Moderated Mediation in SEM help

[Emma Corley]

Hi all, 

I am trying to fit (what I think) is a moderated mediation model with 2 moderators to correspond with the attached diagram. However, I'm relatively new to both R and lavaan and could really do with some help trying to build my model.

I have attached the code I have generated below. I guess my main question is whether I have specified the model correctly. Any help or guidance would be very much appreciated!

I also had a question regarding using a latent variable as a moderator. I found some questions regarding indProd function (in this group) but I am a little confused of how to generate such a code. I was hoping to use the variable ELAS (which consists of 4 items with 4 levels) as a moderator variable. At the moment I've just used a total score as I having a lot of difficulty trying to find a way to use this as a latent variable. 

#moderated mediation

moderated_mediation <-'

#latent variable

IQ=~ in1 + in2 + in3

#regressions

TBV~ a1*gene

IQ~ b1*TBV 

TBV~ a2*ELAS

TBV~a3*gene_ELAS

IQ~ b2*eduageS

IQ~ b3*TBV_eduageS

IQ~ cdash*gene 

#covariates

TBV~ totalB + Sex + Age

#mean of centered moderators (for use in simple slopes)

ELAS ~ ELA.mean*1

eduageS ~eduage.mean*1

#variance of moderators 

ELAS ~~ ELA.var*ELASS

eduageS ~~ eduage.var*eduageS 

#index of moderated mediation 

imm1 := a1*b3

imm2 := b1*a3

#indirect effects conditional on moderators

indirect1.SDbelow := a1*b1* + a3*-sqrt(ELA.var)*b1

indirect1.mean := a1*b1 + a3*ELA.mean*b1

indirect1.SDabove := a1*b1 + a3*sqrt(ELA.var)*b1

indirect2.SDbelow := a1*b1 + a1*-sqrt(eduage.var)*b3

indirect2.mean := a1*b1 + a1*eduage.mean*b3

indirect2.SDabove := a1*b1 + a1*sqrt(eduage.var)*b3

'

#fit model 

sem <- sem(model= moderated_mediation, data= df, fixed.x = FALSE,  missing ='fiml')

#fit measures

summary(sem, fit.measures= TRUE, standardize= TRUE, rsquare= TRUE)

Many thanks again, 

Emma

[Terrence Jorgensen]

I guess my main question is whether I have specified the model correctly. Any help or guidance would be very much appreciated!

Looks like you have separate moderators of the "a" and "b" paths, conforming to Model 4 in Preacher et al.'s (2007) tutorial:  https://doi.org/10.1080/00273170701341316

Inspecting that path diagram will inform you about correct model specification.

 

I was hoping to use the variable ELAS (which consists of 4 items with 4 levels) as a moderator variable. At the moment I've just used a total score as I having a lot of difficulty trying to find a way to use this as a latent variable. 

If by "4 levels" you mean it is ordinal, then you can't use the product-indicator approach while treating it as ordinal.  You would have to treat it as continuous.  

The ?indProd help page has examples, as well as references you can read for more information about the methods.  Here is a book chapter that has example lavaan syntax, where one observed variable is treated as a single-indicator factor so it can interact with a multiple-indicator common factor (I believe that conforms to your situation, but this chapter is about testing measurement invariance, not moderated mediation).

https://doi.org/10.1007/978-3-319-77249-3_20

Terrence D. Jorgensen

Assistant Professor, Methods and Statistics

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

http://www.uva.nl/profile/t.d.jorgensen