This vignettes demonstrates the mediation()-function. Before we start, we fit some models, including a mediation-object from the mediation-package and a structural equation modelling approach with the lavaan-package, both of which we use for comparison with brms and rstanarm.
library(bayestestR)
library(mediation)
library(brms)
library(rstanarm)
# load sample data
data(jobs)
set.seed(123)
# linear models, for mediation analysis
b1 <- lm(job_seek ~ treat + econ_hard + sex + age, data = jobs)
b2 <- lm(depress2 ~ treat + job_seek + econ_hard + sex + age, data = jobs)
# mediation analysis, for comparison with brms
m1 <- mediate(b1, b2, sims = 1000, treat = "treat", mediator = "job_seek")# Fit Bayesian mediation model in brms
f1 <- bf(job_seek ~ treat + econ_hard + sex + age)
f2 <- bf(depress2 ~ treat + job_seek + econ_hard + sex + age)
m2 <- brm(f1 + f2 + set_rescor(FALSE), data = jobs, cores = 4)# Fit Bayesian mediation model in rstanarm
m3 <- stan_mvmer(
list(job_seek ~ treat + econ_hard + sex + age + (1 | occp),
depress2 ~ treat + job_seek + econ_hard + sex + age + (1 | occp)),
data = jobs,
cores = 4,
refresh = 0
)mediation() is a summary function, especially for mediation analysis, i.e. for multivariate response models with casual mediation effects.
In the models m2 and m3, treat is the treatment effect and job_seek is the mediator effect. For the brms model (m2), f1 describes the mediator model and f2 describes the outcome model. This is similar for the rstanarm model.
mediation() returns a data frame with information on the direct effect (median value of posterior samples from treatment of the outcome model), mediator effect (median value of posterior samples from mediator of the outcome model), indirect effect (median value of the multiplication of the posterior samples from mediator of the outcome model and the posterior samples from treatment of the mediation model) and the total effect (median value of sums of posterior samples used for the direct and indirect effect). The proportion mediated is the indirect effect divided by the total effect.
The simplest call just needs the model-object.
# for brms
mediation(m2)
#> # Causal Mediation Analysis for Stan Model
#>
#> Treatment: treat
#> Mediator : job_seek
#> Response : depress2
#>
#> Effect | Estimate | 95% ETI
#> ----------------------------------------------------
#> Direct Effect (ADE) | -0.040 | [-0.124, 0.046]
#> Indirect Effect (ACME) | -0.015 | [-0.041, 0.008]
#> Mediator Effect | -0.240 | [-0.294, -0.185]
#> Total Effect | -0.055 | [-0.145, 0.034]
#>
#> Proportion mediated: 28.14% [-181.46%, 237.75%]
# for rstanarm
mediation(m3)
#> # Causal Mediation Analysis for Stan Model
#>
#> Treatment: treat
#> Mediator : job_seek
#> Response : depress2
#>
#> Effect | Estimate | 95% ETI
#> ----------------------------------------------------
#> Direct Effect (ADE) | -0.040 | [-0.129, 0.048]
#> Indirect Effect (ACME) | -0.018 | [-0.042, 0.006]
#> Mediator Effect | -0.241 | [-0.296, -0.187]
#> Total Effect | -0.057 | [-0.151, 0.033]
#>
#> Proportion mediated: 30.59% [-221.09%, 282.26%]Typically, mediation() finds the treatment and mediator variables automatically. If this does not work, use the treatment and mediator arguments to specify the related variable names. For all values, the 89% credible intervals are calculated by default. Use ci to calculate a different interval.
Here is a comparison with the mediation package. Note that the summary()-output of the mediation package shows the indirect effect first, followed by the direct effect.
summary(m1)
#>
#> Causal Mediation Analysis
#>
#> Quasi-Bayesian Confidence Intervals
#>
#> Estimate 95% CI Lower 95% CI Upper p-value
#> ACME -0.0157 -0.0387 0.01 0.19
#> ADE -0.0438 -0.1315 0.04 0.35
#> Total Effect -0.0595 -0.1530 0.02 0.21
#> Prop. Mediated 0.2137 -2.0277 2.70 0.32
#>
#> Sample Size Used: 899
#>
#>
#> Simulations: 1000
mediation(m2, ci = .95)
#> # Causal Mediation Analysis for Stan Model
#>
#> Treatment: treat
#> Mediator : job_seek
#> Response : depress2
#>
#> Effect | Estimate | 95% ETI
#> ----------------------------------------------------
#> Direct Effect (ADE) | -0.040 | [-0.124, 0.046]
#> Indirect Effect (ACME) | -0.015 | [-0.041, 0.008]
#> Mediator Effect | -0.240 | [-0.294, -0.185]
#> Total Effect | -0.055 | [-0.145, 0.034]
#>
#> Proportion mediated: 28.14% [-181.46%, 237.75%]
mediation(m3, ci = .95)
#> # Causal Mediation Analysis for Stan Model
#>
#> Treatment: treat
#> Mediator : job_seek
#> Response : depress2
#>
#> Effect | Estimate | 95% ETI
#> ----------------------------------------------------
#> Direct Effect (ADE) | -0.040 | [-0.129, 0.048]
#> Indirect Effect (ACME) | -0.018 | [-0.042, 0.006]
#> Mediator Effect | -0.241 | [-0.296, -0.187]
#> Total Effect | -0.057 | [-0.151, 0.033]
#>
#> Proportion mediated: 30.59% [-221.09%, 282.26%]If you want to calculate mean instead of median values from the posterior samples, use the centrality-argument. Furthermore, there is a print()-method, which allows to print more digits.
m <- mediation(m2, centrality = "mean", ci = .95)
print(m, digits = 4)
#> # Causal Mediation Analysis for Stan Model
#>
#> Treatment: treat
#> Mediator : job_seek
#> Response : depress2
#>
#> Effect | Estimate | 95% ETI
#> ------------------------------------------------------
#> Direct Effect (ADE) | -0.0395 | [-0.1237, 0.0456]
#> Indirect Effect (ACME) | -0.0158 | [-0.0405, 0.0083]
#> Mediator Effect | -0.2401 | [-0.2944, -0.1846]
#> Total Effect | -0.0553 | [-0.1454, 0.0341]
#>
#> Proportion mediated: 28.60% [-181.01%, 238.20%]As you can see, the results are similar to what the mediation package produces for non-Bayesian models.
Finally, we also compare the results to a SEM model, using lavaan. This example should demonstrate how to “translate” the same model in different packages or modeling approached.
library(lavaan)
data(jobs)
set.seed(1234)
model <- ' # direct effects
depress2 ~ c1*treat + c2*econ_hard + c3*sex + c4*age + b*job_seek
# mediation
job_seek ~ a1*treat + a2*econ_hard + a3*sex + a4*age
# indirect effects (a*b)
indirect_treat := a1*b
indirect_econ_hard := a2*b
indirect_sex := a3*b
indirect_age := a4*b
# total effects
total_treat := c1 + (a1*b)
total_econ_hard := c2 + (a2*b)
total_sex := c3 + (a3*b)
total_age := c4 + (a4*b)
'
m4 <- sem(model, data = jobs)
summary(m4)
#> lavaan 0.6-9 ended normally after 25 iterations
#>
#> Estimator ML
#> Optimization method NLMINB
#> Number of model parameters 11
#>
#> Number of observations 899
#>
#> Model Test User Model:
#>
#> Test statistic 0.000
#> Degrees of freedom 0
#>
#> Parameter Estimates:
#>
#> Standard errors Standard
#> Information Expected
#> Information saturated (h1) model Structured
#>
#> Regressions:
#> Estimate Std.Err z-value P(>|z|)
#> depress2 ~
#> treat (c1) -0.040 0.043 -0.929 0.353
#> econ_hard (c2) 0.149 0.021 7.156 0.000
#> sex (c3) 0.107 0.041 2.604 0.009
#> age (c4) 0.001 0.002 0.332 0.740
#> job_seek (b) -0.240 0.028 -8.524 0.000
#> job_seek ~
#> treat (a1) 0.066 0.051 1.278 0.201
#> econ_hard (a2) 0.053 0.025 2.167 0.030
#> sex (a3) -0.008 0.049 -0.157 0.875
#> age (a4) 0.005 0.002 1.983 0.047
#>
#> Variances:
#> Estimate Std.Err z-value P(>|z|)
#> .depress2 0.373 0.018 21.201 0.000
#> .job_seek 0.524 0.025 21.201 0.000
#>
#> Defined Parameters:
#> Estimate Std.Err z-value P(>|z|)
#> indirect_treat -0.016 0.012 -1.264 0.206
#> indirct_cn_hrd -0.013 0.006 -2.100 0.036
#> indirect_sex 0.002 0.012 0.157 0.875
#> indirect_age -0.001 0.001 -1.932 0.053
#> total_treat -0.056 0.045 -1.244 0.214
#> total_econ_hrd 0.136 0.022 6.309 0.000
#> total_sex 0.109 0.043 2.548 0.011
#> total_age -0.000 0.002 -0.223 0.824
# just to have the numbers right at hand and you don't need to scroll up
mediation(m2, ci = .95)
#> # Causal Mediation Analysis for Stan Model
#>
#> Treatment: treat
#> Mediator : job_seek
#> Response : depress2
#>
#> Effect | Estimate | 95% ETI
#> ----------------------------------------------------
#> Direct Effect (ADE) | -0.040 | [-0.124, 0.046]
#> Indirect Effect (ACME) | -0.015 | [-0.041, 0.008]
#> Mediator Effect | -0.240 | [-0.294, -0.185]
#> Total Effect | -0.055 | [-0.145, 0.034]
#>
#> Proportion mediated: 28.14% [-181.46%, 237.75%]The summary output from lavaan is longer, but we can find the related numbers quite easily:
treat (c1), which is -0.040indirect_treat, which is -0.016job_seek (b), which is -0.240total_treat, which is -0.056