csemIRT: An R Package for Bootstrap-Validated Conditional Standard Errors of Measurement in Polytomous Item Response Theory Models

Marginal reliability summarizes measurement precision across the full latent trait distribution in a single coefficient, obscuring how precisely a scale measures at different trait levels. The conditional standard error of measurement (CSEM) addresses this limitation, but the CSEM profile is itself an estimate and depends on the item response model chosen. This paper introduces csemIRT, an R package for CSEM analysis under the graded response model, the generalized partial credit model, the partial credit model, and the graded and Rasch rating scale models. The package places all CSEM profiles on a common latent metric, quantifies their sampling uncertainty with delta-method and bootstrap confidence bands, tests pointwise whether competing models imply different measurement precision, and reports the trait regions in which a target conditional reliability is reached. Three simulated datasets are included. csemIRT is applicable to any polytomous scale development and validation context and is freely available via GitHub.