A semiannual overview of recent research contributions in cognitive diagnosis, forced-choice formats, and methodological advances in psychology and psychometrics. Each paper is summarized in plain language with key insights highlighted for a broad academic and applied audience.
“Assessing item-level fit for the sequential G-DINA
model”
Behaviormetrika
This paper addresses a gap in diagnostic classification modeling: how to
assess whether each item in a test fits the assumed model when responses
are graded or sequential (e.g., multi-step open-ended
tasks). The authors adapt three fit indices from classical test
theory—the chi-squared statistic, likelihood-ratio statistic, and
power-divergence index—to work with the Sequential G-DINA
model.
Key points: - The model handles multi-category (polytomous) responses that depend on a latent sequence of cognitive steps. - Fit statistics are computed using posterior pseudo-counts and tested via parametric bootstrap. - Simulation results show the proposed methods are conservative but powerful when detecting major misspecifications.
“A general diagnostic modelling framework for forced-choice
assessments”
British Journal of Mathematical and Statistical
Psychology
This paper proposes an extension of cognitive diagnosis models to handle
forced-choice (FC) formats, which are used to reduce
response biases (e.g., social desirability). It adapts the
G-DINA model to handle paired statements, allowing each
to measure a different latent trait.
Innovations: - Provides a general model for
binary forced-choice blocks that improves on Huang’s
(2023) FC-DCM by allowing more flexible response patterns. -
Accommodates heteropolar and homopolar blocks, enabling
normative interpretation of traits. - Supports practical implementation
with Q-matrix design guidelines, Bayesian estimation,
and software integration via the GDINA R package.
“Equally vs. unequally keyed blocks in forced-choice
questionnaires: Implications on validity and reliability”
Journal of Personality Assessment
This experimental study compares equally-keyed
(homopolar) vs. unequally-keyed (heteropolar)
blocks in FC questionnaires measuring the Big Five. Using IRT-based
models (specifically MUPP-2PL), they assess how item keying direction
impacts reliability, criterion
validity, and ipsativity.
Findings: - No consistent psychometric advantage for heteropolar blocks. - Slight increases in reliability and validity in specific traits, but small overall effect sizes. - Practical difficulties in constructing heteropolar blocks with matched social desirability ratings. - Recommendations: prefer equally-keyed designs unless strong justification exists for heteropolar use.
“Cross-validation and predictive metrics in psychological
research: Do not leave out the leave-one-out”
Behavior Research Methods
This methodological paper critiques standard practices for estimating
predictive accuracy in regression models. It proposes a
reformulated leave-one-out (LOO) cross-validation
approach that computes the out-of-sample R² via a
pooled error term (PRESS/MST), solving known biases in conventional CV
implementations.
Contributions: - Shows that LOO offers more stable
and less biased R² estimates than 5-fold or 10-fold CV, especially in
small samples. - Implements methods in the R package OutR2.
- Simulations and real data (Many Labs Replication Project) confirm the
robustness of the approach.
These works share a common goal: enhancing the precision, interpretability, and fairness of psychological measurement—whether by improving test models, detecting fit issues, or refining how predictions are evaluated. Each contribution balances rigorous methodology with clear application potential in education, clinical, and organizational contexts.
All articles available upon request or via journal links. Summaries using IA by M.A. Sorrel. For collaboration, contact miguel.sorrel@uam.es.