This article argued that traditional estimates in item response theory ignore the influence of sampling error, causing underestimated standard errors of measurement and overestimated marginal reliability. The authors compared three types of methods that have been used to address this problem, and proposed a MI-based procedure to account for such uncertainty in the present study. Two simulated data and one empirical example were used, and findings supported the use of the proposed MI-based model. It is conclusive that the pronounced advantage of the approach over the existing alternatives is its flexibility that the approach can be applied to a variety of IRT models, supported by the three-item artificial data, and that the proposed model connects various methods (e.g., full-pattern or summed-score EAP, 2-PL, 3-PL, and graded-response models) with evidence of the second simulation study. However, I am curious that if the proposed approach functions better than MCMC, in exclusive of simplicity and easy computations. In addition, since Rasch model was found to yield acceptable estimates of latent ability and SEM even with small calibration samples, it might be better to choose Rasch model first before complicated models or the proposed approach are used in data analysis.