Revising the rubric to operationalize pedagogical constructs raises human–machine agreement for openness and depth¶

Claim · All claims
Evidence · 1 study · causal r1 · q2
Currency · newest study 2026 · checked against its sources 2026-10-20 (automated, 1 of 1 entry) · not reviewed by a person · recheck by 2029: tool and model accuracy changes quickly

Subclaims¶

q2 i? Under the revised rubric with direct evaluation, openness agreement rose from 55.7% to 83.3% and depth agreement from 70.5% to 95.0%. → Pei-Yu Tseng 2026

Evidence¶

Pei-Yu Tseng 2026¶

Pei-Yu Tseng, Mahir Akgun, and Peng Liu. (2026). Why Machines Misread Pedagogical Quality: Human–Machine Alignment in LLM-Based Pretest Question Evaluation. arXiv preprint. https://arxiv.org/abs/2606.23629 doi:10.48550/arxiv.2606.23629

q2 · i? · causal · r1

Experiment 2 (revised rubric, direct evaluation) re-scored machine-generated question pools against human instructor ratings. The quote reports the printed agreement, bias and MSE changes; no standardised effect size is printed.

"openness agreement increased from 55.7% to 83.3%, bias decreased from 0.66 to 0.22, and MSE from 1.08 to 0.32. For depth, agreement increased from 70.5% to 95.0%, with very small remaining error."

Discussion¶

Related Claims¶

• Prompt quality and detailed scoring rubrics significantly influence the accuracy and consistency of LLM-generated grades — the general claim this finding bears on, with the studies behind it
• Combining the revised rubric with rationale-first evaluation yields the highest human–machine alignment, with dimension-specific gains — related
• Under the initial rubric with direct evaluation, machine raters systematically underestimate openness while relevance and clarity align near-perfectly — related
• Rationale-first evaluation improves depth alignment under the initial rubric but yields only limited gains for openness — related
• Focused rubric refinement that made required physics and scoring conditions explicit reduced AI–official disagreements on targeted questions (e.g., Grover item MAD 1.1 to 0.4 marks, r 0.70 to 0.82) — related
• Systems Thinking Practices is the hardest CT construct to code reliably for both human raters and the LLM, requiring more explicitly operationalized rubrics — related
• GPT-4o's rubric-guided grading of team communication disagreed with instructors at near-chance levels (55% RMSE), while GPT-5.2 improved but remained insufficiently aligned (35%) — related