Shallowness detectors transfer between the transfer-shallow and PFL-shallow constructs with minimal A′ degradation¶

Claim · All claims
Evidence · 1 study · associational r1 · q2
Currency · newest study 2013 · not yet checked against its sources · not reviewed by a person

Subclaims¶

q2 i? Models built to predict transfer-shallow predict PFL-shallow with A′ 0.75, and models built to predict PFL-shallow predict transfer-shallow with A′ 0.73, suggesting the two constructs are substantially related. → Gowda 2013

Evidence¶

Gowda 2013¶

Gowda, S. M., Baker, R. S., Corbett, A. T., & Rossi, L. M. (2013). Towards Automatically Detecting Whether Student Learning is Shallow. International Journal of Artificial Intelligence in Education. https://doi.org/10.1007/s40593-013-0006-4

q2 · i? · associational · r1

Cross-application analysis applying each shallowness detector to the other construct's labels. Kappa degraded more, dropping to 0.27 for transfer-shallow models predicting PFL-shallow and to 0.41 in the reverse direction, still substantially better than chance.

"When the models built to predict transfer-shallow are applied to predict PFL-shallow, they have very minimal degradation in terms of A ′, achieving an A ′ of 0.75, almost as good as a model trained to predict PFL-shallow."

Discussion¶

Related Claims¶

• K* models outperform step regression for detecting transfer-shallowness, with no-interactions and interaction models each best on different metrics — related
• K* detectors identify no-PFL-shallow learners with cross-validated A′ of 0.790 and kappa 0.55 — related
• K* detectors identify no-transfer-shallow learners at student-level cross-validated A′ of 0.766–0.791 — related