Contributions: Abstract
There are many good reasons for creating parcels in structural equation modeling (SEM). In this
paper, I present clear evidence that how one creates these parcels can have important
implications for the interpretation of the substantive analysis. To the extent that the individual
items do not form a clearly unidimensional domain, it is definitely best to create parcels that are
each representative of all the subdomains within the set of items. I present algebra-based
arguments to support my conclusions, and a brief empirical example that shows how
substantively relevant biases can be introduced if parcels are formed using the "unidimensional"
approach. The results are discussed in terms of assumptions underlying SEM.
|