Package 'rmedsem' reference manual

Title:	Statistical Mediation Analysis for SEMs
Description:	Conducts mediation analysis for structural equation models (SEM) estimated with 'lavaan', 'blavaan', 'cSEM', 'plssem', or 'modsem'. Implements the Baron and Kenny (1986) <doi:10.1037/0022-3514.51.6.1173> and Zhao, Lynch & Chen (2010) <doi:10.1086/651257> approaches to determine the presence and type of mediation. Supports covariance-based SEM, partial least squares SEM, Bayesian SEM, and moderated mediation and mediated moderation models. Tests indirect effects with the Sobel, Delta, Monte-Carlo, and bootstrap methods or, for Bayesian models, with posterior summaries and equal-tailed or highest density credible intervals. Reports the effect size measures RIT, RID, and Upsilon of Lachowicz, Preacher and Kelley (2018) <doi:10.1037/met0000165>. Results can be summarized, extracted with standard methods such as summary(), coef() and confint(), and plotted.
Authors:	Mehmet Mehmetoglu [aut] (ORCID: <https://orcid.org/0000-0002-6092-8551>), Matthias Mittner [aut, cre] (ORCID: <https://orcid.org/0000-0003-0205-7353>), Kjell Slupphaug [aut] (ORCID: <https://orcid.org/0009-0005-8324-2834>)
Maintainer:	Matthias Mittner <[email protected]>
License:	MIT + file LICENSE
Version:	1.1.1
Built:	2026-10-17 07:15:56 UTC
Source:	https://github.com/ihrke/rmedsem

Effect Sizes for Mediation Analysis

Description

Extract the effect sizes computed by rmedsem() (argument effect.size).

Usage

RIT(res, ...)

RID(res, ...)

## Default S3 method:
RIT(res, ...)

## S3 method for class 'rmedsem'
RIT(res, ...)

## Default S3 method:
RID(res, ...)

## S3 method for class 'rmedsem'
RID(res, ...)

Upsilon(res, ...)

## Default S3 method:
Upsilon(res, ...)

## S3 method for class 'rmedsem'
Upsilon(res, adjusted = TRUE, ...)
RIT(res, ...)

RID(res, ...)

## Default S3 method:
RIT(res, ...)

## S3 method for class 'rmedsem'
RIT(res, ...)

## Default S3 method:
RID(res, ...)

## S3 method for class 'rmedsem'
RID(res, ...)

Upsilon(res, ...)

## Default S3 method:
Upsilon(res, ...)

## S3 method for class 'rmedsem'
Upsilon(res, adjusted = TRUE, ...)

Arguments

res

an rmedsem object

...

additional arguments (currently unused)

adjusted

logical; if TRUE (default), return the bias-adjusted estimator of Upsilon; if FALSE, the unadjusted estimator

Details

RIT(): Ratio of the indirect to the total effect, |indirect| / |total|, i.e., the proportion of the total effect that is mediated. Following Kenny (see https://davidakenny.net/cm/mediate.htm), it should only be interpreted if the total effect is not too small (|total| >= 0.2 for standardized coefficients).
RID(): Ratio of the indirect to the direct effect, |indirect| / |direct|.
Upsilon(): The Upsilon effect size (Lachowicz, Preacher & Kelley, 2018), an R-squared-type measure of the variance in Y explained indirectly by X through M, computed from standardized coefficients.

RIT() and RID() give a warning (and the printed output of rmedsem() does not report them) if they should not be interpreted: RIT if the total effect is small (|total| < 0.2), RID if the direct effect is not significant (p-value not below p.threshold), as the ratio is then unstable. Both also warn if the indirect effect is larger than the total effect.

Value

A numeric scalar.

References

Lachowicz, M. J., Preacher, K. J., & Kelley, K. (2018). A novel measure of effect size for mediation analysis. Psychological Methods, 23(2), 244–261. doi:10.1037/met0000165

Examples

mod.txt <- "
read ~ math
science ~ read + math
"
mod <- lavaan::sem(mod.txt, data = rmedsem::hsbdemo)
out <- rmedsem(mod, indep = "math", med = "read", dep = "science")
RIT(out)
RID(out)
Upsilon(out)
Upsilon(out, adjusted = FALSE)

mod.txt <- "
read ~ math
science ~ read + math
"
mod <- lavaan::sem(mod.txt, data = rmedsem::hsbdemo)
out <- rmedsem(mod, indep = "math", med = "read", dep = "science")
RIT(out)
RID(out)
Upsilon(out)
Upsilon(out, adjusted = FALSE)

High School and Beyond Demo Dataset

Description

Demographic information and standardized test scores of 200 students from the High School and Beyond survey, as distributed by the UCLA Statistical Methods and Data Analytics group.

Usage

hsbdemo
hsbdemo

Format

`hsbdemo`

A data frame with 200 rows and 13 columns:

id: Student ID
female: Gender, "female" or "male"
ses: Socio-economic status, "low", "middle" or "high"
schtyp: School type, "public" or "private"
prog: Type of program, "general", "academic" or "vocation"
read: Reading score
write: Writing score
math: Math score
science: Science score
socst: Social studies score
honors: Enrollment in honors program, "enrolled" or "not enrolled"
awards: Number of awards
cid: Class ID

Source

UCLA Statistical Methods and Data Analytics, https://stats.oarc.ucla.edu/stat/data/hsbdemo.dta

Examples

str(hsbdemo)

mod <- lavaan::sem("read ~ math\nscience ~ read + math", data = hsbdemo)
rmedsem(mod, indep = "math", med = "read", dep = "science")
str(hsbdemo)

mod <- lavaan::sem("read ~ math\nscience ~ read + math", data = hsbdemo)
rmedsem(mod, indep = "math", med = "read", dep = "science")

Mate-Choice Survey Data from Trondheim

Description

Data from a survey on mate-choice in Trondheim, collected in 2021 using a convenience sampling procedure. Participants rated how well physical features (smv_attr_face, smv_attr_body, smv_sexy) and personality features (smv_kind to smv_sociable) described them as romantic partners, answered items on self-esteem (⁠ses_*⁠) and reported how often they had experienced feelings related to mental well-being in the past two weeks (⁠mwb_*⁠). All items were measured on a scale from 1 to 5.

Usage

mchoice
mchoice

Format

`mchoice`

A data frame with 1067 rows and 15 columns:

smv_attr_face: How well does this describe you as a partner? (attractive face), [1] very bad - [5] very well
smv_attr_body: How well does this describe you as a partner? (attractive body), [1] very bad - [5] very well
smv_sexy: How well does this describe you as a partner? (sexy), [1] very bad - [5] very well
ses_satis: On the whole, I am satisfied with myself, [1] totally disagree - [5] totally agree
ses_qualities: I feel that I have a number of good qualities, [1] totally disagree - [5] totally agree
ses_able_todo: I am able to do things as well as most other people, [1] totally disagree - [5] totally agree
mwb_optimistic: I have been feeling optimistic about the future, [1] never - [5] always
mwb_useful: I have been feeling useful, [1] never - [5] always
mwb_energy: I have had energy to spare, [1] never - [5] always
smv_kind: How well does this describe you as a partner? (kind), [1] very bad - [5] very well
smv_caring: How well does this describe you as a partner? (caring), [1] very bad - [5] very well
smv_understanding: How well does this describe you as a partner? (understanding), [1] very bad - [5] very well
smv_make_laughh: How well does this describe you as a partner? (make people laugh), [1] very bad - [5] very well
smv_funny: How well does this describe you as a partner? (funny), [1] very bad - [5] very well
smv_sociable: How well does this describe you as a partner? (sociable), [1] very bad - [5] very well

Source

Survey collected in Norway in 2021 using a convenience sampling procedure.

Examples

str(mchoice)

mod.txt <- "
  OwnLook  =~ smv_attr_face + smv_attr_body + smv_sexy
  SelfEst  =~ ses_satis + ses_qualities + ses_able_todo
  MentWell =~ mwb_optimistic + mwb_useful + mwb_energy
  SelfEst  ~ OwnLook
  MentWell ~ OwnLook + SelfEst
"
mod <- lavaan::sem(mod.txt, data = mchoice)
rmedsem(mod, indep = "OwnLook", med = "SelfEst", dep = "MentWell")
str(mchoice)

mod.txt <- "
  OwnLook  =~ smv_attr_face + smv_attr_body + smv_sexy
  SelfEst  =~ ses_satis + ses_qualities + ses_able_todo
  MentWell =~ mwb_optimistic + mwb_useful + mwb_energy
  SelfEst  ~ OwnLook
  MentWell ~ OwnLook + SelfEst
"
mod <- lavaan::sem(mod.txt, data = mchoice)
rmedsem(mod, indep = "OwnLook", med = "SelfEst", dep = "MentWell")

Plot rmedsem Results

Description

Visualize the results of rmedsem().

Usage

## S3 method for class 'rmedsem'
plot(x, type = c("coef", "effect"), ...)

plot_effect(res, description = TRUE)

plot_coef(res)
## S3 method for class 'rmedsem'
plot(x, type = c("coef", "effect"), ...)

plot_effect(res, description = TRUE)

plot_coef(res)

Arguments

x, res

an rmedsem object

type

character; "coef" (default) for a coefficient plot or "effect" for an effect size plot

...

additional arguments passed to plot_coef() or plot_effect()

description

logical, whether to add a caption describing the proportion of the total effect that is mediated (default TRUE)

Details

plot_coef(): A coefficient plot of the indirect effect (for each estimation method), the direct effect and the total effect, with their confidence (or credible) intervals.
plot_effect(): A pie chart of the (absolute) indirect and direct effects, i.e., the proportion of the total effect that is mediated. Requires the effect sizes "RIT" and "RID".

plot() calls plot_coef() (type = "coef") or plot_effect() (type = "effect").

Value

a ggplot object

Examples

mod.txt <- "
read ~ math
science ~ read + math
"
mod <- lavaan::sem(mod.txt, data = rmedsem::hsbdemo)
out <- rmedsem(mod, indep = "math", med = "read", dep = "science")
plot(out)
plot(out, type = "effect")
plot_effect(out, description = FALSE)

mod.txt <- "
read ~ math
science ~ read + math
"
mod <- lavaan::sem(mod.txt, data = rmedsem::hsbdemo)
out <- rmedsem(mod, indep = "math", med = "read", dep = "science")
plot(out)
plot(out, type = "effect")
plot_effect(out, description = FALSE)

Mediation Analysis for Structural Equation Models

Description

Tests the indirect effect of an independent variable X on a dependent variable Y through a mediator M in a fitted structural equation model (SEM), and determines the type of mediation using the Baron and Kenny (1986) and/or Zhao, Lynch & Chen (2010) approaches. Models estimated with lavaan (covariance-based SEM), cSEM and plssem (PLS-SEM), blavaan (Bayesian SEM) and modsem (models with latent interactions) are supported. The model must contain the regression paths X -> M, M -> Y and X -> Y.

Usage

rmedsem(
  mod,
  indep,
  med,
  dep,
  approach = c("bk", "zlc"),
  p.threshold = 0.05,
  effect.size = c("RIT", "RID", "upsilon"),
  ...
)

## Default S3 method:
rmedsem(mod, indep, med, dep, ...)

## S3 method for class 'blavaan'
rmedsem(
  mod,
  indep,
  med,
  dep,
  approach = c("bk", "zlc"),
  p.threshold = 0.05,
  effect.size = c("RIT", "RID", "upsilon"),
  ci.two.tailed = 0.95,
  hdi = FALSE,
  ...
)

## S3 method for class 'cSEMResults'
rmedsem(
  mod,
  indep,
  med,
  dep,
  approach = c("bk", "zlc"),
  p.threshold = 0.05,
  effect.size = c("RIT", "RID", "upsilon"),
  nbootstrap = 1000,
  ci.two.tailed = 0.95,
  seed = NULL,
  ...
)

## S3 method for class 'lavaan'
rmedsem(
  mod,
  indep,
  med,
  dep,
  approach = c("bk", "zlc"),
  p.threshold = 0.05,
  effect.size = c("RIT", "RID", "upsilon"),
  standardized = TRUE,
  mcreps = 5000,
  ci.two.tailed = 0.95,
  ...
)

## S3 method for class 'modsem'
rmedsem(
  mod,
  indep,
  med,
  dep,
  approach = c("bk", "zlc"),
  p.threshold = 0.05,
  effect.size = c("RIT", "RID", "upsilon"),
  moderator = NULL,
  standardized = TRUE,
  mcreps = 5000,
  ci.two.tailed = 0.95,
  ...
)

## S3 method for class 'PlsModel'
rmedsem(
  mod,
  indep,
  med,
  dep,
  approach = c("bk", "zlc"),
  p.threshold = 0.05,
  effect.size = c("RIT", "RID", "upsilon"),
  mcreps = 5000,
  ci.two.tailed = 0.95,
  ...
)
rmedsem(
  mod,
  indep,
  med,
  dep,
  approach = c("bk", "zlc"),
  p.threshold = 0.05,
  effect.size = c("RIT", "RID", "upsilon"),
  ...
)

## Default S3 method:
rmedsem(mod, indep, med, dep, ...)

## S3 method for class 'blavaan'
rmedsem(
  mod,
  indep,
  med,
  dep,
  approach = c("bk", "zlc"),
  p.threshold = 0.05,
  effect.size = c("RIT", "RID", "upsilon"),
  ci.two.tailed = 0.95,
  hdi = FALSE,
  ...
)

## S3 method for class 'cSEMResults'
rmedsem(
  mod,
  indep,
  med,
  dep,
  approach = c("bk", "zlc"),
  p.threshold = 0.05,
  effect.size = c("RIT", "RID", "upsilon"),
  nbootstrap = 1000,
  ci.two.tailed = 0.95,
  seed = NULL,
  ...
)

## S3 method for class 'lavaan'
rmedsem(
  mod,
  indep,
  med,
  dep,
  approach = c("bk", "zlc"),
  p.threshold = 0.05,
  effect.size = c("RIT", "RID", "upsilon"),
  standardized = TRUE,
  mcreps = 5000,
  ci.two.tailed = 0.95,
  ...
)

## S3 method for class 'modsem'
rmedsem(
  mod,
  indep,
  med,
  dep,
  approach = c("bk", "zlc"),
  p.threshold = 0.05,
  effect.size = c("RIT", "RID", "upsilon"),
  moderator = NULL,
  standardized = TRUE,
  mcreps = 5000,
  ci.two.tailed = 0.95,
  ...
)

## S3 method for class 'PlsModel'
rmedsem(
  mod,
  indep,
  med,
  dep,
  approach = c("bk", "zlc"),
  p.threshold = 0.05,
  effect.size = c("RIT", "RID", "upsilon"),
  mcreps = 5000,
  ci.two.tailed = 0.95,
  ...
)

Arguments

mod

a fitted SEM: an object of class lavaan, cSEMResults, blavaan, modsem or PlsModel (plssem). blavaan models containing latent variables must be fitted with save.lvs = TRUE; PlsModel objects must be fitted with bootstrap = TRUE.

indep

a string, the name of the independent variable (X). For modsem and PlsModel models, this can be an interaction term such as "W:X".

med

a string, the name of the mediator (M)

dep

a string, the name of the dependent variable (Y)

approach

approach(es) to determine the type of mediation: "bk" (Baron and Kenny), "zlc" (Zhao, Lynch & Chen), or both (default). Ignored for blavaan models.

p.threshold

a number between 0 and 1, the p-value threshold for significance (default 0.05). A p-value equal to the threshold counts as not significant.

effect.size

character vector with the effect sizes to compute; one or more of "RIT" (ratio of indirect to total effect), "RID" (ratio of indirect to direct effect) and "upsilon" (Lachowicz et al., 2018); see effect-sizes.

...

additional arguments passed to methods (currently unused)

ci.two.tailed

a number between 0 and 1, the level of all confidence (or, for blavaan, credible) intervals (default 0.95)

hdi

(blavaan) a logical. If FALSE (default), equal-tailed credible intervals are computed; if TRUE, highest density intervals (requires the HDInterval package). Applies to the indirect, direct and total effects and to Upsilon.

nbootstrap

(cSEM) the number of bootstrap samples (default 1000)

seed

(cSEM) NULL (default) or a non-negative integer, the seed for the bootstrap. If NULL, the seed is drawn from R's random number generator, so that results can be reproduced with set.seed().

standardized

(lavaan, modsem) a logical, whether to use standardized coefficients (default TRUE). cSEM and blavaan results are always standardized.

mcreps

(lavaan, modsem, PlsModel) the number of Monte-Carlo samples, a positive integer (default 5000). For PlsModel objects, only used for MC-PLS models with delta-method standard errors (see section 'Backends').

moderator

(modsem) NULL (default) or a string, the name of the moderator W for moderated mediation. The model must contain an interaction of the moderator with indep and/or med.

Value

an object of class c("rmedsem_<pkg>", "rmedsem"), where ⁠<pkg>⁠ identifies the backend (lavaan, cSEM, blavaan, modsem or plssem). See rmedsem-methods for functions to print, summarize and extract results, effect-sizes for effect sizes and plot.rmedsem() for plots. The structure of the object is described in section 'Adding a backend'.

Backends

lavaan: The indirect effect is tested with the Sobel, Delta and Monte-Carlo methods. The Zhao, Lynch & Chen approach is based on the Monte-Carlo test.
cSEMResults: The model is re-estimated with nbootstrap bootstrap samples. The indirect effect is tested with the Sobel, Delta and bootstrap methods, and the Zhao, Lynch & Chen approach is based on the bootstrap test. Only single-group, first-order models are supported.
blavaan: Estimates are based on the (standardized) posterior samples. The output reports posterior means, standard deviations, posterior probabilities of a positive and negative indirect effect, evidence ratios and credible intervals; the Baron and Kenny and Zhao, Lynch & Chen approaches are not applied.
modsem: As for lavaan. In addition, moderated mediation (via moderator) and mediated moderation (an interaction term as indep) are supported.
PlsModel: Models estimated with plssem::pls() (PLS-SEM and consistent PLSc-SEM, including models with interaction terms and ordinal indicators). The model must be estimated with bootstrap = TRUE; the number of bootstrap samples is set in plssem::pls() (boot.R), and results are reproducible with its boot.iseed argument. The indirect effect is tested with the Sobel, Delta and bootstrap methods, where the bootstrap test uses the bootstrap samples of plssem. The Zhao, Lynch & Chen approach is based on the bootstrap test. Mediated moderation (an interaction term as indep) is supported. For MC-PLS models (e.g., interaction models with ordinal indicators) estimated with delta-method standard errors (the default mc.delta.se = TRUE in plssem::pls()), the bootstrap samples of plssem do not refer to the Monte-Carlo corrected estimates; a Monte-Carlo test based on the estimates and their variance-covariance matrix (mcreps samples) is used instead of the bootstrap test, also for the Zhao, Lynch & Chen approach. For ordinal indicators, all effects refer to the standardized latent variables.

Multi-group and multilevel models are not supported.

Adding a backend

Support for further model classes is added by writing a method ⁠rmedsem.<class>()⁠ that returns a list of class c("rmedsem_<pkg>", "rmedsem"). For the default print.rmedsem(), plot.rmedsem() and as.data.frame.rmedsem() methods to work, the list must contain the following elements:

package: name of the estimating package (character).
standardized: whether the coefficients are standardized (logical).
vars: list with elements indep, med and dep.
est.methods: character vector naming the estimation methods for the indirect effect, e.g. c("sobel", "delta", "montc").
one element per entry in est.methods: a named numeric vector with elements coef, se, zval, pval, lower and upper. The Baron and Kenny approach requires the element sobel.
zlc.method: (optional) the entry of est.methods whose p-value is used for the Zhao, Lynch & Chen approach; defaults to the last entry of est.methods.
direct.effect: named numeric vector with elements coef, se, pval, lower and upper.
total.effect: named numeric vector with elements coef, se, lower and upper.
med.approach: character vector, a subset of c("bk", "zlc").
med.data: list with elements sig_thresh (the p-value threshold), coefs and pvals; the latter two are lists with elements moi (X -> M), dom (M -> Y) and doi (X -> Y).
effect.size: list with (a subset of) elements RIT, RID and upsilon, as returned by the built-in backends.
nobs: (optional) number of observations, used by stats::nobs().
ci.level: (optional) level of the stored intervals, used by stats::confint() and summary(); defaults to 0.95.
ci.type: (optional) label of the stored intervals, "CI" (default) or "HDI".

A backend whose output does not fit this scheme can provide its own ⁠print.rmedsem_<pkg>()⁠ method, either replacing the default output (as for blavaan) or extending it via NextMethod() (as for modsem).

References

Baron, R. M., & Kenny, D. A. (1986). The moderator-mediator variable distinction in social psychological research: Conceptual, strategic, and statistical considerations. Journal of Personality and Social Psychology, 51(6), 1173–1182. doi:10.1037/0022-3514.51.6.1173

Lachowicz, M. J., Preacher, K. J., & Kelley, K. (2018). A novel measure of effect size for mediation analysis. Psychological Methods, 23(2), 244–261. doi:10.1037/met0000165

Zhao, X., Lynch, J. G., & Chen, Q. (2010). Reconsidering Baron and Kenny: Myths and truths about mediation analysis. Journal of Consumer Research, 37(2), 197–206. doi:10.1086/651257

Examples

## lavaan: observed variables
mod.txt <- "
read ~ math
science ~ read + math
"
mod <- lavaan::sem(mod.txt, data = rmedsem::hsbdemo)
out <- rmedsem(mod, indep = "math", med = "read", dep = "science")
out

# Zhao, Lynch & Chen approach only, unstandardized coefficients
rmedsem(mod, indep = "math", med = "read", dep = "science",
        approach = "zlc", standardized = FALSE, mcreps = 5000)


## cSEM
if (requireNamespace("cSEM", quietly = TRUE)) {
  model <- "
    OwnLook  =~ smv_attr_face + smv_attr_body + smv_sexy
    SelfEst  =~ ses_satis + ses_qualities + ses_able_todo
    MentWell =~ mwb_optimistic + mwb_useful + mwb_energy
    SelfEst  ~ OwnLook
    MentWell ~ OwnLook + SelfEst
  "
  mod <- cSEM::csem(rmedsem::mchoice, model)
  # small number of bootstrap samples to keep the example fast
  rmedsem(mod, indep = "OwnLook", med = "SelfEst", dep = "MentWell",
          nbootstrap = 200)
}

## modsem: mediated moderation and moderated mediation
if (requireNamespace("modsem", quietly = TRUE)) {
  m <- "
    OwnLook =~ smv_attr_face + smv_attr_body + smv_sexy
    SelfEst =~ ses_satis + ses_qualities + ses_able_todo
    MentWell =~ mwb_optimistic + mwb_useful + mwb_energy
    smv =~ smv_kind + smv_caring + smv_understanding +
      smv_make_laughh + smv_funny + smv_sociable
    SelfEst ~ OwnLook + smv + smv:OwnLook
    MentWell ~ OwnLook + SelfEst + smv + smv:OwnLook
  "
  est <- modsem::modsem(m, data = rmedsem::mchoice, method = "lms")

  # mediated moderation
  rmedsem(est, indep = "smv:OwnLook", med = "SelfEst", dep = "MentWell")

  # moderated mediation
  rmedsem(est, indep = "OwnLook", med = "SelfEst", dep = "MentWell",
          moderator = "smv")
}

## plssem (PLS-SEM)
if (requireNamespace("plssem", quietly = TRUE)) {
  model <- "
    OwnLook  =~ smv_attr_face + smv_attr_body + smv_sexy
    SelfEst  =~ ses_satis + ses_qualities + ses_able_todo
    MentWell =~ mwb_optimistic + mwb_useful + mwb_energy
    SelfEst  ~ OwnLook
    MentWell ~ OwnLook + SelfEst
  "
  # small number of bootstrap samples to keep the example fast
  fit <- plssem::pls(model, rmedsem::mchoice, bootstrap = TRUE,
                     boot.R = 200, boot.iseed = 1)
  rmedsem(fit, indep = "OwnLook", med = "SelfEst", dep = "MentWell")
}

## blavaan
if (requireNamespace("blavaan", quietly = TRUE)) {
  # blavaan's fitting functions need the package to be attached
  library(blavaan)
  # short single chain to keep the example fast; use more chains and
  # iterations in practice
  bmod <- bsem(mod.txt, data = rmedsem::hsbdemo, n.chains = 1,
               burnin = 500, sample = 500, seed = 1,
               bcontrol = list(refresh = 0))
  rmedsem(bmod, indep = "math", med = "read", dep = "science")

  # highest density intervals instead of equal-tailed intervals
  if (requireNamespace("HDInterval", quietly = TRUE))
    rmedsem(bmod, indep = "math", med = "read", dep = "science", hdi = TRUE)
}


## lavaan: observed variables
mod.txt <- "
read ~ math
science ~ read + math
"
mod <- lavaan::sem(mod.txt, data = rmedsem::hsbdemo)
out <- rmedsem(mod, indep = "math", med = "read", dep = "science")
out

# Zhao, Lynch & Chen approach only, unstandardized coefficients
rmedsem(mod, indep = "math", med = "read", dep = "science",
        approach = "zlc", standardized = FALSE, mcreps = 5000)


## cSEM
if (requireNamespace("cSEM", quietly = TRUE)) {
  model <- "
    OwnLook  =~ smv_attr_face + smv_attr_body + smv_sexy
    SelfEst  =~ ses_satis + ses_qualities + ses_able_todo
    MentWell =~ mwb_optimistic + mwb_useful + mwb_energy
    SelfEst  ~ OwnLook
    MentWell ~ OwnLook + SelfEst
  "
  mod <- cSEM::csem(rmedsem::mchoice, model)
  # small number of bootstrap samples to keep the example fast
  rmedsem(mod, indep = "OwnLook", med = "SelfEst", dep = "MentWell",
          nbootstrap = 200)
}

## modsem: mediated moderation and moderated mediation
if (requireNamespace("modsem", quietly = TRUE)) {
  m <- "
    OwnLook =~ smv_attr_face + smv_attr_body + smv_sexy
    SelfEst =~ ses_satis + ses_qualities + ses_able_todo
    MentWell =~ mwb_optimistic + mwb_useful + mwb_energy
    smv =~ smv_kind + smv_caring + smv_understanding +
      smv_make_laughh + smv_funny + smv_sociable
    SelfEst ~ OwnLook + smv + smv:OwnLook
    MentWell ~ OwnLook + SelfEst + smv + smv:OwnLook
  "
  est <- modsem::modsem(m, data = rmedsem::mchoice, method = "lms")

  # mediated moderation
  rmedsem(est, indep = "smv:OwnLook", med = "SelfEst", dep = "MentWell")

  # moderated mediation
  rmedsem(est, indep = "OwnLook", med = "SelfEst", dep = "MentWell",
          moderator = "smv")
}

## plssem (PLS-SEM)
if (requireNamespace("plssem", quietly = TRUE)) {
  model <- "
    OwnLook  =~ smv_attr_face + smv_attr_body + smv_sexy
    SelfEst  =~ ses_satis + ses_qualities + ses_able_todo
    MentWell =~ mwb_optimistic + mwb_useful + mwb_energy
    SelfEst  ~ OwnLook
    MentWell ~ OwnLook + SelfEst
  "
  # small number of bootstrap samples to keep the example fast
  fit <- plssem::pls(model, rmedsem::mchoice, bootstrap = TRUE,
                     boot.R = 200, boot.iseed = 1)
  rmedsem(fit, indep = "OwnLook", med = "SelfEst", dep = "MentWell")
}

## blavaan
if (requireNamespace("blavaan", quietly = TRUE)) {
  # blavaan's fitting functions need the package to be attached
  library(blavaan)
  # short single chain to keep the example fast; use more chains and
  # iterations in practice
  bmod <- bsem(mod.txt, data = rmedsem::hsbdemo, n.chains = 1,
               burnin = 500, sample = 500, seed = 1,
               bcontrol = list(refresh = 0))
  rmedsem(bmod, indep = "math", med = "read", dep = "science")

  # highest density intervals instead of equal-tailed intervals
  if (requireNamespace("HDInterval", quietly = TRUE))
    rmedsem(bmod, indep = "math", med = "read", dep = "science", hdi = TRUE)
}

Methods for rmedsem Objects

Description

Print, summarize and extract the results of rmedsem().

Usage

## S3 method for class 'rmedsem'
summary(object, ...)

## S3 method for class 'summary.rmedsem'
print(x, digits = 3, ...)

## S3 method for class 'rmedsem'
coef(object, method = NULL, ...)

## S3 method for class 'rmedsem'
confint(object, parm, level = NULL, method = NULL, ...)

## S3 method for class 'rmedsem'
nobs(object, ...)

## S3 method for class 'rmedsem'
as.data.frame(x, ...)

## S3 method for class 'rmedsem'
print(x, digits = 3, indent = 3, ...)

## S3 method for class 'rmedsem_blavaan'
print(x, digits = 3, indent = 3, ...)

## S3 method for class 'rmedsem_modsem'
print(x, digits = 3, indent = 3, ci_moderation = FALSE, ...)
## S3 method for class 'rmedsem'
summary(object, ...)

## S3 method for class 'summary.rmedsem'
print(x, digits = 3, ...)

## S3 method for class 'rmedsem'
coef(object, method = NULL, ...)

## S3 method for class 'rmedsem'
confint(object, parm, level = NULL, method = NULL, ...)

## S3 method for class 'rmedsem'
nobs(object, ...)

## S3 method for class 'rmedsem'
as.data.frame(x, ...)

## S3 method for class 'rmedsem'
print(x, digits = 3, indent = 3, ...)

## S3 method for class 'rmedsem_blavaan'
print(x, digits = 3, indent = 3, ...)

## S3 method for class 'rmedsem_modsem'
print(x, digits = 3, indent = 3, ci_moderation = FALSE, ...)

Arguments

object

an rmedsem object

...

additional arguments (currently unused)

x

an rmedsem object; for print.summary.rmedsem() a summary.rmedsem object

digits

an integer, the number of decimal places to print

method

estimation method for the indirect effect, one of object$est.methods (e.g., "sobel", "delta", "montc", "boot" or "bayes"); see section 'Extracting results' for the default

parm

character vector; a subset of c("indirect", "direct", "total")

level

the confidence level. The intervals are computed by rmedsem() (argument ci.two.tailed), so level can only be used to check that the stored intervals have the requested level.

indent

an integer, the number of spaces to indent

ci_moderation

a logical, whether to print confidence intervals for the moderation effects (moderated mediation with modsem only)

Value

print() returns x invisibly.

summary() returns an object of class summary.rmedsem, a list with elements

package, vars, standardized, nobs, ci.level: copied from object.
ci.type: type of the intervals: "CI" or, for blavaan models fitted with hdi = TRUE, "HDI".
p.threshold: the p-value threshold (NULL for blavaan models).
effects: a data frame with columns effect, method, estimate, se, zval, pval, lower and upper; NA where a quantity is not available. For blavaan models, pval is the posterior probability of the opposite sign.
mediation: a list with elements bk and zlc giving the type of mediation (NULL if the approach was not requested). bk is one of "none", "complete" or "partial"; zlc is one of "indirect-only", "direct-only", "no-effect", "complementary" or "competitive".
zlc.method: the estimation method used for the Zhao, Lynch & Chen approach.
effect.size: a named numeric vector with the requested effect sizes (RIT, RID, upsilon (adjusted) and upsilon.unadjusted).
effect.size.problems: a named character vector with an entry for each of RIT and RID that should not be interpreted (see effect-sizes), describing the reason; empty if there is none.

coef() returns a named numeric vector with elements indirect, direct and total.

confint() returns a matrix with one row per effect and columns giving the lower and upper limits, labelled by their probabilities (e.g., "2.5 %" and "97.5 %") or, for highest density intervals, "lower" and "upper".

nobs() returns an integer.

as.data.frame() returns a data frame with one row per estimation method of the indirect effect and columns package, method and the estimates stored for that method (coef, se, zval, pval, lower and upper; blavaan results additionally contain the posterior probabilities pvpos and pvneg and the evidence ratios ERpos and ERneg).

Printing and summarizing

print() gives a detailed, step-by-step description of the results: a table of the tests of the indirect effect (one column per estimation method), the steps and conclusions of the Baron and Kenny and/or Zhao, Lynch & Chen approaches, and the effect sizes. For blavaan models, the table reports posterior summaries instead; for moderated mediation with modsem, the moderation effects are printed in addition.

summary() collects the same results in compact form: a table of the indirect (for each estimation method), direct and total effects, the type of mediation, and the effect sizes.

Extracting results

coef() returns the estimated indirect, direct and total effects, confint() their confidence (or, for blavaan models, credible) intervals, and nobs() the number of observations. as.data.frame() returns the estimates of the indirect effect for all estimation methods.

The indirect effect is estimated with several methods (see object$est.methods), which all give the same point estimate but different standard errors and intervals. By default, coef() and confint() use the method that also underlies the Zhao, Lynch & Chen approach: "montc" (Monte-Carlo) for lavaan and modsem, "boot" (bootstrap) for cSEM and plssem, and "bayes" for blavaan.

Extending the printed output

print.rmedsem() handles all backends that provide the elements described in section 'Adding a backend' of rmedsem(). Backends that need a different output provide a method for their subclass, either replacing the default output (print.rmedsem_blavaan()) or extending it with NextMethod() (print.rmedsem_modsem()).

Examples

mod.txt <- "
read ~ math
science ~ read + math
"
mod <- lavaan::sem(mod.txt, data = rmedsem::hsbdemo)
out <- rmedsem(mod, indep = "math", med = "read", dep = "science")

# detailed output
print(out)

# compact summary and its elements
s <- summary(out)
s
s$mediation
s$effects

# extract estimates
coef(out)
confint(out)
confint(out, parm = "indirect", method = "sobel")
nobs(out)
as.data.frame(out)

mod.txt <- "
read ~ math
science ~ read + math
"
mod <- lavaan::sem(mod.txt, data = rmedsem::hsbdemo)
out <- rmedsem(mod, indep = "math", med = "read", dep = "science")

# detailed output
print(out)

# compact summary and its elements
s <- summary(out)
s
s$mediation
s$effects

# extract estimates
coef(out)
confint(out)
confint(out, parm = "indirect", method = "sobel")
nobs(out)
as.data.frame(out)

Fitness Center Survey Data from Trondheim

Description

Data from a survey among the members of a fitness center in Trondheim, Norway. Participants rated how important different reasons for working out are to them and how well two physical features describe them. All items except age are measured on a scale from 1 to 6. The data contain missing values; 187 of the 246 rows are complete.

Usage

workout
workout

Format

`workout`

A data frame with 246 rows and 12 columns:

age: Age in years
lweight: How important is the following to you to work out: to lose weight, [1] not important at all - [6] very important
calories: How important is the following to you to work out: to burn calories, [1] not important at all - [6] very important
cweight: How important is the following to you to work out: to control my weight, [1] not important at all - [6] very important
body: How important is the following to you to work out: to have a good body, [1] not important at all - [6] very important
appear: How important is the following to you to work out: to improve my appearance, [1] not important at all - [6] very important
attract: How important is the following to you to work out: to look more attractive, [1] not important at all - [6] very important
muscle: How important is the following to you to work out: to develop my muscles, [1] not important at all - [6] very important
strength: How important is the following to you to work out: to get stronger, [1] not important at all - [6] very important
endur: How important is the following to you to work out: to increase my endurance, [1] not important at all - [6] very important
face: How well does the following describe you as a person: attractive face, [1] very badly - [6] very well
sexy: How well does the following describe you as a person: sexy, [1] very badly - [6] very well

Source

Survey among the members of a fitness center in Trondheim, Norway.

Examples

str(workout)

mod.txt <- "
  Attractive =~ face + sexy
  Appearance =~ body + appear + attract
  Muscle     =~ muscle + strength + endur
  Appearance ~ Attractive + age
  Muscle     ~ Appearance + Attractive + age
"
mod <- lavaan::sem(mod.txt, data = workout)
rmedsem(mod, indep = "Attractive", med = "Appearance", dep = "Muscle")
str(workout)

mod.txt <- "
  Attractive =~ face + sexy
  Appearance =~ body + appear + attract
  Muscle     =~ muscle + strength + endur
  Appearance ~ Attractive + age
  Muscle     ~ Appearance + Attractive + age
"
mod <- lavaan::sem(mod.txt, data = workout)
rmedsem(mod, indep = "Attractive", med = "Appearance", dep = "Muscle")

Package 'rmedsem'

Help Index

Effect Sizes for Mediation Analysis

Description

Usage

Arguments

Details

Value

References

Examples

High School and Beyond Demo Dataset

Description

Usage

Format

hsbdemo

Source

Examples

Mate-Choice Survey Data from Trondheim

Description

Usage

Format

mchoice

Source

Examples

Plot rmedsem Results

Description

Usage

Arguments

Details

Value

See Also

Examples

Mediation Analysis for Structural Equation Models

Description

Usage

Arguments

Value

Backends

Adding a backend

References

Examples

Methods for rmedsem Objects

Description

Usage

Arguments

Value

Printing and summarizing

Extracting results

Extending the printed output

See Also

Examples

Fitness Center Survey Data from Trondheim

Description

Usage

Format

workout

Source

Examples

`hsbdemo`

`mchoice`

`workout`