Course Text: Pages 9-10
If a model does not provide an adequate fit to data overall, this means that the local independence assumption is violated because the model fails to account for some of the associations among the indicators. In this case, a measure of local fit known as bivariate residuals (BVR) can be useful to examine the association between each pair of variables to see which of these local associations the model fails to explain. Later, in Topic K, we will describe an approach where one or more large BVRs are singled out and used to add model parameters called direct effects to a model to improve the model fit. This approach is an alternative to the standard strategy of increasing the number of latent classes.
Assigned Reading:
LG tutorial 1:
D: Bivariate Residuals (BVRs), pages 17-18