Abstract This issue's cover features one of the entries to the 2026 Educational Measurement: Issues and Practice ( EM:IP ) Cover Graphic/Data Visualization Competition. This graphic entitled “Response Dominance Map: Cut Scores for a Polytomously Scored Item Bank Based on Modal Category Transitions” was submitted by Enis Dogan of the National Center for Education Statistics and Burhan Ogut of the American Institutes for Research.
Module Abstract Unstructured data, for example, text, audio, handwriting and drawings, video, and interaction logs, are increasingly abundant in education. However, they lack the predefined schema that conventional psychometric tools assume. This module introduces a principled, end‐to‐end workflow for turning such data into defensible measures. In this module, we first define unstructured data (Section 1) and then proceed to four analytic stages: preprocessing across modalities (Section 2), data representation (Section 3), making inferences (Section 4), and evaluating the resulting measures against psychometric standards (Section 5). Finally, Section 6 gives an overview of data privacy, consent, and governance. We demonstrate the methodologies using a dataset collected from a collaborative problem‐solving task. By the end, learners will have a conceptual map of the analysis pipeline from raw signals to valid inference, a solid understanding of modern representation methods, and a framework for evaluating and governing measures built from unstructured data in the field of education.
Abstract Bayesian networks emerged as a psychometric modeling paradigm in the 1990s and have been gainfully employed in a variety of educational assessment contexts and applications. This article surveys Bob Mislevy's work on Bayesian networks and related graphical models in educational assessment. It is argued that Mislevy made seminal and influential contributions in this space, shaping how these models would be conceived, used, and further developed.
Abstract Multidimensional multistage testing (MMST) combines multidimensional item response theory (MIRT) with adaptive multistage testing design to measure multiple related constructs efficiently. Despite its potential for large‐scale assessments, little research has examined calibration methods under MMST. This study compared concurrent calibration, fixed item parameter calibration, and concurrent calibration with multiple panels using a simulated three‐stage MMST design with a within‐item two‐dimensional MIRT structure. Sample size, trait correlation, and routing stage length were manipulated across 36 conditions. Results showed that larger sample sizes and longer routing stages generally reduced RMSE for discrimination parameters, whereas higher trait correlations increased discrimination RMSE. In contrast, ability RMSE decreased as trait correlation increased, while the effects of sample size and routing stage length were less uniform. Difficulty RMSE remained relatively stable across conditions. Differences among calibration methods were generally small, with no significant pairwise differences among methods.
Abstract Low‐stakes assessments are particularly vulnerable to rapid guessing behavior, where responses may reflect minimal cognitive engagement, potentially biasing scores and weakening inferences. Drawing on a scoping review of more than 70 studies, we summarize how rapid guessing is identified, its impact on key parameters, and how it is treated. Detection methods range from time‐based thresholds to statistical models using response‐time or residual data, each balancing practicality and precision. Treatments include filtering, downweighting, and model‐based adjustments, with differing consequences for score meaning and comparability. However, decisions about identifying rapid guessing, evaluating its impact, and selecting treatments are often not explicitly linked, making it difficult to arrive at coherent analytic choices. To address this, we propose a four‐step workflow and reporting guidelines: (1) clarify construct relevance, (2) screen using context‐appropriate methods, (3) assess impact via sensitivity analysis, and (4) justify a treatment. Together, these tools provide a practical framework for defensible educational measurement practice.
