Help with Complex Model containing interactions between OVs

[Cleno Couto]

Hi everyone,

I'm new to SEM and lavaan, and any help would be appreciated. I'm trying to test a complex model through lavaan only with observed variables. From what I could gather in the package website and this forum, I came up with the following model:

model1 <- ' 

          ICA ~ g*PE + c*NAS + e*LHS
          NAS ~ b*RD + f*RD.LHS
          LHS ~ d*RD
          RD ~ a*SJ

          # Indirect Effects:
          abc := a*b*c
          ade := a*d*e

          # Conditional Effect:
          cond := (a*b*c) + (a*d*e) + f
          cond_below := (a*b*c) + (a*d*e) + (-14.81*f)
          cond_upper := (a*b*c) + (a*d*e) + (14.81*f)

          '

Still, I don't know if my script is correct, especially considering the interaction RD.LHS and the conditional effect, as well as how to probe it. I am aware of semTools, but its functions only work with latent variables, so I looked up in the forum and found the suggestion of doing -1 SD / + 1 SD in the conditional effects (14.81 being the SD of the moderator).

Again, any help would be greatly appreciated.

Cleno Couto

[Terrence Jorgensen]

I'm trying to test a complex model

A path diagram would help communicate what model you are trying to fit.

          NAS ~ b*RD + f*RD.LHS

RD.LHS cannot be interpreted as an interaction unless both lower-order terms are included.  You only have RD, reflecting the effect of LHS being nested under RD.

http://quantpsy.org/interact/interactions.htm

the suggestion of doing -1 SD / + 1 SD in the conditional effects (14.81 being the SD of the moderator).

It should be the mean +/- 1 SD from the mean.  Also, you should define conditional specific indirect effects, then define conditional total indirect effects, noting that the moderator affects the size of one of the paths being multiplied (i.e., it is not merely an additive effect; see link above for moderation primer).  Suppose the mean of your moderator is 38:

## specific conditional indirect effects

abc_low  := a*(b + (38 - 14.81)*f)*c

abc_mean := a*(b +  38         *f)*c

abc_high := a*(b + (38 + 14.81)*f)*c

ade_low  := a*(d + (38 - 14.81)*f)*e

ade_mean := a*(d +  38         *f)*e

ade_high := a*(d + (38 + 14.81)*f)*e

## specific total indirect effects

ind_low  := abc_low + ade_low

ind_mean := abc_mean + ade_mean

ind_high := abc_high + ade_high

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

[Cleno Couto]

Hi Terrence,

Thanks for the help and sorry for the delay!

A path diagram would help communicate what model you are trying to fit.

I'm attaching the path diagram in this email. I'll also try the suggestions you made, thank you again.

-- Cleno Couto