Targeting Toward Inferential Goals in Bayesian Rasch Models for Estimating Person-Specific Latent Traits

In educational assessment, researchers employ measurement models to estimate individuals’ latent abilities, rank these attributes, and analyze group-level distributions to meet various inferential objectives. This paper investigates two strategies for enhancing person-specific latent trait estimation: (a) semiparametric modeling using Dirichlet process mixtures to provide more flexible priors, and (b) alternative posterior summaries, such as constrained Bayes and triple-goal estimators, designed to minimize specific loss functions. We evaluate the performance of these strategies, both individually and in combination, within conditions that are common to curriculum-based assessments. Our results indicate that their effectiveness is highly dependent on test reliability, influenced by the level of within- and between-person variability in the data. Lastly, we emphasize the need to align posterior summaries with particular inferential goals, particularly when balancing trade-offs related to shrinkage effects.