Volunteers using an immersive 3D graph to navigate engineering information took significantly longer than those viewing a conventional slide presentation, according to a new preprint. On the measured retrieval task, the two groups had the same median error score, while recall and confidence also showed no significant differences.
The study asked whether an interactive, layered 3D graph in virtual reality could make model-based digital threads easier to navigate and understand than slide-based presentations. Its findings concern a focused comparison of two presentation setups and the specific outcomes measured in the experiment.
The two ways of exploring the model
The researchers ran a between-subjects controlled comparison: each participant used only one setup. One group worked with GraphXplore, a layered 3D graph presented in a virtual-reality headset with interactive controls. The comparison group viewed a PowerPoint slide deck containing model screenshots.
The experiment enrolled 33 volunteers, mainly industrial product-design postgraduates and professors. The researchers recorded objective measures such as errors, recall and completion time, as well as participants’ ratings of confidence, usability and cognitive workload.
Participants were assigned at random, and the study had local ethics approval and registration before it began. The researchers checked for possible outliers and tested whether the results met normality assumptions. They used Mann-Whitney U tests for non-parametric or ordinal outcomes and an independent-samples t-test for normally distributed completion time.
Speed was the clearest difference
The strongest result concerned completion time. The mean was 15.471 for the 3D Graph condition, compared with 11.375 for Slides. The reported test gave p < .001 and a large standardized difference, Cohen’s d = 1.348, favoring the slide condition.
The study also reported an efficiency comparison, but that difference was not statistically significant. Median efficiency was 0.080 for the 3D Graph condition and 0.199 for Slides, with p = .081 and a reported effect size of r = .310.
The timing result should be read as a comparison between complete setups. The experiment did not test the layered graph separately from the virtual-reality display and interaction system, so the reported difference cannot be attributed to the graph representation alone.
Accuracy and memory showed no clear edge
On the error measure, both conditions had a median score of 0. The difference was not significant: the Mann-Whitney U statistic was 132, with p = .901 and a small reported effect size of r = .031. For this measured retrieval task, the graph and slides produced comparable error scores.
Recall scores were similarly close. The median was 11 for the 3D Graph condition and 10 for Slides, but the difference was not significant, with p = .465 and r = .132. The result applies to the recall measure used after the exploration task.
Participants’ confidence ratings did not differ significantly either. During retrieval, median confidence was 18 for the 3D Graph condition and 16.5 for Slides. After the recall task, the medians were 7 and 10. Neither comparison was significant: p = .260 for retrieval confidence and p = .110 for recall-task confidence.
Higher usability ratings did not produce a significant gap
Usability ratings leaned toward the immersive graph in the descriptive results. The median System Usability Scale score was 77.5 for the 3D Graph condition and 67.5 for Slides. But the difference was not statistically significant, with p = .382 and a reported effect size of r = .015.
Reported workload was also close. Median NASA-TLX scores, the study’s measure of perceived cognitive workload, were 47.7 for the 3D Graph condition and 44.5 for Slides. The difference was not significant, with p = .510 and r = .120.
The abstract gives a separate descriptive summary in which the virtual environment had an average SUS score of 73.1, compared with 66.4 for slides. It reports mean NASA-TLX scores of 43.8 for GraphXplore and 50.4 for the traditional setup. Those averages are descriptive and do not alter the non-significant inferential comparisons reported for the outcomes.
A focused comparison, not a verdict on design reviews
Taken together, the results offer no measured evidence of better accuracy, recall or confidence with the immersive graph, while completion time favored Slides. The usability and workload measures were descriptively more favorable in some summaries, but the reported between-group tests did not establish significant differences.
The sample was made up mainly of industrial product-design postgraduates and professors. That makes the experiment useful for a controlled human-computer interaction comparison, but the results do not by themselves establish how the setup would perform for every engineering audience or every model-navigation task.
The comparison also tested specific measured outcomes—errors, recall, completion time, confidence, usability and workload. It therefore leaves open whether different tasks, especially ones requiring more extensive navigation or deeper understanding of linked information, would produce the same pattern.
Because the graph, display and interaction method were bundled into the immersive condition, another experiment would be needed to identify which element, if any, accounts for the differences. The current comparison cannot isolate the effect of the layered 3D graph from the rest of the setup.
For the task tested, the preprint is best read as an exploratory benchmark rather than evidence that virtual reality improves industrial design-review effectiveness. It shows a clear timing difference and several non-significant comparisons, but it does not demonstrate a broad performance advantage for GraphXplore over slides.
The paper reports that its data, including the SPSS statistical analysis file, are available from the corresponding author on request. No funding statement is reported in the supplied text or metadata.
Paper data and sources
Original title: Navigating and Retrieving Information in Immersive Model-Based Design Reviews: An Exploratory Study
Authors: Victor Romero, Romain Pinquié, Frédéric Noel
Journal/Repository: arXiv
Status: Preprint, not yet peer-reviewed
First online: 2026-08-20
DOI: Not available
Original paper · Full text