At the median cutoff, proactive and reactive tutoring show no significant difference in MAP growth, with proactive benefits increasing for students farther below the cutoff¶

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

Subclaims¶

q2 i? At the cutoff, the difference between proactive and reactive tutoring on relative MAP growth was not statistically significant (-0.33, p = .579), while reactive students at the cutoff achieved 1.74x expected growth. → Ashish Gurung 2026 q2 i? The proactive-by-prior-performance interaction was not statistically significant (-1.61, p = .247), but its magnitude suggests the proactive benefit increased by approximately 1.6x per SD below the cutoff. → Ashish Gurung 2026 (2)

Evidence¶

Ashish Gurung 2026¶

Ashish Gurung, Ge Gao, Jordan Gutterman, Danielle R. Thomas, Shivang Gupta, Lee Branstetter, Emma Brunskill, Vincent Aleven, Kenneth R. Koedinger. (2026). Improving Hybrid Human-AI Tutoring by Differentiating Human Tutor Roles Based on Student Needs. arXiv preprint. https://arxiv.org/abs/2605.11155 doi:10.48550/arxiv.2605.11155

q2 · i? · causal · r2

DiDC analysis (Equation 4) on the IK-bandwidth subsample estimating the treatment effect at the median cutoff, with prior performance standardized within grades. The proactive-reactive difference at the cutoff was "not statistically significant (-0.33, p = .579)"; equivalence was not tested.

"At the cutoff, students in the reactive group achieved 1.74×expected MAP growth (+74%, p = .007). Although proactive students at the cutoff showed numerically lower growth, this difference was not statistically significant (-0.33, p = .579)."

Ashish Gurung 2026 (2)¶

Ashish Gurung, Ge Gao, Jordan Gutterman, Danielle R. Thomas, Shivang Gupta, Lee Branstetter, Emma Brunskill, Vincent Aleven, Kenneth R. Koedinger. (2026). Improving Hybrid Human-AI Tutoring by Differentiating Human Tutor Roles Based on Student Needs. arXiv preprint. https://arxiv.org/abs/2605.11155

q2 · i? · causal · r2

DiDC heterogeneity estimate (Equation 4) on the IK-bandwidth subsample. The interaction was "not statistically significant (-1.61, p = .247)"; the authors interpret the point estimate as indicating the proactive benefit increased by approximately 1.6x per SD below the cutoff.

"The in- teraction between proactive tutoring and prior performance was also not statistically significant (-1.61, p = .247); how- ever, the magnitude of this estimate suggests a meaningful pattern."

Discussion¶

Related Claims¶

• Proactive tutoring produced larger relative MAP growth than reactive tutoring on average (75% marginally significant in the IK bandwidth) — related
• Introducing human-AI tutoring improves relative MAP academic growth by 61% over an AI-only baseline within the IK bandwidth — related
• Human-AI tutoring increases IXL time on task by 25% with no significant difference between proactive and reactive groups — related
• Human-AI tutoring is associated with a 36% rise in IXL skill proficiency, but the gap widened in favor of reactively tutored students (difference-in-differences preprint) — reports the opposite
• Proactive tutoring increases learners' use of metacognitive help tools compared with reactive tutoring — related
• Proactive tutoring increases learners' use of lexical help tools compared with reactive tutoring — related
• High dosage tutoring can produce accelerated academic growth for academically at-risk students if implemented effectively and at proper scale — a broader claim this one bears on
• Grade cutoffs are arbitrary because adjacent scores receiving different grades are not significantly different, per the author's statistical argument — related