Research

Table of Contents

1. Research Overview
2. Cognitive Diagnosis and Adaptive Assessment
3. Item Response Theory and Knowledge Space Theory
4. Sequential Methods for Assessment and Learning
5. Current Research
6. Applied and Collaborative Methodology
7. Selected Publications
8. Additional Information

Research Overview

My research concerns statistical models for educational and psychological measurement, with particular emphasis on cognitive diagnosis models (CDMs), item response theory (IRT), knowledge space theory (KST), and sequential decision methods.

The central problems in my work concern latent structure and statistical decision making: how knowledge or mastery should be represented, when latent states can be identified from observed responses, how assessment information should be collected, and when accumulated evidence is sufficient to make a classification or instructional decision.

These questions arise naturally in computerized adaptive assessment and learning systems, where responses are observed sequentially, and assessment decisions can affect subsequent instruction. My work therefore combines psychometric modeling with sequential analysis, Bayesian methods, and adaptive experimental design.

Back to Top

Cognitive Diagnosis and Adaptive Assessment

Cognitive diagnosis models represent proficiency in terms of discrete skills or attributes rather than a single continuous latent score. My work in this area concerns both the statistical structure of these models and their use for adaptive assessment and learning.

Specific topics include:

• classification of multidimensional mastery profiles;
• Q-matrix specification and identification;
• adaptive item selection under cognitive diagnosis models;
• sequential detection of changes in mastery;
• diagnostic assessment using multiple-choice and other structured response formats;
• incorporation of response process information into mastery decisions; and
• relationships between assessment, instruction, and latent skill transitions.

An early part of this work developed sequential procedures for detecting learning under cognitive diagnosis models. More recent work extends this framework to computerized adaptive learning, where assessment and instructional decisions are made repeatedly as evidence about mastery accumulates.

The objective is not simply to estimate a latent profile more accurately. The broader statistical problem is to determine how diagnostic information should be collected and used when assessment itself is part of a learning process.

Back to Top

Item Response Theory and Knowledge Space Theory

A second line of research examines the relationship between item response theory and discrete latent-variable approaches, particularly knowledge space theory.

IRT typically represents individual differences through continuous latent variables, whereas knowledge space and related diagnostic models represent proficiency through collections of admissible knowledge states. These frameworks make different assumptions about latent structure, dependence, and response generation, but they can also produce closely related observable response distributions.

My work has examined:

• parameter estimation for KST-IRT models;
• local dependence within KST-IRT formulations;
• identification of three- and four-parameter IRT models from a knowledge-space perspective;
• formal relationships among IRT, knowledge structures, and other dichotomous latent-variable models; and
• conditions under which alternative psychometric representations are observationally distinguishable.

This line of work is concerned primarily with model structure and identifiability rather than with selecting one framework as universally preferable.

Back to Top

Sequential Methods for Assessment and Learning

Sequential analysis provides the principal statistical framework connecting much of my psychometric research.

In a conventional assessment, the number of observations is generally determined before data collection. In adaptive assessment and learning, however, responses arrive sequentially, and the relevant question is often whether enough evidence has accumulated to justify a decision.

My research considers procedures for:

• sequential change detection, particularly detection of changes in latent mastery;
• sequential hypothesis testing for classification and certification decisions;
• adaptive stopping rules for computerized assessment;
• Bayesian sequential decision procedures;
• adaptive allocation and item selection; and
• statistical control of decision errors when assessments continue over time.

These methods are especially useful when the timing of a decision is itself important. For example, delaying a mastery decision unnecessarily increases assessment burden, whereas making the decision too early increases the probability of incorrect classification.

My work therefore treats assessment length, classification accuracy, and learning efficiency as related statistical decision problems.

Back to Top

Current Research

Dual-threshold sequential designs for adaptive learning

Current work develops sequential designs that use evidence of mastery both to guide continued learning and to determine when mastery can be certified. The design uses separate decision criteria for instructional adaptation and final mastery detection.

The statistical problem is to balance learning efficiency, detection delay, and classification error while allowing assessment and instruction to evolve jointly over time.

Identification of Q-matrices and latent profiles

I am studying conditions under which Q-matrix structures that satisfy conventional identification requirements may nevertheless fail to distinguish particular latent profiles under general or mixed cognitive diagnosis models.

This work concerns the distinction between parameter identification at the model level and practical separability of the latent classifications that the model is intended to produce.

Information-guided adaptive learning

Another line of work examines item and instructional selection rules for accelerating mastery. Current approaches include information-gain criteria, posterior predictive quantities, and Bayesian adaptive allocation procedures.

The objective is to select observations that are useful not only for measurement but also for the progression of learning.

Response-process information

I continue to examine procedures that combine response accuracy with additional process information, including response time, for sequential mastery detection. Process variables may provide evidence about changes in proficiency before those changes are fully reflected in response accuracy.

Skill-profile analysis in vocational assessment

In collaborative work, cognitive diagnosis methods are being applied to extended-reality vocational training data for adults with intellectual and developmental disabilities. The objective is to characterize multidimensional skill profiles rather than reduce performance to a single aggregate score.

Back to Top

Applied and Collaborative Methodology

My primary research program is psychometric, but I also work as a statistician on interdisciplinary studies in education, rehabilitation, behavioral science, and health research.

Methodological work in these collaborations has included:

• study design and statistical analysis for randomized clinical trials;
• longitudinal and correlated-data analysis;
• survival analysis;
• mixed-effects and machine-learning models;
• reliability and allocation problems;
• educational and vocational assessment; and
• quantitative design for externally funded research.

These collaborations provide applied settings in which statistical and measurement problems can be studied outside conventional testing environments.

Back to Top

Selected Publications

Adaptive assessment and cognitive diagnosis

• Ye, S. & de la Torre, J. (2025). Two-stage computerized adaptive learning detection. Journal of Educational and Behavioral Statistics.
  doi:10.3102/10769986251397632

• Ye, S., Fellouris, G., Culpepper, S., & Douglas, J. (2016). Sequential detection of learning in cognitive diagnosis. British Journal of Mathematical and Statistical Psychology, 69(2), 139–158.
  doi:10.1111/bmsp.12065

Item response theory and knowledge space theory

• Noventa, S., Heller, J., Ye, S., & Kelava, A. (2025). Toward a unified perspective on assessment models, part II: Dichotomous latent variables. Journal of Mathematical Psychology, 125, 102926.
  doi:10.1016/j.jmp.2025.102926

• Noventa, S., Ye, S., Kelava, A., & Spoto, A. (2024). On the identifiability of 3- and 4-parameter item response theory models from the perspective of knowledge space theory. Psychometrika, 89(2), 486–516.
  doi:10.1007/s11336-024-09950-z

• Ye, S., Kelava, A., & Noventa, S. (2023). Parameter estimation of KST-IRT model under local dependence. Psych, 5(3), 908–927.
  doi:10.3390/psych5030060

Sequential statistical methodology

• Ye, S. & Rekab, K. (2024). On the Bayes risk of a sequential design for estimating a mean difference. Communications for Statistical Applications and Methods, 31(4), 427–440.
  doi:10.29220/CSAM.2024.31.4.427

A complete publication record is available through my Google Scholar profile.

Back to Top

Additional Information

• Curriculum Vitae
  A complete record of publications, presentations, funded research, teaching, and professional service is available in my CV.

• Teaching
  Courses and methodological training are summarized on the Teaching page.

• Invited Talks and Workshops
  Invited lectures, workshops, and conference training sessions are listed under Talks & Workshops.

• Contact
  Contact information and professional profiles are available in the site sidebar.

Back to Top