Using measurement parcels in model estimation. Our hypotheses related to the structural model, and prior work suggests that use of multi-item parcels as indicators for latent variables is defensible in such situations (Bandalos & Finney, 2001; Marsh & O’Neill, 1984). Use of parcels in these circumstances addresses rater effects and reduces the number of estimated paths in the model. A domain-representative approach to parcel construction treats information from each reporter as equally valid (or equally biased) and unit-weights the raters by distributing their information across the parcels (for an empirical example see Schofield et al, 2009). Following the procedures outlined by Kishton and Widaman (1994), domain representative parcels were created, which allowed rater-specific variance and variance common across raters to contribute to the latent factor. As a further test of this procedure, we first ran the models using parcels as just described. We then ran the models separately for each informant. The results from these preliminary analyses supported the results from the parcel model we present here.
