Skip to contents

anovapowersim

anovapowersim is designed to make determining a priori power for ANOVAs as easy as possible. You can add as many within/between factors with as many levels as you would like. There’s no need to estimate condition means, SDs, or repeated-measures correlations; just enter the target partial eta squared.

The package simulates data and estimates power based on the specified design. It also provides direct power calculations for comparison.

Getting a priori power for a 2 × 2 × 3 mixed interaction effect is as simple as running the following:

install.packages("anovapowersim") # if not already installed

library(anovapowersim)

power_n(
  between = c(group = 2), # group has 2 levels
  within = c(stim = 2, cond = 3), # stim has 2 levels, cond has 3
  term = "group:stim:cond", # three-way interaction term
  target_pes = 0.08, # target effect size
  n_sims = 5000, # increase to 10000+ for more precise estimates
  power = .90,
  alpha = .05,
  parallel = TRUE, # simulations will be run in parallel for speed
  seed = 123 # for reproducibility
)
#><anovapowersim_curve>
#>  term:          'group:stim:cond'
#>  target power:  0.900
#>  alpha:         0.05
#>  effect size:   pes = 0.08
#>  n values:      8 per-cell sample sizes visited
#>  sims per cell size: 5000
#>  SS type:       III
#>  n needed for between-subjects cell: 38
#>  total N needed: 76
#>
#> n_per_cell total_n n_sims num_df den_df    ncp power_calc power_sim
#>         31      62   5000      2    120 10.435      0.823     0.825
#>         37      74   5000      2    144 12.522      0.890     0.885
#>         38      76   5000      2    148 12.870      0.899     0.903
#>         39      78   5000      2    152 13.217      0.907     0.901
#>         40      80   5000      2    156 13.565      0.915     0.918
#>         41      82   5000      2    160 13.913      0.922     0.916
#>         46      92   5000      2    180 15.652      0.949     0.947
#>         62     124   5000      2    244 21.217      0.989     0.988

Additional power analyses

The development version includes several experimental power-analysis options. Their full examples and guidance are kept in the dedicated guides linked below.

Achieved power and sensitivity

At a fixed sample size, power_achieved() estimates power for a chosen partial eta squared, while power_sensitivity() estimates the minimum detectable partial eta squared. See the fixed-sample tutorial.

Calculation-only functions

The _calc() functions skip simulations and use calculated noncentral-F power. They also support planned nonsphericity through epsilon. See the calculated-power tutorial.

Power for unbalanced designs

power_unbalanced() simulates one exact allocation from user-defined cell means and sample sizes under a common standard deviation and optional within-subject correlations. It is simulation-only and does not extrapolate how unequal cell sizes should scale. See the unbalanced-design tutorial.

Installation

anovapowersim can be installed from CRAN:

install.packages("anovapowersim")

You can install the development version from GitHub:

install.packages("pak")
pak::pak("shaheedazaad/anovapowersim")

Or, with remotes:

install.packages("remotes")
remotes::install_github("shaheedazaad/anovapowersim")

Citation

Azaad, S. (2026). A priori power analysis for ANOVA interaction effects with the anovapowersim R package: a short introduction. https://doi.org/10.31234/osf.io/86rsy_v1.

Limitations

anovapowersim is designed to be simple and easy to use first, which means it has some limitations for now. It does not support:

  • Covariates (ANCOVAs)
  • Sample-size searches or power curves for unbalanced designs
  • Huynh-Feldt corrections in power simulations. Greenhouse-Geisser-corrected simulated tests are supported for sums-of-squares type II or III when a custom covariance implies epsilon < 1; type I tests remain uncorrected.
  • Heteroskedastic ANOVA. Simulation functions require one common marginal variance; unequal correlations and Greenhouse–Geisser corrections remain supported for repeated-measures designs.

Simulation functions warn when their default common sd = 1 or default within-subject correlation of 0.5 is used. Covariance specifications also warn when only some correlations are defined; the default correlation fills only the undefined pairs. - Simple main effects/pairwise comparisons

Other packages

I recommend checking out Superpower, which handles some of the limitations above.