Development of a virtual reality model for STEM education
TL;DR
Given the single-group design and concurrent measurement of the SEM constructs, the findings indicate within-cohort temporal changes and theoretically consistent associations rather than causal effects of VR.
Abstract
Immersive virtual reality (VR) enables interactive three-dimensional representations of abstract scientific phenomena, but its educational value depends on how it is integrated into instructional practice. This study developed and longitudinally evaluated a VR-supported instructional model for undergraduate electromagnetism. The baseline cohort comprised 193 students from two physics-related programmes; 187 completed the immediate post-test (T2), and 181 completed the two-month follow-up (T3). The intervention consisted of four 40–50-minute instructional sessions delivered over two weeks, with approximately 5–10 min of active VR interaction per participant during each session. The instructional sequence combined guided exploration, manipulation of a current-carrying loop in a magnetic field, prediction tasks, instructor-facilitated discussion, and structured reflection. Repeated-measures analyses based on 181 complete cases showed significant changes over time in electromagnetism achievement, cognitive processing, learning motivation, and perceived learning outcomes (all p < .001; partial η p 2 = .385–.465 ) . Cognitive processing, learning motivation, and perceived learning outcomes remained relatively stable from T2 to T3, whereas electromagnetism achievement showed a small descriptive decline while remaining above its baseline mean. Structural equation modelling using T2 data ( n = 187) showed positive associations between perceived visualisation quality and cognitive processing ( β = .61), cognitive processing and learning motivation ( β = .57), and learning motivation and perceived learning outcomes ( β = .63). The sequential indirect association between perceived visualisation quality and perceived learning outcomes through cognitive processing and learning motivation was also significant [ β = .22, 95% CI (.142,.319), p < .001]. Given the single-group design and concurrent measurement of the SEM constructs, the findings indicate within-cohort temporal changes and theoretically consistent associations rather than causal effects of VR.