Advancing Educational Assessment

Advanced Psychometrics and Quantitative Methodologies

Comprehensive Research Program

• Methodological Pillars: Second-order latent structures, theta parameterization, Many-Facet Rasch Modeling, Latent Class Analysis (LCA), and BSEM GPU acceleration (Sinval and Merkle 2026).
• Empirical Grounding: Data gathered across major international initiatives spanning over 20,000 participants in Europe, Asia, and the Americas (Sinval et al. 2022; Monteiro et al. 2026).
• Systemic Impact: Bridging different measurement frameworks with evidence-based policy in healthcare, higher education, and occupational health (Melnik et al. 2025; Rasooli et al. 2025).

Psychometric Evaluation of Burnout: BAT-12 & BAT-23

Cross-National Measurement Invariance (N = 10,138)

• Large-Scale Adaptation: Evaluating national quota samples across seven countries (Netherlands, Belgium, Germany, Austria, Ireland, Finland, Japan) (Beer et al. 2020).
• Hierarchy of Invariance: Establishing configural, metric, and full scalar invariance (\(\Delta CFI \leq 0.008, \Delta RMSEA \leq 0.060\)).
• Latent Mean Comparisons: Finding statistically significant cross-national latent mean differences, with Japan exhibiting elevated burnout (\(\beta = 0.71, p < .001\)).

Multi-Group CFA & Theta Parameterization

• Parameterization Mechanics: Estimating indicator thresholds (\(\tau\)) rather than continuous intercepts for 5-point Likert response formats (Sinval et al. 2022).
• Identification Constraints: Resolving complex second-order identification by fixing first-order latent means and setting marker variable indicator loadings.
• Fit Metric Evaluation: Prioritizing Standardized Root Mean Square Residual (\(SRMR < .050\)) over \(RMSEA\) due to known ordinal estimation biases (Beer et al. 2020).

Transcultural Validation in Brazil and Portugal

• Bicontinental Assessment: Testing BAT-23 and BAT-12 in multi-occupational cohorts across Brazil (\(n = 2,217\)) and Portugal (\(n = 886\)) (Sinval et al. 2022).
• Strict Invariance Achievement: Establishing full uniqueness measurement invariance across countries and gender groups.
• Internal Consistency Reliability: Demonstrating high second-order omega estimators (\(\omega_{L2} = .90, ω_{partial~L1} = .97\)) (Sinval et al. 2022).

Longitudinal & Serial Mediation in Student Success

%%{init: {
  "theme": "base",
  "themeVariables": {
    "fontSize": "28px"
  }
}}%%
graph LR
    classDef var fill:#f3f4f6,stroke:#004aad,stroke-width:2px;
    A(PsyCap) --> B(Engagement)
    B --> C(Burnout)
    A --> C
    class A,B,C var;

Organisational Health: Georgia School Personnel Survey

• Large-Scale Application: Adapting the 29-item GSPS across multi-occupational school staff in 2023 (\(N = 1,965\)) and 2024 (\(N = 2,884\)) (Mendes et al. 2025).
• Second-Order Architecture: Confirming a six-dimension first-order structure loading onto a second-order School Climate factor (\(CFI = .97, TLI = .96\)).
• Occupational Group Invariance: Proving scalar and full uniqueness invariance between teachers (\(72.1\%\)) and school support staff (\(27.9\%\)) (Mendes et al. 2025).

Multidimensional IRT & Rasch Modeling

• IRT Framework: Deploying the Multidimensional Random Coefficients Multinomial Logit Model (MRCMLM) via TAM in R (Mendes et al. 2025).
• Item & Person Fit Statistics: Evaluating Infit and Outfit Mean Square (MNSQ) statistics within acceptable \(0.6–1.4\) rating scale boundaries.
• Latent Trait Precision: Estimating Expected A Posteriori (\(EAP\)) reliability indices (\(EAP \geq .80\)) across multidimensional constructs.

Item-Person Mapping & Diagnostic Utility (Wright Maps)

• Visualizing Construct Coverage: Utilizing Wright Maps (WrightMap package) to map individual respondent trait levels (\(\theta\)) against item threshold difficulties (\(\tau\)) (Mendes et al. 2025).
• Targeting & Scale Coverage: Identifying response scale coverage across the full continuum of student and teacher populations (Sinval et al. 2021).
• Diagnostic Refinement: Using IRT item difficulty parameters to refine short-form screeners and adaptive diagnostic tests.

Academic Engagement Validation (USEI, N = 908)

• Construct Mapping: Evaluating the University Students Engagement Inventory (USEI) across Behavioral, Emotional, and Cognitive Engagement (Sinval et al. 2021).
• Hierarchical Structure: Confirming a second-order Academic Engagement meta-construct (\(\chi²_{(87)} = 286.665, CFI = .987, RMSEA = .051\)).
• Educational Field Invariance: Achieving full measurement invariance across diverse academic majors (Engineering, Law/Econ, Humanities) and gender.

Person-Oriented Methodologies: Latent Class Analytics

• Typological Diagnostics: Transitioning from variable-centered SEM to person-centered Latent Class Analysis (LCA) via depmixS4 and poLCA (Monteiro et al. 2022).
• Model Selection Criteria: Utilizing Bayesian Information Criterion (\(BIC = 14,807.84\)) and \(AIC\) to determine optimal class solutions (Melnik et al. 2025).
• Empirical Subgroup Profiles: Identifying distinct graduate employability profiles and practitioner typologies.

Clinical Reasoning & Evidence-Based Practice Diagnostics

• Diagnostic Psychometrics: Evaluating accuracy, sensitivity, and specificity of psychometric tools used in evidence-based practice (Melnik et al. 2025).
• Methodological Standards: Aligning clinical diagnostic evaluations with the AERA/APA/NCME Standards for Educational and Psychological Testing.
• Translational Impact: Demonstrating how psychometric calibration prevents diagnostic misclassification in clinical and school psychology.

Assessment Fairness Literacy (CAFI Framework)

• Fairness Literacy: Investigating preservice teachers’ assessment decision-making via the Classroom Assessment Fairness Inventory (CAFI, N = 228) (Rasooli et al. 2025).
• Exploratory Factor Analytics: Deploying polychoric correlation matrices (\(\rho_{PC}\)), parallel analysis, and Very Simple Structure (\(VSS\)) via psych and EFAtools.
• Methodological Findings: Resolving item wording effects versus true construct multidimensionality in educational assessment fairness.

Frequentist Limitations in Complex SEM Architecture

• Frequentist Failure Modes: Highlighting Maximum Likelihood (ML) and WLSMV susceptibility to Heywood cases, non-convergence, and sample size constraints (Sinval and Merkle 2026).
• The Invariance Dilemma: Explaining the “Scylla and Charybdis” of exact measurement invariance — where minor parameter differences force model rejection.
• Computational Inflexibility: Standard ML inability to incorporate cumulative prior scientific knowledge into structural parameter estimation.

Bayesian Structural Equation Modeling (BSEM) & Priors

• Epistemological Shift: Treating parameters as probability distributions rather than fixed unknown constants (Sinval and Merkle 2026).
• Informative & Small-Variance Priors: Replacing rigid zero constraints (\(parameter = 0\)) with flexible small-variance priors (\(\mathcal{N}(0, 0.01)\)).
• Stabilizing Estimation: Eliminating Heywood cases, accommodating small sample sizes, and naturally propagating measurement uncertainty.

Approximate Measurement Invariance via BSEM

• Approximate Invariance Framework: Utilizing Bayesian MCMC to allow minor cross-group threshold and loading differences via small-variance priors (Sinval and Merkle 2026).
• Posterior Predictive Checks: Evaluating model fit using Posterior Predictive p-values (\(PPP ≈ 0.50\)) and \(95\%\) credibility intervals.
• Cross-Cultural & Longitudinal Utility: Enabling meaningful latent mean comparisons when exact frequentist invariance fails.

The Computational Bottleneck in Ordinal BSEM

• The Computational Barrier: BSEM estimation via Markov Chain Monte Carlo (MCMC) sampling requires hours, days, or weeks on standard CPUs (Sinval and Merkle 2026).
• The Ordinal Bottleneck: Estimating continuous latent response variables (\(y^\ast\)) and threshold vectors for categorical items multiplies matrix complexity.
• The No-U-Turn Sampler (NUTS): Demanding dense gradient evaluation steps in Hamiltonian Monte Carlo (HMC) within Stan.

High-Performance Computing & GPU Acceleration

• Strategic FCT/CPCA Supercomputing Project: Principal Investigator on high-performance computing initiatives leveraging MareNostrum 5 and Deucalion supercomputers (Sinval and Merkle 2026).
• Software Architecture: Integrating OpenCL GPU parallel processing into blavaan (R package interfacing with Stan C++ backend).
• Targeted Hardware: Utilizing NVIDIA Hopper H100 and Ampere A100 GPU architectures to execute parallelized MCMC matrix algebra.

Algorithmic Architecture & Benchmarking

• Simulation Matrix: Benchmarking \(20,000\) simulation cells crossing sample sizes (\(N = 200, 1000, 5000\)), factor complexities, and categories (Sinval and Merkle 2026).
• Execution Metrics: Measuring Wall-clock time, Effective Sample Size per second (ESS/s), and GPU memory transfer overhead.
• Diagnostic Equivalency: Proving absolute parameter convergence (\(\hat R \leq 1.01\)) and relevant speedups.

Methodological Synthesis: Translating Computing to Education

• Methodological Triad: Synthesizing second-order SEM, Multidimensional IRT/Rasch, and GPU-accelerated BSEM into a unified psychometric toolkit (Sinval and Merkle 2026).
• Open-Source Dissemination: Delivering fully optimized GPU-enabled blavaan code and vignettes directly to the global R statistical repository.
• Transition to NIE: Applying world-class quantitative computing to solve complex structural measurement problems in Singapore’s educational ecosystem.

Current Synergies: Collaborative Research at NIE

Strategic Transition: Advanced Psychometrics for Educator Well-Being

• The Research Plan: Teacher and principal burnout fundamentally undermines systemic educational quality and student outcomes (Storti et al. 2025; Alves et al. 2022).
• The Multi-Level Reality: Schools are hierarchical. Teachers are nested within schools, governed by principals.
• The Paradigm Shift: Applying my Bayesian modeling architecture (ML-BSEM) to properly evaluate occupational health and climate within hierarchical educational data.

The Dilemma: Small Data & Hierarchical Complexity in Schools

• The “Small N” Problem at Level-2: Quantitative studies of school leadership often lack the statistical power to model school-level (Level-2) variances effectively using frequentist ML.
• Hierarchical Confounding: Failing to separate within-school (teacher-level) variance from between-school (e.g., principal-level) variance produces biased standard errors.
• Psychometric Distortion: Standard frequentist models either collapse the hierarchy or fail to converge when \(N_{\text{schools}} < 50\).

Research Paradigm: Multilevel Bayesian Well-Being

• Core Vision: Establishing a robust quantitative framework at NIE for assessing organizational health and leadership climate (Sinval and Marôco 2020).
• Methodological Pillars:

1. Multilevel Bayesian SEM (ML-BSEM) for nested educational data.
2. Informative Priors to overcome the principal “Small N” bottleneck.
3. Implicit Measures & Vignettes to capture true occupational distress.

• Systemic Output: Providing MOE and NIE with precise, actionable data on school climate interventions.

Multilevel BSEM (ML-BSEM): Principals and Teachers

%%{init: {
  "theme": "base",
  "themeVariables": {
    "fontSize": "28px"
  }
}}%%
graph TD
    classDef l2 fill:#e0e7ff,stroke:#4338ca,stroke-width:2px;
    classDef l1 fill:#dcfce7,stroke:#15803d,stroke-width:2px;
    
    A(Principal Support L2) --> B(School Climate L2)
    C(Teacher PsyCap L1) --> D(Teacher Burnout L1)
    B -.-> D
    
    class A,B l2;
    class C,D l1;

Overcoming the “Small N” Problem with Informative Priors

• Eliciting Informative Priors: Incorporating prior knowledge (from meta-analyses or previous MOE datasets) into the Level-2 covariance matrices.
• Stabilizing the Hierarchy: Using Bayesian algorithms to allow model convergence even when evaluating a small network of targeted schools (\(N_{\text{schools}} = 15\text{--}30\)).
• Context-Aware Diagnostics: Providing school leaders with posterior latent trait distributions that explicitly account for their unique organizational context.

Systemic Impact & Grant Architecture at NIE

• Alignment with National Priorities: Directly supporting MOE’s focus on 21st Century Competencies (21CC), educator retention, and holistic school environments.
• Interdisciplinary Collaboration: Partnering with NIE research clusters in leadership, occupational health, and organizational psychology.
• External Grant Trajectory: Targeting funding via MOE Education Research Funding Programme (ERFP) and NRF Social Science Research Council (SSRC) Tier 2 grants.

Conclusion & Roadmap for NIE Faculty

• Summary of Portfolio: A bridge from completed methodological rigor (SEM, MI, MIRT, LCA, GPU-BSEM) to innovations in modeling educator well-being.
• Teaching & Mentorship Commitment: Educating the next generation of Singaporean eduational researchers and teachers in quantitative literacy, assessment fairness, and advanced statistics (Rasooli et al. 2025).
• Final Vision: Contribution to NIE as the premier international hub for Educational Modeling.