Multidimensional IRT scores that account for latent variable covariances over time produce better recovery of growth parameters in longitudinal growth models than sum scores and other IRT approaches¶

Claim · All claims
Evidence · 1 study · causal r? · q2
Currency · newest study 2020 · checked against its sources 2026-10-20 (automated, 1 of 1 entry) · not reviewed by a person · overdue for a recheck since 2023: tool and model accuracy changes quickly

Subclaims¶

q2 i? Using multidimensional IRT scores accounting for latent variable covariances over time leads to better recovery of growth parameters than sum scores and other IRT approaches in longitudinal growth models. → Megan Kuhfeld 2020

Evidence¶

Megan Kuhfeld 2020¶

Megan Kuhfeld, James Soland. (2020). Avoiding bias from sum scores in growth estimates: An examination of IRT-based approaches to scoring longitudinal survey responses. Psychological Methods. https://doi.org/10.1037/met0000367

q2 · i? · causal · r?

A series of simulation and empirical studies compared growth parameter recovery across scoring approaches in a single-cohort repeated measure design. The article reports that multidimensional IRT scores "leads to better recovery of growth parameters relative to models using sum scores and other IRT approaches."

"Results indicate that using scores from multidimensional IRT approaches that account for latent variable covariances over time in growth models leads to better recovery of growth parameters relative to models using sum scores and other IRT approaches."

Discussion¶

Related Claims¶

• The multi-rater growth model shows promise relative to cruder approaches like estimating growth from a single rater — related
• Junction-tree factorization reduces EM algorithm complexity for latent growth IRT models from exponential to linear in the number of measurement occasions — related
• Computational complexity of the graph-based MML procedure scales with the number of latent variables within a conditionally independent subset, and brute-force integration scales exponentially with dimensionality — related
• Longitudinal survey growth is typically measured with sum scores or IRT scale scores from a consistent item set administered across multiple school years — related
• Bifactor and second-order structures within measurement occasions keep multidimensional growth IRT models computationally tractable — related
• The effect of response styles on growth estimates varies by psychological construct, response style, and IRT model used — related
• In the empirical growth mindset study, MIRT scoring yields substantively higher intercept and slope means and variances, including an intercept-slope covariance of .13 that is indistinguishable from zero under other approaches — a narrower finding that bears on this claim
• Student growth modeling and teacher value-added modeling are complex due to non-random assignment, measurement error, multidimensionality, and covariates — related
• In a three-timepoint simulation, MIRT-based scores recover true linear growth slope parameters better than sum scores and unidimensional IRT approaches, which understate the slope — a narrower finding that bears on this claim
• Self-management and self-efficacy show highly collinear growth during middle school, with slopes correlated .95 and intercepts .74 — related
• Both cognitive domains showed significant growth, negative self-feedback, and substantial individual differences in change over the interval — related
• Measurement error in the mediator reduces statistical power for mediator-achievement association estimates — related
• Meaningful longitudinal linking depends on the construct not changing across measurement occasions — related
• Little of the existing IRT scaling and linking expertise from large-scale educational assessment is applied to scaling psychological and social-emotional survey constructs — related
• Summing all test scores into one general-intelligence rating manufactures the appearance of a general factor — related