Author ORCID Identifier
Brahim Es-Sabery: https://orcid.org/0009-0004-3594-5407
Abstract
Recommender systems for university electives are trained on records of what students chose, which is not the same as a record of what they were assigned. We examine what that difference does to the estimated relationship between academic unfamiliarity and elective performance, using eight academic years of enrolment records from one university (4,568 students, 5,591 courses, 208,949 selections). Administrative boundary crossing is measured only from strictly prior terms. Before adjustment, electives in a novel academic unit are associated with a within-course standardized score +0.161 SD above familiar electives. After student-grouped cross-fitted augmented inverse-probability weighting, the same comparison is -0.157 SD (95% CI [-0.189, -0.125]). The familiar group has a weighted effective sample size of only 611, so we take the novel–moderate contrast of -0.059 SD (95% CI [-0.089, -0.029]) as the better-supported estimate. Even this contrast would be explained away by an unmeasured confounder accounting for 5.9% of the residual variance of both treatment and outcome. Raw scores and student fixed effects are inconclusive, administrative and semantic distance correlate at r = .43, and the semantic measure does not reproduce the pattern. A temporal-holdout optimizer stress test produces implausible extrapolative scores. We conclude that these data do not establish an absolute learning penalty and do not support a deployable exploration policy; they show instead how fragile observational evidence for course recommendation can be.
First Page
19
Last Page
41
Ethics Approval
No
Declaration Statement
N/A
Recommended Citation
Es-Sabery, B., Moumane, A., Ouhda, M., & Baslam, M. (2026). When course-recommendation evidence reverses: Selection, overlap, and the measurement of academic unfamiliarity. Journal of Educational Technology Development and Exchange (JETDE), 19(4), 19-41. https://doi.org/10.18785/jetde.1904.02
Creative Commons License

This work is licensed under a Creative Commons Attribution-NonCommercial-No Derivative Works 4.0 International License.
Included in
Curriculum and Instruction Commons, Educational Assessment, Evaluation, and Research Commons, Educational Technology Commons