I am trying to do latent growth curve modelling using lavaan on two latent variables with three time points. For each latent variable (EF and PEF), there are three indicators.
1st problem:
How do we specify individually varying time intervals? Participants come in every 6 months but realistically, some come after 5 months, some come after 8 months, ending up with individually varying time intervals.
I tried using
intP =~ 1*inh1+ 1*inh2 + 1*inh3 + 1*shif1 + 1*shif2 + 1*shif3 + 1*wm1 + 1*wm2 + 1*wm3
intE =~ 1*gs1+ 1*gs2 + 1*gs3 + 1*ss1 + 1*ss2 + 1*ss3 + 1*sop1 + 1*sop2 + 1*sop3
slopeP=~0*inh1+ inh2 + 1*inh3 + 0*shif1 + shif2 +1*shif3 + 0*wm1 + wm2 + 1*wm3
slopeE=~0*gs1 + gs2 + 1*gs3 + 0*ss1 + ss2 + 1*ss3 + 0*sop1 + sop2 + 1*sop3
- I wonder if this code interprets what I want to specify?
2nd problem: cross-lagged panel modelling for directionality
How do we specify the directionality (perhaps bi-directionality) between the two latent variables over time?
First of all, to figure out what the directionality will be like, I've done panel modelling. In the cross-lagged panel modelling context, I've tried to compare the 4 models:
a. no coupling
b. EF->PEF
c. PEF->EF
d. bidirectoinality - BUT I couldn't figure out how to specify the reciprocity.
What I've tried is to include both directions in one model. Is this right?
- PEF predicting EF
EF2+EF3 ~ PEF1
EF3 ~ PEF2
- EF predicting PEF
PEF2+PEF3 ~ EF1
PEF3 ~ EF2
3rd problem: Specifying the directionality in latent growth curve context?
I've tried:
intP ~~ intP
intE ~~ intE
slopeP ~~ slopeP
slopeE ~~ slopeE
intE ~~ slopeE
intE ~~ slopeP
slopeP ~~slopeE
Does this interpret what I mean? I want to see if: a. intercept of EF predicts slope of EF and/or of PEF b. intercept of PEF predicts slope of EF and/or of PEF c. whether they are reciprocal or whether either of a or b is better fitting
4th question: Model not converging
So the last code is for random intercept, random slope model. But when I run the code, it doesn't converge - any suggestions as to why? I can think of sample-specific reasons (E.g., small/incomplete sample size) but anything else?
Thank you so much for any comments!!