Bayesian Mediation Analysis With Latent Variables

[Julian Gaviria]

Dear list members,

I wonder whether I could run a mediation Analysis including latent variables in blavaan, such as it is easily implemented with lavaan :

library(lavaan)

  model <- '

   # latent variables

     X =~ x1 + x2 + x3

     M =~ y1 + y2 + y3 + y4

     Y =~ y5 + y6 + y7 + y8

   

  fit <- sem(model,data=PoliticalDemocracy)

  summary(fit)

  

  model <- ' # direct effect

             Y ~ c*X

           # mediator

             M ~ a*X

             Y ~ b*M

           # indirect effect (a*b)

             ab := a*b

           # total effect

             total := c + (a*b)

         '

  fit <- sem(model, data = PoliticalDemocracy)

  summary(fit)

P.D. I think one option in the Bayesian context would be the R2OpenBUGS package:

https://psycnet.apa.org/record/2019-78964-001

However, it would be nice to know whether I could use blavaan for the same analysis. 

Thanks in advance and the best regards,  

[Mauricio Garnier-Villarreal]

Yes, you can directly estimate this in blavaan, and no need to use boostrap or monte-carlo method. The inference from the posterior for the indirect effects have no assumption on its distribution

Not sure why you have 2 syntaxes called model, you only need 1

library(blavaan)

  model <- '

   # latent variables

     X =~ x1 + x2 + x3

     M =~ y1 + y2 + y3 + y4

     Y =~ y5 + y6 + y7 + y8

 # direct effect

             Y ~ c*X

           # mediator

             M ~ a*X

             Y ~ b*M

           # indirect effect (a*b)

             ab := a*b

           # total effect

             total := c + (a*b)

         '

  fit <- bsem(model, data = PoliticalDemocracy, std.lv=T)

  summary(fit)

[Julian Gaviria]

Thanks Mauricio,

It nicely worked out. Indeed, I had a typo with the string : "model <- ' # direct effect".

Best regards,