Manifest variable's error term regression weight in a structural regression model

Question

AMOS automatically places a value of 1 on the regression weights for the relationship between an error term and a manifest variable. When I convert my measurement model into a path model, do I keep this regression weights at 1 or do they become unconstrained?

Answer 1

In SEM, both common and unique factors (the latter being "errors" or "residuals") are latent variables. A unique factor only has one indicator (by definition), so you must either fix its variance or "loading" (regression weight) to 1.