come straight from a pilot fit; its limit is that you need a pilot model to start from, and the interface is R-only with no GUI. MCPower covers overlapping ground while also supporting pilot-data upload, GUI access, and the other model families.

WebPower (R package + web app) is formula-based for a wide range of tests and exposes them through a web form. It covers mixed models for some standard structures. It does not do simulation.

Speed

MCPower's native engine runs simulations in a compiled, multi-core Rust kernel, so it is typically faster than pure-R or pure-Python simulation tools for the same number of simulations.

[needs measurement] Quantitative benchmarks — wall-clock time per 1,000 simulations for a common OLS, logistic, and LME design — have not yet been measured against each alternative. Any future figure must state the machine (CPU, core count), the simulation count, and the exact designs timed alongside the number. Self-reported figures in tool documentation are not compared here.

For the browser app, the same engine runs in a worker pool inside the browser tab — no server round-trips, no installation required. Simulation count versus precision is the only knob you control; see why it's fast for how that trade-off works.