(Version 0.2.4.1, updated on 2024-10-22, release history)
Functions for estimating indirect effects, conditional indirect effects, and conditional effects in a model with moderation, mediation, and/or moderated mediation fitted by structural equation modelling (SEM) or estimated by multiple regression. The package was introduced in:
- Cheung, S. F., & Cheung, S.-H. (2023). 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. https://doi.org/10.3758/s13428-023-02224-z
What Can It Do?
Compute an unstandardized or standardized indirect effect or conditional indirect effect in a path model.
Form the confidence interval for this effect. Nonparametric bootstrapping is fully supported, while Monte Carlo is supported for models fitted by
lavaan::sem().Multigroup models fitted by
lavaan::sem()are also supported in 0.1.14.2 and later versions. Details can be found in this article.
Advantages
-
A Simpler Workflow
No need to define any parameters or similar code when fitting a model in
lavaan::sem(). Just focus on fitting the model first. After a model has been selected, users can compute the effect for nearly any path, from nearly any variable, to nearly any other variables, conditional on nearly any moderators, and at any levels of the moderators. (Seevignette("manymome")for details.) This is particularly convenient for multigroup models fitted bylavaan::sem(), which are supported in 0.1.14.2 and later versions (see this guide, for an illustration). -
Supports Both SEM-Based and Regression-Based Analysis
Supports structural equation models fitted by
lavaan::sem()or by path models fitted by regression usinglm(), although the focus of this package is on structural equation models. The interface of the main functions are nearly the same for both approaches. -
Flexible in the Form of Models
No limit on the number of predictors, mediators, and outcome variables, other than those by
lavaan::sem()andlm(). For multigroup models fitted bylavaan::sem(), there is no inherent limit on the number of groups, other than the limit due to `lavaan::sem(), if any (supported in 0.1.14.2 and later versions). -
Supports Standardized Effects
Can estimate standardized indirect effects and standardized conditional indirect effects without the need to standardize the variables. The bootstrap and Monte Carlo confidence intervals for standardized effects correctly take into account the sampling variation of the standardizers (the standard deviations of the predictor and the outcome variable) by recomputing them in each bootstrap sample or replication.
-
Supports Missing Data
Supports datasets with missing data through
lavaan::sem()with full information maximum likelihood (fiml).In version 0.1.9.8 or later, it also supports missing data handled by multiple imputation if the models are fitted by
semTools::sem.mi()orsemTools::runMI()(seevignette("do_mc_lavaan_mi")). -
Supports Numeric and Categorical Moderators
Supports numeric and categorical moderators. It has a function (
factor2var()) for the easy creation of dummy variables inlavaan::sem(), and can also capitalize on the native support of categorical moderators inlm(). -
Less Time for Bootstrapping
Bootstrapping, which can be time consuming, can be conducted just once. The main functions for computing indirect effects and conditional indirect effects can be called as many times as needed without redoing bootstrapping because they can reuse pregenerated bootstrap estimates (see
vignette("manymome")andvignette("do_boot")). -
Supports Latent Variables Mediation
Supports indirect effects among latent variables for models fitted by
lavaan::sem()(seevignette("med_lav")). -
Support Treating Group As a Moderator
For multigroup models fitted by
lavaan::sem(), it supports comparing the direct or indirect effects along any path between any two groups. That is, it uses the grouping variable as a moderator (illustrated here; supported in 0.1.14.2 and later versions).
Limitations
Despite the aforementioned advantages, the current version of manymome has the following limitations:
Does not (officially) support categorical
x-variables with more than two categories. Note that most functions will work with dichotomousx-variables, and the effect ofxis simply the difference in predictedybetween the two levels if coded as 0 and 1 (or any coding with a difference of 1, e.g., 1 and 2).Does not support multilevel models (although
lavaandoes).For bootstrapping, only supports nonparametric bootstrapping, and supports only percentile and bias-corrected confidence interval. Does not support other bootstrapping methods such parametric bootstrapping.
Only supports OLS estimation when
lm()is used.
We would add more to this list (suggestions are welcomed by adding GitHub issues) so that users (and we) know when other tools should be used instead of manymome, or whether we can address these limitations in manymome in the future.
How To Use It?
A good starting point is the Get-Started article (vignette("manymome")).
There are also articles (vignettes) on special topics, such as how to use mod_levels() to set the levels of the moderators. More will be added.
Installation
The stable version at CRAN can be installed by install.packages():
install.packages("manymome")The latest developmental-but-stable version at GitHub can be installed by remotes::install_github():
remotes::install_github("sfcheung/manymome")Background
We developed the package stdmod in 2021 for moderated regression. We included a function (stdmod::stdmod_lavaan()) for standardized moderation effect in path models fitted by lavaan::sem(). However, in practice, path models nearly always included indirect effects and so moderated mediation is common in path models. Moreover, stdmod is intended for moderated regression, not for structural equation modeling. We thought perhaps we could develop a more general tool for models fitted by structural equation modelling based on the interface we used in stdmod::stdmod_lavaan(). In our own projects, we also need to estimate indirect effects in models frequently. Large sample sizes with missing data are also common to us, for which bootstrapping is slow even with parallel processing. Therefore, we developed manymome to address these needs.