Jayden Nord posted on Tuesday, March 14, 2017 - 11:49 am
In Web Note 6, a random slope approach for latent interaction estimation is described and is accompanied by example syntax. Alternatively, the XWITH command may be used to define an interaction term. Using these two approaches on the same dataset, I get slightly different results in the parameter estimates, standard errors, etcetera. Are these approaches inherently different? If not, what causes the differences; if yes, how are they different?
First check if they use the same number of parameters. If they don't, see where the difference lies. If they do, sharpen convergence criterion.
Jayden Nord posted on Tuesday, March 14, 2017 - 6:53 pm
Both approaches have the same number of parameters. Despite using a smaller convergence criterion, differences persist. There are, however, no differences in the loadings and factor variance for just one of the two interacting latent predictors.