• Analyses & Alternatives
• /
• v.5 no.1
• /
• pp.3-24
• /
• 2021

This paper introduces a multilevel item-response-theory (IRT) model as a unifying model for hypothesis testing using legislative voting data. This paper shows that a probit or logit model is a special type of multilevel IRT model. In particular, it is demonstrated that, when a probit or logit model is applied to multiple votes, it makes unrealistic assumptions and produces incorrect coefficient estimates. The advantages of a multilevel IRT model over a probit or logit model are illustrated with a Monte Carlo experiment and an example from the U.S. House. Finally, this paper provides a practical guide to fitting this model to legislative voting data.

• Korean Journal of Comparative Education
• /
• v.28 no.5
• /
• pp.1-19
• /
• 2018

The purpose of educational assessments such as the Trends in International Mathematics and Science Study (TIMSS) is to compare groups such as countries. When the unit of measurement is above the student level, group-level diagnostic assessment based on multilevel item response theory (ML-IRT) can be considered just as cognitive diagnosis models are developed from item response theory. This study suggests an ML-IRT-based group-level diagnostic assessment model by modifying an item feature model by Park and bolt (2008). The model is illustrated on the recently released TIMSS 2015 Grade 8 mathematics assessment. The results provide skill profiles for the studied countries and the nine cognitive attributes; that is, the attribute effects can be compared across the countries and also across the attributes. By controlling unexplained variance, the suggested model may provide more reliable and more informative group-level comparisons. The results are interpreted using an example. Limitations and directions for future research are also discussed.