Effects of a Hybrid Immersive Virtual Reality and Conventional Football Training Program on Passing Performance, Decision-Making, and Game Performance in Undergraduate Physical Education Students

Introduction

Football passing is a perception–action skill. A player must detect the positions and movements of teammates and opponents, judge pressure and available space, select an appropriate option, and execute the pass within a limited time. Technical repetition can refine movement execution, but practice that removes the information regulating the action may not prepare learners for the uncertainty of play [1].

This distinction matters in university physical education, where students often differ widely in prior football experience and receive limited instructional time. Conventional drills are useful for repetition and feedback, yet highly predictable exercises may underrepresent visual search, temporal pressure, and tactical choice. Representative tasks are therefore needed to connect technical execution with the information on which successful decisions depend [2].

Ecological dynamics provides a theoretical basis for this approach. Skilled behaviour emerges from interactions among learner, task, and environmental constraints, and performers become attuned to affordances through repeated engagement with relevant information. Representative Learning Design similarly argues that practice should preserve the informational variables that regulate behaviour in the target setting. Both perspectives emphasize functional perception–action coupling rather than treating decision-making and movement execution as independent processes [3]

Immersive virtual reality (VR) can present standardized three-dimensional scenarios while allowing systematic manipulation of defensive pressure, passing options, and task difficulty. Compared with passive video, a head-mounted display may support active visual exploration from a first-person perspective. Its training value nevertheless depends on representativeness, response fidelity, feedback, and whether learning transfers beyond the simulated task [4, 5]

Sport research has reported changes in visual search, anticipation, reaction time, tactical awareness, and decision-making after interactive simulation. Football studies have also examined scanning and passing choices in virtual environments. However, better performance on a device-based assessment does not necessarily demonstrate improved technical execution or game transfer. Transfer is more defensible when training and assessment preserve the information–movement relations of the target activity [6, 7].

The evidence base remains heterogeneous. Many studies use small samples, short exposure periods, different hardware and software, and outcomes with limited ecological or psychometric support. Reuse of identical test stimuli can produce familiarity effects, while immediate post-testing provides little evidence about persistence. These features complicate both causal interpretation and comparison across studies [8].

Comparator design is equally important. In a substitution design, where VR replaces part of field practice, the effect estimate pertains to the hybrid curriculum as a whole. Even with equal scheduled duration, conditions may differ in content, novelty, feedback, cognitive engagement, and on-field exposure. Such a study cannot identify a VR-specific mechanism unless these co-interventions are controlled.

Undergraduate physical education students are a relevant educational population because they are developing both practical competence and knowledge of instructional methods. VR could broaden access to standardized decision scenarios within a course, but evidence from university students should not be generalized directly to trained football players [9].

Accordingly, this assessor-blinded randomized trial compared eight weeks of hybrid training—30 minutes of conventional practice plus 30 minutes of immersive VR per session—with 60 minutes of conventional practice. The outcomes were LSPT completion time, video-based decision-making accuracy and reaction time, GPAI-based game performance, post-intervention intrinsic motivation, and performance at a four-week follow-up.

The estimand was the comparative effect of assignment to the two complete training packages. The design was not intended to isolate the independent effect of VR.

Materials and Methods

Research Design

This assessor-blinded, parallel-group randomized controlled trial compared a hybrid immersive-VR and conventional football program with conventional football training alone. Outcome assessors were blinded to allocation; participants and instructors could not be blinded.

This randomized controlled trial was not prospectively registered in a public clinical trial registry. Consequently, no prospective trial-registration number or registry URL is available. The trial was reported in accordance with the CONSORT 2025 statement [10]. The CONSORT 2025 participant flow diagram is presented in Figure 1, and the completed CONSORT 2025 checklist is provided as Supplementary File S1.

Participants were allocated 1:1 to a hybrid group completing 30 minutes of field practice plus 30 minutes of VR, or a control group completing 60 minutes of field practice. Session frequency and scheduled duration were matched, but on-field exposure and training content were not equivalent.

Outcomes were assessed at baseline, immediately after the eight-week intervention, and four weeks later. Because outside football and physical activity were not documented during the interval, the final occasion is described as a follow-up rather than definitive evidence of motor-learning retention.

Participants

Undergraduate students enrolled in the compulsory football course at Universitas Negeri Medan were recruited during the 2024 academic year through classroom announcements and direct invitations. Prospective participants received written and verbal study information before enrolment.

An a priori G*Power 3.1 calculation for a two-group, three-occasion within–between interaction used f = 0.25, α = .05, power = .80, a repeated-measure correlation of .50, and ε = 1.00. The calculation indicated 28 participants; 72 were recruited to improve precision and allow for attrition.

Eligibility required age 18–23 years, enrolment in the football course, no prior sport-training experience with immersive VR, normal or corrected vision, and no lower-extremity injury during the previous six months. Exclusion criteria were vestibular or neurological disorders, severe motion sickness, a recent lower-limb injury affecting performance, structured competitive football training outside the university during the intervention, or attendance below 80%.

All participants gave written informed consent and could withdraw without academic consequences. The Universitas Negeri Medan Institutional Research Ethics Committee approved the protocol (No. 128/UN33.8/KEP/2024). Procedures followed the Declaration of Helsinki, and no financial incentive was provided.

Randomization and Blinding

After baseline assessment, an independent researcher generated a simple 1:1 random allocation sequence using Random Allocation Software. The available study records do not contain the software version, randomization seed, or documentation of additional sequence restrictions; therefore, these details could not be verified retrospectively. Allocation was concealed using sequentially numbered, opaque, sealed envelopes prepared by an independent investigator. Envelopes were opened after baseline testing by the intervention coordinator. The available records do not permit retrospective verification of the identity of the individual who retained the complete allocation list.

Participants and instructors could not be blinded to group assignment. Outcome assessors remained unaware of allocation during baseline, post-intervention, and follow-up testing, and participants were reminded not to reveal their group. Groups trained separately and were asked not to discuss intervention content with members of the other group.

Intervention

The intervention comprised three supervised 60-minute sessions per week for eight weeks (24 scheduled sessions). The groups had equal scheduled time and frequency but different amounts of field practice.

All participants completed a familiarization session. The hybrid group also received headset-safety instruction and VR practice trials; the control group received no equivalent technology-based orientation, so novelty and expectancy effects remain possible.

Hybrid VR and Conventional Training Group

Each hybrid session combined 30 minutes of conventional football practice with 30 minutes of immersive VR-based perception–action training. VR replaced half of the on-field practice rather than being added to an otherwise identical program.

The VR component was developed in Unity 2022 LTS and delivered through a Meta Quest 3 headset. Participants viewed interactive football scenarios, scanned for teammates and defenders, evaluated passing options, and selected responses under varying defensive pressure. Each session contained approximately 12–15 scenarios lasting 90–120 seconds, separated by brief recovery intervals.

The study records retained for the present analysis did not contain sufficient technical documentation to retrospectively verify the formal application name and release identifier, software developer or vendor, exact response-interface configuration, or the specific motor sequence used to execute a virtual pass. These details were therefore not reconstructed from external sources. Similarly, the retained records did not permit independent verification of the precise modality and timing of application-generated feedback, the application-specific scoring algorithm, or the exact scenario-randomization procedure.

The VR intervention was structured to increase perceptual and decision-making demands progressively across the eight-week intervention. Weeks 1–2 involved static defenders and limited passing options; Weeks 3–4 introduced moving defenders and additional passing options; Weeks 5–6 increased defensive and perceptual uncertainty; and Weeks 7–8 incorporated small-sided-game scenarios, transitions, multiple options, and greater time pressure. The corresponding training characteristics are summarized in Table 1.

A complete session-level technical-failure log was not retained in the study records. Consequently, the frequency and specific type of headset, tracking, software, or other technical failures could not be reconstructed retrospectively, and no unverified failure rate is reported. The VR component was treated as an intervention modality rather than as a separately validated measurement instrument.

Control Group

The control group completed 60 minutes of conventional football practice, including passing drills, ball control, positional movement, possession games, and small-group activities. Coaching staff, frequency, scheduled duration, and general environment were matched.

The conditions nevertheless differed in modality, task content, feedback, novelty, cognitive engagement, and on-field practice. The comparison therefore concerns the complete training packages, not an isolated VR effect.

Outcome Measures

Trained assessors who were blinded to allocation administered outcomes at baseline, post-intervention, and four-week follow-up. Participants were asked to avoid vigorous activity for 24 hours before testing. Assessor training and current-study reliability should be reported where applicable.

Football Passing Performance

Football passing performance was assessed using the Loughborough Soccer Passing Test (LSPT), with LSPT completion time prespecified as the primary outcome. Participants completed the standardized LSPT procedure after a 10-minute warm-up and one familiarization trial. Completion time was recorded in seconds, with lower values indicating faster task performance. The same testing procedure was applied at baseline, immediately after the intervention, and at the four-week follow-up.

Decision-Making Performance

Decision-making was assessed using 30 football video clips lasting approximately 5–8 seconds. Three experienced football coaches independently reviewed the scenarios and agreed on the expected tactical response for each clip. The assessment approach was informed by established video-based soccer decision-making methodologies, including studies demonstrating content, construct, and test–retest evidence for video-based decision-making assessments [11], [12]. For example, the TacticUP Video Test for Soccer demonstrated content, construct, and face validity, with test–retest reliability across individual video scenes yielding Cohen’s kappa values ranging from 0.622 to 1.00 [12]. These findings provide methodological support for the use of video-based decision-making assessment in soccer but do not constitute validation of the specific 30-clip assessment used in the present study.

Clips were presented individually on a computer monitor and stopped immediately before a critical action. Participants selected the most appropriate tactical response, and the computerized assessment recorded response accuracy and reaction time. Decision-making accuracy (%) was calculated as the percentage of correct responses across the 30 clips, whereas decision-making reaction time (ms) represented the elapsed time between video occlusion and participant response. Higher accuracy and shorter reaction time indicated better performance.

The available study documentation did not permit retrospective verification of the specific response interface, software timing procedure, clip-order procedure, use of alternate forms, or formal reliability coefficients for the assessment used in the present study. Accordingly, these measurement properties were not estimated retrospectively and no reliability coefficient from another instrument was attributed to the present assessment. Because the same assessment was administered at baseline, post-intervention, and four-week follow-up, repeated exposure and familiarity with the assessment may have contributed to longitudinal changes. This limitation was considered when interpreting the decision-making outcomes, which were therefore treated as study-specific performance indicators rather than measurements from a fully validated diagnostic instrument.

Game Performance Assessment Instrument

Game performance was assessed using the Game Performance Assessment Instrument (GPAI), with scoring focused on decision-making and skill execution during football performance. The GPAI has been used in contemporary physical-education research to assess game-performance components, including decision-making and skill execution[13]. In the present study, predefined observational criteria and standardized scoring procedures were used to obtain an overall GPAI score at each assessment occasion. Higher GPAI scores indicated better observed game-performance quality. The same assessment procedure was applied at pre-intervention, post-intervention, and four-week follow-up.

The available participant-level dataset retained only the final GPAI scores and did not include the original item-level scoring records, independent ratings from multiple observers, or assessment videos. Consequently, the exact item-level scoring process, operational coding records, and current-sample inter-rater reliability could not be reconstructed retrospectively. An intraclass correlation coefficient (ICC) with a 95% confidence interval therefore could not be calculated from the available records. We do not report a study-specific inter-rater reliability coefficient or attribute reliability estimates from other GPAI implementations to the present sample. The GPAI findings are consequently interpreted as observed performance scores under the study conditions rather than as evidence of independently verified measurement reliability or definitive transfer to unrestricted authentic match performance.

Intrinsic Motivation

Intrinsic motivation was assessed immediately after the intervention using a 7-point response scale, with higher scores indicating greater intrinsic motivation. Previous research has reported acceptable psychometric properties for IMI adaptations used in physical-education contexts [14]. The assessment was administered as a post-intervention measure and was therefore not used to evaluate change over time or improvement from baseline. An overall intrinsic-motivation score was retained in the available participant-level dataset and was analyzed as a post-intervention between-group outcome.

The available study records did not retain the original questionnaire, item-level responses, or sufficient documentation to retrospectively verify the exact IMI version, number and wording of items, subscale structure, reverse-scored items, language or translation procedures, or the original scoring instructions. Consequently, internal consistency for the present sample could not be recalculated using Cronbach’s alpha or McDonald’s omega, and subscale-level scores could not be reconstructed. The reported overall score is therefore treated as a study-specific post-intervention indicator of intrinsic motivation rather than as evidence of the psychometric validity of a particular IMI version. This limitation is acknowledged when interpreting the intrinsic-motivation findings.

VR Tolerability

After each VR training session, participants in the Hybrid VR group rated VR-related motion sickness on a study-specific 0–5 scale, with higher scores indicating greater discomfort. Three session-level ratings were obtained for each participant and averaged to derive a participant-level motion-sickness score. The resulting participant-level scores were summarized descriptively. The measure was used as a descriptive indicator of VR tolerability rather than as a comparative safety outcome.

Internal Training Load

Internal training load was estimated using session rating of perceived exertion (RPE), recorded approximately 30 minutes after each training session using the Borg 6–20 scale. Each training session lasted 60 minutes. Weekly internal training load was calculated as session RPE × session duration (min) × number of sessions completed per week. Training exposure was summarized descriptively and was not treated as an efficacy outcome.

Attendance was recorded at every session, with 80% participation specified as the minimum required for inclusion.

Four-Week Follow-Up

Performance outcomes were reassessed four weeks after the intervention. The research team provided no VR or structured football practice during this interval, but other football and physical activity were not monitored.

The same equipment, conditions, and assessors were used across occasions. This may reduce procedural variation, but repeating identical decision clips could create familiarity or memory effects.

The study lasted approximately 12 weeks: recruitment and baseline testing, eight weeks of training, immediate post-testing, and a four-week follow-up.

At each occasion, participants completed the LSPT, video decision test, and GPAI-based game assessment in a standardized order. Randomization followed baseline testing.

Both groups trained three times weekly. Attendance and session RPE were recorded after each session, although matched scheduled time did not constitute equivalent training content or field exposure.

Outcomes were repeated immediately after training and four weeks later. Outside activity during the follow-up interval was not recorded.

Statistical Analysis

All statistical analyses were conducted using IBM SPSS Statistics on the participant-level dataset. Seventy-two participants were included in the efficacy analyses. All 72 participants had complete outcome data for the longitudinal efficacy analyses; therefore, no imputation was required and no participant was excluded because of missing outcome data. Data were screened for missing values, data-entry errors, and implausible observations before analysis; no participant was excluded solely because of statistical extremeness. The prespecified primary outcome was LSPT completion time. Decision-making accuracy, decision-making reaction time, and GPAI score were prespecified as secondary performance outcomes, whereas intrinsic motivation was analyzed separately as a post-intervention secondary outcome.

Longitudinal changes in LSPT completion time, decision-making accuracy, decision-making reaction time, and GPAI were examined using two-way mixed repeated-measures ANOVA, with Group (Hybrid VR vs. control) as the between-participants factor and Time (pre-intervention, post-intervention, and retention) as the within-participants factor. The Group × Time interaction was the principal test of intervention effects, with main effects of Group and Time also reported. Only outcomes with independently verifiable scoring procedures and participant-level data were included in the inferential analyses.

Mauchly’s test was used to assess sphericity, with the Greenhouse–Geisser correction applied when this assumption was violated. Homogeneity of variance was assessed using Levene’s test. Effect sizes for the longitudinal omnibus effects were reported as partial eta squared (ηp²).

Estimated marginal means were used to report between-group differences at post-intervention and retention, including 95% confidence intervals and adjusted p-values. Pairwise longitudinal contrasts included pre–post, pre–retention, and post–retention comparisons, with Bonferroni adjustment applied to the corresponding pairwise comparisons.

Residual and outlier diagnostics were performed for each longitudinal outcome using descriptive statistics, histograms, Q–Q plots, boxplots, extreme-value listings, and Shapiro–Wilk tests. Potential extreme observations were checked against participant-level records, and no observations were excluded solely on statistical grounds.

Training adherence, session-RPE, weekly internal training load, motion-sickness scores, and adverse events were summarized descriptively. All tests were two-sided. To account for multiplicity across the four longitudinal performance outcomes, a Bonferroni-adjusted significance threshold of α = .0125 was applied to the Group × Time interaction tests.

Results

Participant Characteristics

A total of 72 randomized participants completed the post-intervention and four-week follow-up assessments and were included in the reported analyses. Mean attendance was approximately 22.9 of 24 scheduled sessions, so completion of outcome assessments should be distinguished from attendance at every intervention session. All 72 randomized participants met the ≥80% attendance criterion and had complete longitudinal outcome data; therefore, all 72 participants were included in the efficacy analyses. Missed sessions were not imputed.

Baseline demographic characteristics are presented in Table 3. Independent-samples t-tests revealed no significant between-group differences in age, height, body mass, body mass index (BMI), or football playing experience (all p > .05). Similarly, chi-square analyses demonstrated comparable distributions of sex and dominant foot between groups (p > .05). These findings indicate that the two groups were well balanced before the intervention, suggesting successful randomization and comparable baseline characteristics.

Note. Continuous variables are presented as mean ± standard deviation (SD). Categorical variables are presented as frequency (percentage). Between-group differences for continuous variables were analyzed using independent-samples t-tests, whereas categorical variables were analyzed using Pearson's chi-square tests. No statistically significant baseline differences were observed between groups (all p > 0.05).

Training Adherence and Intervention Safety

Training attendance was high in both groups. Participants in the hybrid group completed 22.89 ± 1.33 of 24 scheduled sessions (95.4 ± 5.5%), while control participants completed 22.86 ± 0.93 sessions (95.3 ± 3.9%). Internal training load was monitored using the Borg 6–20 Rating of Perceived Exertion scale. Mean session RPE was 16.03 ± 0.65 in the hybrid group and 15.93 ± 0.68 in the control group. Weekly internal training load was calculated as session RPE × session duration × three scheduled sessions per week, yielding 2885.00 ± 116.78 AU and 2866.50 ± 122.26 AU, respectively.

No intervention-related adverse events or musculoskeletal injuries were reported. The hybrid group reported a mean study-specific simulator-discomfort score of 2.47 ± 1.93. Because this measure was not validated and was not administered to the control group, it was treated as a descriptive tolerability indicator rather than a comparative safety outcome.

Note. Values are presented as mean ± standard deviation unless otherwise indicated. Attendance represents the number of completed sessions out of 24 scheduled sessions. Session RPE was assessed using the Borg 6–20 Rating of Perceived Exertion scale. Weekly internal training load was calculated as session RPE × 60 min × 3 sessions per week. Simulator discomfort was assessed only in the hybrid group using a study-specific 0–5 scale and is reported descriptively rather than as a comparative safety outcome.

Panel A. Attendance

• Hybrid group = 22.89 sessions
• Control = 22.86 sessions

Panel B. Weekly Training Load

• Hybrid group = 2885.00 AU
• Control = 2866.50 AU

Mean attendance and weekly internal training load during the eight-week intervention are displayed. Attendance and internal training load were similar between groups. Error bars represent standard deviations.

Passing Performance

Descriptive statistics and mixed repeated-measures ANOVA results for LSPT completion time are presented in Tables 5 and 6. LSPT completion time decreased across the three assessment occasions in both groups, with a greater reduction observed in the Hybrid VR group. The Group × Time interaction was significant, F(1.157, 81.008) = 438.08, p < .001, ηp² = .862. At the four-week follow-up, LSPT completion time remained lower than baseline in both groups. Between-group differences at post-intervention and follow-up, together with 95% confidence intervals, are reported in Table 8. Bonferroni-adjusted post-intervention versus follow-up contrasts are also reported in Table 8.

Decision-Making Performance

Mean decision-making accuracy increased and reaction time decreased more in the Hybrid VR group than in the control group. Because the Group × Time interaction compares the effects of the two complete training packages, these findings are interpreted as comparative intervention effects rather than isolated effects of the VR component. Given the need to further document the assessment's validity, reliability, alternate-form use, and potential repeated-exposure effects, the decision-making findings are interpreted cautiously.

Mean decision-making outcomes at follow-up remained close to post-intervention values. Bonferroni-adjusted post-intervention versus retention comparisons showed a 1.208-percentage-point decrease in decision-making accuracy (95% CI, 0.964 to 1.453; p < .001) and an 11.597-ms increase in reaction time (95% CI, 10.418 to 12.776 ms; p < .001).

Note. Data are presented as mean ± standard deviation (SD).

Note. F values for within-subject effects are reported using Greenhouse–Geisser-corrected degrees of freedom because Mauchly's test indicated significant violations of the sphericity assumption for all outcomes. Partial eta squared (ηp²) was used as the effect-size measure. Mauchly's test results were as follows: LSPT completion time, W = 0.272, χ²(2) = 89.891, p < .001, εGG = 0.579; decision-making accuracy, W = 0.321, χ²(2) = 78.391, p < .001, εGG = 0.596; decision-making reaction time, W = 0.140, χ²(2) = 135.558, p < .001, εGG = 0.538; GPAI, W = 0.218, χ²(2) = 105.169, p < .001, εGG = 0.561.

All four Group × Time interactions remained statistically significant after accounting for multiplicity across the four longitudinal outcomes using a Bonferroni-adjusted significance threshold of α = .0125 (all p < .001).

Game Performance

GPAI-based scores increased more in the hybrid group than in the conventional group, with a reported Group × Time interaction. Interpretation as transfer to representative game performance remains conditional on complete reporting of the game protocol, scoring algorithm, assessor blinding, and current-study inter-rater reliability.

Note. Mean differences are calculated as VR − Control. Negative values indicate lower values in the VR group; therefore, negative differences represent faster LSPT completion and shorter decision-making reaction time, whereas positive differences represent higher decision-making accuracy and GPAI scores in the VR group. Confidence intervals and p-values are Bonferroni-adjusted. Comparisons were based on estimated marginal means from the mixed repeated-measures ANOVA.

Note. Mean differences are based on estimated marginal means. Pairwise comparisons were adjusted using the Bonferroni procedure.

Bonferroni-adjusted pairwise comparisons showed significant differences between all assessment occasions for each outcome (all adjusted p < .001). From post-intervention to retention, LSPT completion time increased by 0.746 s (95% CI, 0.651 to 0.840 s), decision-making accuracy decreased by 1.208 percentage points (95% CI, 0.964 to 1.453), decision-making reaction time increased by 11.597 ms (95% CI, 10.418 to 12.776 ms), and GPAI decreased by 0.095 points (95% CI, 0.080 to 0.110).

Intrinsic Motivation

At post-intervention, the hybrid group had a higher overall intrinsic-motivation score than the control group (6.41 ± 0.30 vs. 5.28 ± 0.43). Welch’s independent-samples t-test indicated a significant between-group difference, t(62.60) = 13.15, p < .001, with a Cohen’s d of 3.08. Because intrinsic motivation was assessed only after the intervention, this result represents a post-intervention between-group difference and does not indicate within-participant change over time.

Note. Values are presented as mean ± standard deviation (SD). Welch's independent-samples t-test was used because Levene's test indicated unequal variances (F = 12.41, p = .001). Cohen's d is reported as the standardized between-group effect size.

Mean ± SD values for (A) LSPT completion time, (B) decision-making accuracy, (C) decision-making reaction time, and (D) GPAI score at pre-test, post-test, and four-week follow-up. The Hybrid VR group received 30 minutes of conventional football practice plus 30 minutes of immersive VR training per session, whereas the Control group received 60 minutes of conventional football practice per session. The four-week follow-up represents the assessment conducted four weeks after completion of the intervention.

Discussion

This assessor-blinded randomized trial compared a hybrid curriculum of conventional football practice and immersive VR with time-matched conventional training. Reported trajectories favored the hybrid group for passing-related, decision-making, and GPAI-based outcomes, and its post-intervention IMI score was higher. These findings apply to the complete hybrid package, not the independent contribution of VR, because the groups also differed in content, feedback, novelty, cognitive demands, and on-field practice [15], [16].

The LSPT completion-time outcome showed a different trajectory between the hybrid and conventional training packages, with greater reductions observed in the hybrid group. Because the LSPT completion time represents the time required to complete the standardized passing task, the finding indicates faster task completion following the hybrid program [17]. However, this result should be interpreted as a comparative effect of the complete training packages rather than evidence of an independent VR effect.

The video-based outcomes showed large differences in accuracy and reaction-time trajectories. Repeated exposure to representative situations may have improved responses to similar scenarios [18]. However, coach agreement establishes content relevance rather than validity, reliability, or responsiveness. If the same clips and sequence were used at all assessments, memory and procedural familiarity may have contributed to improvement, particularly given the exceptionally large effects. These outcomes should remain exploratory unless validated alternate forms, timing accuracy, test–retest reliability, and safeguards against repeated-test learning are documented.

The GPAI-based outcome provides an observational measure of game-performance behavior beyond the computerized assessments [19], [20]. Higher scores indicate better performance according to the scoring procedures used in the present study. However, the available records do not provide sufficient detail to establish the extent to which these scores represent transfer to authentic football performance [21]. In addition, independent ratings were not retained, preventing estimation of inter-rater reliability in the present sample. Therefore, the GPAI findings should be interpreted as changes in observed game-performance scores within the study context rather than as definitive evidence of transfer to authentic match performance.

Mean values at four-week follow-up remained close to post-intervention values in the hybrid group. This is compatible with short-term persistence, but similarity of means and an omnibus interaction do not formally test maintenance. The adjusted longitudinal analyses showed that both groups changed significantly across the three assessment occasions, while the significant Group × Time interactions indicated that the pattern of change differed between the hybrid and control conditions. The estimated between-group differences at post-intervention and four-week follow-up, together with the adjusted pairwise contrasts and 95% confidence intervals, provide a more complete description of the magnitude and direction of these changes. Because football and physical activity during the interval were not monitored, this occasion is better described as follow-up than proof of permanent motor learning [22], [23].

The post-intervention IMI comparison also requires restraint. The hybrid group reported a higher overall score, but motivation was measured only once; consequently, the study cannot demonstrate that motivation increased from baseline. The difference may reflect the training experience, novelty, expectations, or unmeasured pre-existing differences. Because the original IMI questionnaire, item-level responses, and complete instrument documentation were not retained, the exact IMI version, item composition, subscales, reverse-scored items, scoring structure, and internal consistency of the present sample could not be independently verified. Therefore, the IMI finding should be interpreted as a study-specific post-intervention motivational indicator rather than definitive evidence from a fully psychometrically verifiable IMI measurement.

From an applied perspective, the findings suggest that replacing half of a conventional session with the evaluated VR activities may be feasible in an undergraduate course and may yield different performance trajectories from a fully conventional session [4]. This is not equivalent to showing that VR is effective as an add-on or that it should generally replace field practice. The hybrid group received less time for physical ball contact and more exposure to standardized perceptual scenarios. Decisions about implementation should therefore consider the instructional objective, access to equipment, technical support, adverse symptoms, and the opportunity cost of reduced on-field practice.

Several limitations materially affect confidence in the estimates. The active comparator does not isolate a VR-specific effect. The video-based decision test and GPAI protocol require additional documentation of validity and reliability; independent ratings were not retained, preventing estimation of current-sample inter-rater reliability for the GPAI. Motivation was measured only after training, and the original IMI questionnaire and item-level responses were unavailable for retrospective psychometric verification. External activity during follow-up was not monitored, and the large effect sizes observed for the longitudinal outcomes should be interpreted in relation to the substantial within-participant changes and the corresponding residual variation. Importantly, the reported statistics were derived from the participant-level analysis and are presented together with the Greenhouse–Geisser correction, adjusted pairwise comparisons, and confidence intervals to provide a transparent account of the observed effects. The revised analytical reporting incorporates the prespecified primary outcome, analysis population, missing-data handling, multiplicity adjustment, Greenhouse–Geisser correction where required, adjusted pairwise comparisons, estimated group differences, effect sizes, and 95% confidence intervals. These reporting elements strengthen the transparency of the longitudinal analysis and allow the magnitude and uncertainty of the observed effects to be evaluated more directly. Finally, the single-institution sample of physical education students limits generalizability to trained players and other settings.

In summary, the reported data favor the hybrid curriculum over the time-matched conventional curriculum on several study outcomes. The clearest defensible inference is a comparative effect of two training packages, not a specific causal effect of VR [24]. Publication-quality interpretation requires verification of the RPE scale and workload calculation, complete documentation of outcome reliability, and independent verification of the statistical results. A future prospectively registered trial should use validated outcomes, alternate test forms, an attention-matched comparator, transparent data and analysis code, and longer follow-up in football-trained populations.

Conclusions

The hybrid immersive-VR and conventional football training program was associated with more favorable trajectories in LSPT completion time, decision-making accuracy, decision-making reaction time, and GPAI scores than conventional football training alone. The significant longitudinal effects and Group × Time interactions, together with the adjusted between-group comparisons and post-intervention versus four-week follow-up contrasts, indicate that the hybrid approach was associated with a distinct pattern of performance change across the assessment period. The hybrid group also reported higher post-intervention intrinsic motivation, while the observed RPE values indicated comparable perceived exertion between groups. VR-related discomfort was generally low to moderate and was summarized descriptively because the study-specific measure was not a validated comparative safety instrument. Performance differences between groups were still observed at the four-week follow-up; however, football and physical activity during the interval were not monitored, limiting attribution of these differences to the intervention. Overall, the findings support further evaluation of hybrid immersive-VR and conventional training as a time-matched option in undergraduate football instruction.

Declarations

Ethics Approval and Consent to Participate
Ethical approval was obtained from the Research Ethics Committee of Universitas Negeri Medan (No. 128/UN33.8/KEP/2024). All study procedures were conducted in accordance with the principles of the Declaration of Helsinki. Participants received written and verbal information about the study before enrolment, and written informed consent was obtained from all participants before participation. Participants were informed that participation was voluntary and that they could withdraw from the study without academic consequences. No participants were younger than 18 years; therefore, parental or legal-guardian consent was not required.

Funding
This research received no external funding.

Role of the Funder
The study received no external funding; therefore, no external funder had any role in the study design, data collection, data analysis, interpretation of the findings, manuscript preparation, or decision to submit the manuscript for publication.

AI Use Statement

During the revision of this manuscript, the authors used OpenAI’s ChatGPT to assist with English-language editing, improve clarity and readability, and support the formatting and graphical presentation of the results. The tool was not used to generate or modify research data, perform statistical analyses, fabricate references, or determine the scientific conclusions of the study. All AI-assisted content was critically reviewed, verified, and revised by the authors, who take full responsibility for the accuracy, integrity, and final content of the manuscript.

Conflict of Interest
The authors declare that they have no competing financial or non-financial interests that could have influenced the conduct, analysis, or reporting of this study.

Data Availability
The raw participant-level dataset and full SPSS output files are retained by the authors. De-identified data and relevant statistical output may be made available upon reasonable request, subject to ethical and institutional requirements. No original SPSS syntax file was retained.