Authors
Yu Deng, Qingwen Zhang
Published in
Complementary therapies in medicine. Pages 103478. Nov 15, 2026. Epub Nov 15, 2026.
Abstract
An indicator can appear prominent in a fitted model without its wider domain adding distinct predictive information. We examined both repeated indicator prominence and domain-level incremental prediction of later adolescent physical activity beyond recent participation.
Four assessments from 3,755 Grade 7-8 adolescents at nine schools supported two approximately 3-month prediction windows: June-September (across summer vacation) and September-December. Participant-blocked rolling validation kept each adolescent in one fold and used earlier transitions to predict later activity. Elastic net and XGBoost compared 12 candidate characteristics in five domains with recent-activity-plus-sex baselines.
The recent-activity-plus-sex linear comparator explained 55.4% and 64.9% of held-out variance. Adding all characteristics increased elastic-net R² to.588 and.665 (absolute gains.034, 95% CI [.025,.043], and.017 [.008,.026]). Removing the motivation domain reduced R² across the two windows by.037 and.024 for elastic net and.039 and.026 for XGBoost (2.4-3.9 percentage-point losses). Intentions, beliefs about capabilities, and affective appraisal were recurrent prominent indicators; removing the wider affective domain changed R² by approximately zero.
Motivation-related measures contributed domain-level predictive information in both windows, whereas prominent affective appraisal did not imply a comparable contribution from its wider domain. Independent school-based replication should assess generalisability, calibration, assessment burden, and decision utility.
PMID:
43035191
Bibliographic data and abstract were imported from PubMed on 16 Nov 2026.
Read full publication at:
Please sign in
to see all details.
Advertisement
Stats
- Recommendations n/a n/a positive of 0 vote(s)
- Views 6
- Comments 0