MCPower answers two questions about a study before you run it: how much statistical power a given sample size buys, and what sample size reaches a power target. It works by simulating the study many times on a native engine, so the same short workflow covers OLS, logistic, mixed, and factorial models.

Install

install.packages("mcpower", repos = "https://r.mcpower.app")

This installs a prebuilt binary on Windows, macOS, and Linux. Need a specific R version, an offline install, or the previous R release on Linux? See manual binary downloads. The package is also on R-universe. On other systems (Intel Mac, older Linux, older R) compile the current source instead:

remotes::install_github("pawlenartowicz/mcpower", subdir = "ports/r", build = FALSE)

A first analysis

Describe the model as a formula, mark which predictors are binary, set the effect sizes you want to be able to detect, and ask for power:

library(mcpower)

# Does a treatment lift satisfaction, controlling for age?
model <- MCPower$new("satisfaction ~ treatment + age")
model$set_variable_type("treatment=binary")
model$set_effects("treatment=0.5, age=0.3")

result <- model$find_power(sample_size = 120, target_test = "treatment")

Power Analysis — OLS  N=120  sims=1600  α=0.05  target=80%
formula: satisfaction ~ treatment + age

───────────────────────────────────
Test                 Power   Target
───────────────────────────────────
treatment            77.4%      80%
───────────────────────────────────

That is the whole loop. At n = 120 this design has 77.4% power to detect the treatment effect — just under the conventional 80% target, so you would want a few more participants. The next page asks both questions properly.

The tutorial ladder

Each rung adds one modelling idea and powers it. Climb from the top:

  1. First analysis — the two questions, in full
  2. Interactions — a:b terms
  3. Correlations — correlated predictors
  4. Logistic regression — a binary outcome
  5. Mixed models — grouped / hierarchical data
  6. ANOVA & post-hoc — factors and pairwise contrasts
  7. Multiple testing — corrections across many tests
  8. Custom scenarios — robustness sweeps
  9. Upload data — drive simulation from a CSV or dataframe
  10. Model misspecification — what testing the wrong model costs

New to power analysis? The Concepts pages explain effect sizes, model specification, and what the power number actually means.