BANOVA.multi.mediation {BANOVA}

Mediation analysis with multiple possibly correlated mediators

Description

BANOVA.multi.mediation is a function for analysis of multiple possibly correlated mediators. These mediators are assumed to have no causal influence on each other. Both single-level and multi-level models can be analyzed.

Usage

BANOVA.multi.mediation(sol_1, sol_2, xvar, mediators, individual = FALSE)

Arguments

sol_1

an object of class "BANOVA" returned by BANOVA.run function with a fitted model for an outcome variable regressed on a causal variable, a mediator, and, possibly, moderators and control variables. The outcome variable can follow Normal, T, Poisson, Bernoulli, Binomial, Truncated Normal and ordered Multinomial distributions.

sol_2

an object of class "BANOVA" returned by BANOVA.run function, which contains an outcome of the analysis for multiple Multivariate Normal mediators regressed on a casual variable and other possible moderators and control variables.

xvar

a character string that specifies the name of the causal variable used in both models.

mediators

a vector with character strings, which specifies the names of the mediator variables used in the models.

individual

logical indicator of whether to output effects for individual units in the analysis (TRUE or FALSE). This analysis requires a multilevel sol_1.

Details

The function extends BANOVA.mediation to the case with multiple possibly correlated mediators. For details about mediation analysis performed in BANOVA see the help page for the BANOVA.mediation.

BANOVA.multi.mediation estimates and tests specific indirect effects of the causal variable conveyed through each mediator. Furthermore, the total indirect effect of the causal variables are computed as a sum of the specific indirect effects.

The function prints multiple tables with mediated effects. Tables with direct effects of the causal variable and mediators on the outcome variable, as well as direct effects of the causal variable on the mediators include a posterior mean and 95% credible intervals of the effects. Next, the function displays on the console tables with specific indirect effects and effect sizes of the mediators, followed by the TIE of the causal variable. These tables include the mean, 95% credible intervals, and two-sided Bayesian p-values.

Value

Returns an object of class "BANOVA.multi.mediation". The returned object is a list containing:

dir_effects

table or tables with the direct effect.

individual_direct

is returned if individual is set to TRUE and the causal variable is a within-subject variable. Contains a table or tables of the direct effect at the individual levels of the analysis

m1_effects

a list with tables of the effects of the mediator on the outcome

m2_effects

a list with tables of the effect of the causal variable on the mediator

indir_effects

tables of the indirect effect

individual_indirect

is returned if individual is set to TRUE and the mediator is a within-subject variable. Contains the table or tables with the indirect effect

effect_sizes

a list with effect sizes on individual mediators

total_indir_effects

table or tables with the total indirect effect of the causal variable

xvar

the name of the causal variable

mediators

the names of the mediating variables

individual

the value of the argument individual (TRUE or FALSE)

Author(s)

Anna Kopyakova

Examples

# Use the colorad data set
data(colorad)
# Add a second mediator to the data set
colorad$blur_squared <- (colorad$blur)^2
# Prepare mediators to be analyzed in the Multivariate Normal model
mediators <- cbind(colorad$blur, colorad$blur_squared)
colnames(mediators) <- c("blur", "blur_squared")
colorad$mediators <- mediators

# Build and analyze the model for the outcome variable
model <- BANOVA.model('Binomial')
banova_binom_model <- BANOVA.build(model)
res_1 <- BANOVA.run(y ~ typic, ~ color + blur + blur_squared, fit = banova_binom_model,
                    data = colorad, id = 'id', num_trials = as.integer(16), 
                    iter = 2000, thin = 1, chains = 2)
# Build and analyze the model for the mediators
model <- BANOVA.model('multiNormal')
banova_multi_norm_model <- BANOVA.build(model)
res_2 <- BANOVA.run(mediators ~ typic, ~ color, fit = banova_multi_norm_model,
                    data = colorad, id = 'id', iter = 2000, thin = 1, chains = 2)
                    
# Calculate (moderated) effects of "typic" mediated by "blur" and "blur_squared"
results <- BANOVA.multi.mediation(res_1, res_2, xvar='typic', mediators=c("blur", "blur_squared"))