This talk covers identification issues in structural models and discusses setting a variance to one (for exogenous latent variables), fixing a loading to unity (for endogenous vars) as well as the effects coding strategy and normalized basis methods for securing an identified solution.
topics at variosu tims are:
1:40 example of mathematical identification using Onyx
4:59 Same model but in Amos
7:12 fixing a reference loading to one
12:20 for exogenous variables you might want to fix variance to one. For endogenous latents, you'll need to fix a variable loaidng to unity because the sling on the latent variable is an error term.
14:41 identification using effects coding
18:28 identification via normalized basis (i.e., eigenvalues and eigenvectors)
31:13 empirical nonidentification issues
42:32 local underidentification
43:52 preview of iterative solution in structural models

свернуть