Hello,
Iam testing a reflective model with 2 latent variables and 32 indicators on model fit and measurement invariance between two groups. My sample size is approximately N = 130 (n1 = 80, n2 = 50).
Is there a way to test this in CBSem or is my sample size too small for testing this model? If not, can I test it with PLS-Sem/ MICOM alternatively or are there other and better methods to do it out-of-sample?
Thanks!
Here is what I think. Because you seem to plan to use SEM, some people would recommend a sample of at least 200 (e.g., Hinkin 1998, Hoelter 1983) per group. However, Hoelter (1983) also seems to suggest that a sample of less than 200 can still give you reasonably good co-variance structure if the assumptions of normality are not violated.
I couldn't understand what you meant by "out-of-sample." I think people use this term while developing models for forecasting or prediction. But I am not very sure about that.
Best,
Rajiv
Rajiv Kumar Thank you, for your answer! For data collecting I used a 4-point Likertscale (from 1 to 4) and I fear that my data does not meet the criteria of normality as Shapiro-Wilk-Test measured in one of two groups tells me. Is there any possibility still to investigate measurement invariance with non-parametric methods?
The statistical program JASP allows to evaluate the analysis of invariance or multigroup analysis in SEM models in a simple way, there are tutorials in youtube.
Hello Janina:
Unfortunately I don't know much about that. A quick search led me to a paper by Roger Millsap in 2007 (DOI: 10.1007/S11336-007-9039-7), and as I scanned this paper, I could not get a clear sense.
Regards,
Rajiv
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis