Impact of homogeneous and heterogeneous parceling strategies when latent variables represent multidimensional constructs.

Many researchers are interested in using structural equation models to test theoretical relations among multidimensional constructs; however, logistical constraints often prevent researchers from obtaining multiple measures. The current study examines the implications for such models when a latent variable is extracted from carefully constructed parcels of items obtained from a single multidimensional measure of the multidimensional target construct. Two parceling methods are compared. One is homogeneous parceling, in which items are pooled so that each parcel represents a single lower order dimension of the higher order construct. The other is heterogeneous parceling, in which items are pooled so that each parcel represents all lower order dimensions of the higher order construct. Results of simulated and real data analysis reveal that both approaches can result in models that fit the data well; however, conceptual-theoretical differences exist in the nature of the latent variables that are extracted from homogeneous versus heterogeneous parcels. Additionally, compared with homogeneous parceling, heterogeneous parceling generates smaller (i.e., closer to zero) but tighter estimates of structural path coefficients, the net result of which is greater statistical power to test substantive relations among latent variables. Beyond parceling, implications surface about the nature of latent variables that emerge when the underlying constructs are multidimensional. (PsycINFO Database Record