Simple mediation example in ctsem - frequentist and hierarchical Bayes approaches.

mediation.R

install.packages('ctsem')
library(ctsem)

#generate data (columns of drift matrix 'cause' rows)
genmodel <- ctModel(n.manifest=3,n.latent=3,LAMBDA=diag(3),Tpoints=10,
  DRIFT=matrix(c(
    -.3, .1,0,
    0,-.3,.1,
    0,0,-.3), nrow=3, ncol=3),
  DIFFUSION=diag(1,3))

dat <- ctGenerate(ctmodelobj=genmodel,n.subjects=50,burnin=50,wide=FALSE)

#specify full and mediated models (real data probably needs between person variance specification, eg:
#MANIFESTTRAITVAR='auto', or use Bayesian random effects approach below...

fullmodel <- ctModel(n.manifest=3,n.latent=3,LAMBDA=diag(3),Tpoints=10,
  DRIFT=matrix(c(
    'a11', 'a12','a13',
    0,'a22','a23',
    0,0,'a33'), nrow=3, ncol=3))

mediatedmodel <- ctModel(n.manifest=3,n.latent=3,LAMBDA=diag(3),Tpoints=10,
  DRIFT=matrix(c(
    'a11', 'a12',0,
    0,'a22','a23',
    0,0,'a33'), nrow=3, ncol=3))


#fit
fullfit <- ctFit(dat=dat,ctmodelobj=fullmodel,dataform='long')

mediatedfit <- ctFit(dat=dat,ctmodelobj=mediatedmodel,dataform='long')


#compare model fit
mxCompare(fullfit$mxobj,mediatedfit$mxobj)

#summarise
summary(mediatedfit)

plot(mediatedfit)


#use Bayesian approach with random effects instead (slower!)
bayesmediatedmodel<-ctStanModel(ctmodelobj=mediatedmodel)
bayesmediatedfit <- ctStanFit(datalong=dat,ctstanmodel=bayesmediatedmodel,iter=200,chains=3,cores=3)

plot(bayesmediatedfit)
summary(bayesmediatedfit)