Use predictions to open doors
Could a prediction model identify students missing a worthwhile STEM opportunity? Researchers trained models on 282 German Physics Olympiad participants, then estimated first-round success potential for 1,103 nonparticipants in academic-track secondary schools. At the selected threshold, 43 nonparticipants were flagged as potentially successful. Thirty-two had already entered another selective STEM competition; eleven had no such prior participation. The finding complicates the idea that every missing entrant represents undiscovered talent: many were already engaged elsewhere. Prediction quality was modest, with a cross-validated area under the curve of .605 for the chosen model. These are estimates of unobserved outcomes, not proof that the selected students would advance or that others lack potential.
Why it matters for leaders
Ask enrichment leaders to examine who receives invitations, transport and preparation, including students without prior competition experience. Use any analytics to offer additional opportunities, with teacher and student input. Do not let an uncertain score become an eligibility barrier; track uptake and success after outreach.
The evidence
The preprint compared three classifiers using repeated nested cross-validation and 31 indicators. A regularized regression model was selected. The threshold was calibrated using participants and then transferred to nonparticipants.
Who was studied
German academic-track secondary schools; competition access, not U.S. district validation.
What the evidence cannot tell us
Selection and population differences may weaken that transfer. Threshold choice changes the identified group; actual competition outcomes for nonparticipants are unavailable.
Read the original sources
Summary and leadership analysis from the Week ending 2026-10-02 brief. The edition’s selection notes explain source access and evidence limitations.