(Version 0.2.0, updated on 2026-03-18, release history)
This package is for power analysis and sample size determination for moderation, mediation, and moderated mediation.
It includes functions for power analysis and sample size determination for moderation, mediation, and moderated mediation effects in models fitted by structural equation modeling (SEM) or multiple linear regression. For SEM, both latent variable models and path models of observed variables are supported.
For more information on this package, please visit its GitHub page:
https://sfcheung.github.io/power4mome/
A get-started guide illustrates how to use this package:
https://sfcheung.github.io/power4mome/articles/power4mome.html
Code templates are also available for common mediation, moderation, and moderated mediation models:
https://sfcheung.github.io/power4mome/articles/
Major new features can be found below.
The package was developed with this philosophy:
Easy to specify the population models, even with latent variables.
As few manually-set numbers as possible when specifying the population models, with convenient default values.
As few restrictions as possible on the form of the models, even when the models have latent factors.
As automatic as possible in finding the sample size with the target power.
To achieve this comes with some costs, and some of the goals conflict with other goals: Being flexible usually means being less user-friendly, and being easy to specify the model usually means not supporting some models.
Therefore, we also try to
balance these goals, and
allow the functions to be used in different ways, to accommodate scenarios that prioritize these goals differently.
These are some major new features added in power4mome
0.2.0 (see the release
history for a full list of changes).
All-in-one “quick” functions for common mediation models. Instead of
specifying the model and effect size and then call
power4test(), these quick functions combine all the steps
into one. Learn more about them in these
examples.
Support for multigroup models has been substantially improved. For example, power analysis for detecting the group difference in an indirect path can be conducted. Several test functions have also been updated to support testing group differences. See this article for an illustration. More demonstrations will be added.
Real studies rarely involve continuous variables. Datasets with
missing data is also common. The argument process_data has
been added to power4test() and friends for generating more
realistic data for power analysis. For example, missing data can be
simulated, and ordinal variables can be created from the continuous
variables to simulate binary or Likert scale items. Power analysis can
then be conducted by taking missing data into account. See these articles
for illustrations.
The stable version at CRAN can be installed by
install.packages():
install.packages("power4mome")The latest developmental version of this package can be installed by
remotes::install_github:
remotes::install_github("sfcheung/power4mome")Some of us the developers have developed the package manymome
(Cheung & Cheung, 2024) for computing and testing effects in models
with mediation, mediation, or moderated mediation. The tests are usually
done by simulation-based methods such as Monte Carlo or bootstrap
confidence intervals, due to the complicated sampling distributions of
the effects. Therefore, there are no simple ways to determine the power
of the test analytically and accurately. The computation becomes more
complicated when latent variables are involved, necessitating a
simulation-based method to estimate the sample size.
There are already many excellent packages out there for estimating power in structural equation modeling in general, and some are also specifically for mediation or moderated mediation. We are not intended to replace with them or reinvent the wheel. We just want to have a tool that meet our own needs:
It leverages on the flexibility of manymome in
testing an indirect effect or conditional effect with little limitations
on the model.
It allows users (us and our collaborators) to specify the population model as easy (quickly) as typical power analysis programs.
We ourselves know how to do the power estimation on our own by simulation, if necessary. However, time is usually a concern, and we would like to have a tool that, though specifically designed with mediation, moderation, and moderated mediation in mind and may be limited in scope (though it is a “big” scope), is easy for our daily use in estimating power.
So here it is, power4mome, developed with we ourselves
as the users, but we believe are also useful for others who need to do
power analysis for mediation, moderation, and moderated mediation.
But power4mome is not just for mediation, moderation,
and moderated mediation. We avoided writing the functions just for these
effects, and have left room for testing other effects, as hinted in some
examples in the help pages. They may be introduced later. For now,
supporting effects that can be tested by manymome is our
priority.
If you have any suggestions and found any bugs, please feel free to open a GitHub issue:
https://github.com/sfcheung/power4mome/issues
Thanks.
Cheung, S. F., & Cheung, S.-H. (2024). manymome: An R package for computing the indirect effects, conditional effects, and conditional indirect effects, standardized or unstandardized, and their bootstrap confidence intervals, in many (though not all) models. Behavior Research Methods, 56(5), 4862–4882. https://doi.org/10.3758/s13428-023-02224-z