Morgan Casal Ribeiro
Hello everyone.
I am trying to
perform cfa, sem and latent growth modeling using lavaan and SemTools, but have
a few questions. Although I have done a good bit of research, this is my first
time using this methodology, so I’m sorry if anything I say is a bit confusing.
I would be
really thankful if the group could confirm if my approach is correct:
- I have measured data for two different years, so first I tested my measurement model with a cfa (separately for each year, and for a combined dataset, all producing good fit).
- Then I checked for measurement invariance between years, establishing configural, metric and scalar invariance, but residual invariance was not shown*. From what I understand, this is not a concern for my analysis.
- I proceeded to structural equation modeling, and the structural model showed good fit to both years. For one of the years, the model showed good fit but none of the relationships between variables were significant (is that strange?).
- I created a second-order latent growth model for each of my latent variables, to see if they changed significantly from one year to another, as well as to check if any sociodemographic variable I also measured influenced that change. One of the latent variables changed significantly from one year to another (both slope and intercept showed statistical significance) and was significantly related to a sociodemographic variable**.
Question 1: Does that process make sense, and are there any important steps
that I may be missing?
*Question 2:
Is the absence of residual invariance a problem for the rest of the
methodology?
Question 3:
After establishing measurement invariance for both years, is it valid to improve
the measurement model of one of the years, adding a theoretically valid
modification due to it having a high modification index and standardized expected
parameter change, while keeping the model of the other year the same? (the change
is adding something similar to “measured_variable1 ~~ measured_variable2”)
**Question 4:
Do I also need to establish measurement invariance across the groups of the
sociodemographic variable that was related to the latent variable change from
one year to the other (latent growth model)? It is a numeric variable (household
size, ranging from 1 to 10), but I have also divided it into groups (1-2, 3-4,
5-6, 7-8, 9-10) and the relationship with the latent variable change was still
significant.
Sorry for the
long post! If needed I can provide the data or the script, but it is a bit of a
mess at the moment.
Thank you in advance!
Terrence Jorgensen
*Question 2: Is the absence of residual invariance a problem for the rest of the methodology?
Question 3: After establishing measurement invariance for both years, is it valid to improve the measurement model of one of the years
**Question 4: Do I also need to establish measurement invariance across the groups of the sociodemographic variable that was related to the latent variable change from one year to the other (latent growth model)? It is a numeric variable (household size, ranging from 1 to 10), but I have also divided it into groups (1-2, 3-4, 5-6, 7-8, 9-10) and the relationship with the latent variable change was still significant.