You must parcel carefully if you have to! Comparing eight item parceling strategies with the item-level model for bifactor predictive models.

The bifactor predictive model has been an appealing method for simultaneously examining the validity of the general factor and group factors of hierarchical constructs. Item parceling has been frequently applied in the bifactor context, but all of these applications in bifactor models were based on parceling strategies developed for simple factor structures where one item only loads on one factor. It remains unexplored regarding how to create parcels that can consider the double-loading structure of bifactor models and still allow users to maintain the bifactor structure after item parceling. Moreover, it is unclear whether item parceling itself is beneficial or detrimental in the context of bifactor predictive models. In the present study, we propose eight parceling strategies for bifactor models and investigate (a) whether parceling performs better than item-level modeling and (b) the impact of the eight item parceling strategies on the statistical performance of bifactor predictive models. Contrary to prior studies showing that parceling generally enhances statistical performance for nonbifactor models, the current simulations and the empirical example clearly demonstrate (a) that the item-level model tends to outperform parceling strategies overall and (b) that different parceling strategies demonstrate substantially different statistical performance. Among the eight parceling strategies, the three parceling strategies that maximize factor loading differences within each group factor performed relatively the best, often close to the item-level model. In contrast, strategies that minimize factor loading differences within each group factor performed substantially worse. This study also provides an in-depth discussion and practical guidance for applied researchers. (PsycInfo Database Record (c) 2026 APA, all rights reserved).