I have four latent variables. However, two latent variables (according to the theoretical constructs) have only one measured variable each. Is it advised to proceed with SEM with latent variables having only one measured variable in each? If so, what shall be the identification criteria?
There is no way to identify the loading for a single variable in SEM. Instead, you have to "fix" it at some predetermined value that reflects its reliability.
There is no way to identify the loading for a single variable in SEM. Instead, you have to "fix" it at some predetermined value that reflects its reliability.
It should be better to avoid and try for simple path analysis if you want to describe your model.
For your two measures where you have multiple variables, you should use coefficient alpha to ascertain their reliability, which you can also compare to a pair of Exploratory Factor Analyses. In particular, alpha assumes equal weighting for all of the variables, while factor analysis gives them separate weights (loadings), so when all the variables have approximately the same loading, the variables can be simply added together to form a scale.
You can then either simply add the variables to create scales or use the appropriate factor scores. Either way, you can use the resulting measures in path analysis, which would be the same as ordinary least squares regressions.
The latent variable (unobserved variable) is only for the construct that having more than two items. If you have only a single item, you may proceeds your analysis by apply the observed variable.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Techniques
- Structural Equation Modeling