probe3WayMC, can't get it to produce the simple intercepts

[Melvin Sim]

Hello everyone,

I fit the model below:

##Create product terms

stumain.df3$gxb1<-with(stumain.df3, STgrit1cent * STbelong1cent)
stumain.df3$gxb2<-with(stumain.df3, STgrit2cent * STbelong4cent)
stumain.df3$gxb3<-with(stumain.df3, STgrit3cent * STbelong6cent)

stumain.df3$gxs1<-with(stumain.df3, STgrit1cent * PAReduyearscent)
stumain.df3$gxs2<-with(stumain.df3, STgrit2cent * PARjobindexcent)
stumain.df3$gxs3<-with(stumain.df3, STgrit3cent * homeposcent)

stumain.df3$sxb1<-with(stumain.df3, PAReduyearscent * STbelong1cent)
stumain.df3$sxb2<-with(stumain.df3, PARjobindexcent * STbelong4cent)
stumain.df3$sxb3<-with(stumain.df3, homeposcent * STbelong6cent)

stumain.df3$gxbxs1<-with(stumain.df3, STgrit1cent * STbelong1cent * PAReduyearscent)
stumain.df3$gxbxs2<-with(stumain.df3, STgrit2cent * STbelong4cent * PARjobindexcent)
stumain.df3$gxbxs3<-with(stumain.df3, STgrit3cent * STbelong6cent * homeposcent)

##Threeway interaction model

st.ach7<-'
math =~ pv1m + pv2m + pv3m + pv4m + pv5m + pv6m + pv7m + pv8m + pv9m + pv10m
ses =~ PAReduyearscent + PARjobindexcent + homeposcent
grit =~ STgrit1cent + STgrit2cent + STgrit3cent + STgrit4cent
belong =~ STbelong1cent + STbelong2cent + STbelong3cent + STbelong4cent + STbelong5cent + STbelong6cent
##Interaction terms
gxs =~ gxs1 + gxs2 + gxs3
gxb =~ gxb1 + gxb2 + gxb3
sxb =~ sxb1 + sxb2 + sxb3
gxbxs =~ gxbxs1 + gxbxs2 + gxbxs3

##Covariance
grit ~~ cgxs*ses
grit ~~ cgxb*belong
ses ~~ csxb*belong

##Fix latent variable cov with interact at 0
grit ~~ NA*gxs
grit ~~ NA*gxb
grit ~~ NA*gxbxs
ses ~~ NA*gxs
ses ~~ NA*sxb
ses ~~ NA*gxbxs
belong ~~ NA*gxb
belong ~~ NA*sxb
belong ~~ NA*gxbxs
gxs ~~ NA*gxbxs
gxb ~~ NA*gxbxs
sxb ~~ NA*gxbxs

##Intercepts of int terms
gxs ~ cgxs*1
gxb ~ cgxb*1
sxb ~ csxb*1
gxbxs ~ cgxbxs*1

##Intercepts on Latent var
ses ~ NA*1
grit ~ NA*1
belong ~ NA*1

#Intercept of items fix to 0
pv1m ~ 0*1
pv2m ~ NA*1
pv3m ~ NA*1
pv4m ~ NA*1
pv5m ~ NA*1
pv6m ~ NA*1
pv7m ~ NA*1
pv8m ~ NA*1
pv9m ~ NA*1
pv10m ~ NA*1

PAReduyearscent ~ 0*1
PARjobindexcent ~ NA*1
homeposcent ~ NA*1

STgrit1cent ~ 0*1
STgrit2cent ~ NA*1
STgrit3cent ~ NA*1
STgrit4cent ~ NA*1

STbelong1cent ~ 0*1
STbelong2cent ~ NA*1
STbelong3cent ~ NA*1
STbelong4cent ~ NA*1
STbelong5cent ~ NA*1
STbelong6cent ~ NA*1

gxs1 ~ 0*1
gxs2 ~ NA*1
gxs3 ~ NA*1

gxb1 ~ 0*1
gxb2 ~ NA*1
gxb3 ~ NA*1

sxb1 ~ 0*1
sxb2 ~ NA*1
sxb3 ~ NA*1

gxbxs1 ~ 0*1
gxbxs2 ~ NA*1
gxbxs3 ~ NA*1

##Latent regressions
grit ~ ses
belong ~ ses
math ~ grit + ses + belong + gxs + gxb + sxb + gxbxs
'

st.ach7.fit<-sem(st.ach7, data=stumain.df3, meanstructure =T)
summary(st.ach7.fit, standardized = T, fit.measures = T)

st.ach7.probe<-probe3WayMC(st.ach7.fit, c("grit","ses","belong","gxs","gxb","sxb","gxbxs"), "math" ,c("ses","belong") ,
                           valProbe1 = c(-0.379*2,-0.379,0,0.379,0.379*2), valProbe2 = c(-0.7301,0,0.7301))

I am however unable to the probe3WayMC function to produce the simple intercepts of the model. Is there something wrong with my syntax? I am able to get a model that converges.

Melvin Sim

[Alex Schoemann]

Hi Melvin,

I'd double check your output to make sure you're estimating all the latent means you plan to. It looks like the latent mean (really intercept here) for math isn't estimated in the syntax. That's likely leading to the issue you're seeing.

Alex

[Melvin Sim]

Hi Alex,

Yes you were right indeed! Thank you so much for your help!

Melvin