multiple mediators in lavaan?

[NewInMediation]

Hi,

Can we put multiple mediators in lavaan (instead of multiple models with only one mediator)? I heard psych package can do it, but I wonder if it's possible in lavaan as well to continue to use lavaan.

Thanks in advance,

NewInMediation

[Terrence Jorgensen]

In principle, there is no limit to the number of mediators, nor to the functional form of the causal model (given logical idenfiability conditions).  In other words, you should be able to write the syntax that fits the model you want to fit.  

http://lavaan.ugent.be/tutorial/mediation.html

That example only uses the classic 3-variable scenario, but you could search this forum for "mediation" threads that have other examples.

Terrence D. Jorgensen

Postdoctoral Researcher, Methods and Statistics

Research Institute for Child Development and Education, the University of Amsterdam

UvA web page: http://www.uva.nl/profile/t.d.jorgensen

[Adam Omidpanah]

Is it a parallel or sequential mediation analysis?

[Terrence Jorgensen]

Is it a parallel or sequential mediation analysis?

Not having seen your model, I would not know.  But these are not lavaan issues.  You can post general SEM questions on SEMNET:

http://www2.gsu.edu/~mkteer/semnet.html

[NewInMediation]

Thanks everyone for your input,

But I think misunderstanding happened. My question was lavaan specific, not SEM general: how we can put multiple mediators in lavaan. 

Here is my model:

p16a18c24 <- ' # direct effect

myOutcome ~ c*myPredictor

# mediator

myMediator ~ a*myPredictor

myOutcome ~ b*myMediator

# indirect effect (a*b)

ab := a*b

# total effect

total := c + (a*b)

'

fit.p16a18c24 <- sem(p16a18c24,data=m,missing="fiml",se="bootstrap")

summary(fit.p16a18c24,fit.measures=TRUE,standardized=TRUE,rsquare=TRUE)

Am I supposed to add mediators like this?

p16a18c24 <- ' # direct effect

myOutcome ~ c*myPredictor

# mediator

myMediator1 ~ a1*myPredictor

myOutcome ~ b1*myMediator1

myMediator2 ~ a2*myPredictor

myOutcome ~ b2*myMediator2

myMediator3 ~ a3*myPredictor

myOutcome ~ b3*myMediator3

# indirect effect (a*b)

a1b1 := a1*b1

a2b2 := a2*b2

a3b3 := a3*b3

# total effect

total := c + (a1*b1+a2*b2+a3*b3)

'

fit.p16a18c24 <- sem(p16a18c24,data=m,missing="fiml",se="bootstrap")

summary(fit.p16a18c24,fit.measures=TRUE,standardized=TRUE,rsquare=TRUE)

Or is there another way?

I would really appreciate your clarification.

NewInMediation

[NewInMediation]

Thanks Yves, I really appreciate your reply.

But I don't understand what you meant by "use different labels, for different paths." Did you mean a1, a2, and a3 or names of variables (myMediator1, 2, & 3)? 

Thank you so much,

NewInMediation

On Wed, Aug 30, 2017 at 11:44 AM, Yves Rosseel wrote:

Am I supposed to add mediators like this?

p16a18c24 <- ' # direct effect
myOutcome ~ c*myPredictor
# mediator

myMediator1 ~ a*myPredictor
myOutcome ~ b*myMediator1
myMediator2 ~ a*myPredictor
myOutcome ~ b*myMediator2
myMediator3 ~ a*myPredictor
myOutcome ~ b*myMediator3

# indirect effect (a*b)
ab := a*b
# total effect
total := c + (a*b)
'

yes, BUT you may want to use different labels, for different paths:

p16a18c24 <- ' # direct effect
myOutcome ~ c*myPredictor
# mediator
myMediator1 ~ a1*myPredictor
myOutcome ~ b1*myMediator1
myMediator2 ~ a2*myPredictor
myOutcome ~ b2*myMediator2
myMediator3 ~ a3*myPredictor
myOutcome ~ b3*myMediator3

# indirect effects (a*b)
ab1 := a1*b1
ab2 := a2*b2
ab3 := a3*b3
# total effect
total := c + (a1*b1) + (a2*b2) + (a3*b3)
'

Yves.