Midterm self-efficacy (comfort and confidence) did not differ between conditions after three assignments, though both varied significantly across algorithmic topics¶

Claim · All claims
Evidence · 1 study (2 entries) · causal r2 · q3
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q3 i? No significant between-group differences emerged on the Comfort composite (U=5,788, p=.374) or Confidence composite (U=5,152, p=.561) after three assignments. → Ethan Dickey 2026 q3 i? Within subjects, Asymptotic Analysis and Divide & Conquer were rated significantly higher than Dynamic Programming, Greedy Methods, and Graph Algorithms on both scales (Holm-adjusted p<.001). → Ethan Dickey 2026 (2)

Evidence¶

Ethan Dickey 2026¶

Ethan Dickey, Marios Mertzanidis, Alexandros Psomas. (2026). Is Solving Better Than Evaluating GenAI Solutions? https://arxiv.org/abs/2607.27586 doi:10.48550/arxiv.2607.27586

q3 · i? · causal · r2

Pre-midterm Qualtrics survey (N=208, 95% response rate) with five-point Likert items; Comfort (alpha=0.79) and Confidence (alpha=0.84) composites compared via Mann-Whitney U tests between crossover conditions.

"No statistically significant differences were observed between groups on either the Comfort composite (𝑈= 5,788,𝑝= .374) or the Confidence composite (𝑈= 5,152,𝑝=. 561) after three assignments under their respective conditions."

Ethan Dickey 2026 (2)¶

Ethan Dickey, Marios Mertzanidis, Alexandros Psomas. (2026). Is Solving Better Than Evaluating GenAI Solutions? https://arxiv.org/abs/2607.27586

q3 · i? · causal · r2

Topic-level analysis of the same survey found Asymptotic Analysis (Comfort median=4) and Divide & Conquer rated highest; Comfort and Confidence composites were strongly associated (Spearman's rho=0.714, p<.001).

"Friedman tests indicated significant within-subject variation across topics for both Comfort ( 𝜒 2(4)= 156.78,𝑝<. 001) and Confidence (𝜒 2(4)= 166.50,𝑝<. 001)."

Discussion¶

Related Claims¶

• Proximal goals and ability attributional feedback for early learning raise self-efficacy more than distant goals and effort feedback, by a review's report of earlier studies — the general claim this finding bears on, with the studies behind it
• Prior GenAI-solution evaluation showed no significant benefit on exam problems structurally aligned with the homework, with the midterm comparison inconclusive due to a floor effect — related
• Evaluating GenAI-generated solutions produced no significant differences on midterm scores, final exam scores, course percentages, or normalized exam change relative to solving problems directly — related
• Students who used LLMs on assignments scored significantly lower than non-users on the first midterm (large effect size), but this gap narrowed and was non-significant on the later midterms — related
• Structural paths did not differ significantly across AI usage-frequency groups, though exploratory coefficients varied numerically — related
• Students' self-efficacy for using generative AI increased over the semester, with current self-efficacy significantly higher after the fifth assignment than after the first — related
• Most students reported no change in study habits, but students who adapted their study strategies rated GenAI-evaluation assignments as significantly more helpful — related
• Students rated the LLM-based assignments as useful, challenging, engaging, and enjoyable at both the first and fifth iterations, with no significant change across the semester — related
• Group summary scores did not differ significantly between algorithm-formed and randomly formed groups on content, vocabulary, or grammar rubrics — related
• Intervention students reported significantly higher exam confidence than controls (M=4.3 vs 3.4 on a 5-point scale, d=0.76) — related