The ESDM's unidirectional trait-ACH-to-trait-ASC path is not empirically identified, and the confounder-based RI-LTVC alternative shows poor fit with no support for a time-varying confounder¶

Claim · All claims
Evidence · 1 study (2 entries) · associational r2 · q2
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q2 i? Identical fit and parameter estimates are obtained whether the ESDM trait path is reversed or replaced by a correlation, so its direction is not empirically identified. → Núñez-Regueiro 2026 q2 i? The RI-LTVC model had the second worst fit and a 0% probability of reflecting the true ASC-ACH relations, providing no support for a time-varying confounder explanation. → Núñez-Regueiro 2026 (2)

Evidence¶

Núñez-Regueiro 2026¶

Núñez-Regueiro, F., Marsh, H. W., Pekrun, R., Lüdtke, O., & Guo, J. (2026). On Model Interpretability and Time-Reversal Testing in Analyses of Reciprocal Effects: Clarifying Modeling Issues in Sorjonen et al.'s (2025a, 2025b) Commentaries. Educational Psychology Review, 38(80). https://doi.org/10.1007/s10648-026-10179-8

q2 · i? · theoretical · r3

Analytical equivalence demonstration reported by the authors: because trait factors are time-invariant random intercepts, reversing or correlating the ESDM path yields "identical fit and parameter estimates", so the direction of the path is not empirically identified.

"identical fit and parameter estimates are obtained if the effect is reversed ( ASC T rait → ACH T rait) or replaced by a correlation ( ACH T rait ↔ ASC T rait), as we showed before"

Núñez-Regueiro 2026 (2)¶

Núñez-Regueiro, F., Marsh, H. W., Pekrun, R., Lüdtke, O., & Guo, J. (2026). On Model Interpretability and Time-Reversal Testing in Analyses of Reciprocal Effects: Clarifying Modeling Issues in Sorjonen et al.'s (2025a, 2025b) Commentaries. Educational Psychology Review, 38(80). https://doi.org/10.1007/s10648-026-10179-8

q2 · i? · associational · r2

Model-comparison result reported from the authors' prior response (Núñez-Regueiro et al., 2025): the random-intercept time-varying covariate model "had the second worst fit" and a 0% probability of being the true model, disfavouring the confounder account.

"This RI-LTVC had the second worst fit to the data compared to other models and a 0% probability of reflecting the true relations between ASC and ACH"

Discussion¶

Related Claims¶

• The reciprocal effects are robust to alternative models attributing them to latent confounding or residual error (RI-LTVC, STARTS) — related
• Reanalyses of simulated and PALMA data provide converging support for the extended reciprocal effects model, including an effect of academic self-concept on achievement — reports the opposite
• Simulation evidence shows RI-CLPM estimates are biased when the true model is a CLPM with lag-2 effects, and fit statistics cannot decide between the two — a broader claim this one bears on
• The extended REM was robust to several alternative modeling specifications proposed by Sorjonen et al. — reports the opposite
• Random-intercept and lag2 panel models rest on different assumptions and offer alternative perspectives on controlling covariates — related
• The ANCOVA and CHANGE models can be transformed into one another, so empirical model fit is no criterion for choosing between them — possibly the same claim (merge candidate)
• Sensitivity analyses showed most sequential CVT paths were spurious due to confounders; only interest→enjoyment was robust — related