Virtual Reality Titration Training Shows No Clear Advantage Over Real Lab Work in Pilot Study

Students in a mandatory first-semester Life Sciences course who began titration training in immersive virtual reality before moving to the physical laboratory did not show a clear knowledge advantage over classmates who completed the two activities in the reverse order, according to a pilot study published in Frontiers in Virtual Reality. The research team, led by Urs Chalupny of the Centre de Formation Professionnelle Neuchâtelois and colleagues at the University of Applied Sciences and Arts Northwestern Switzerland, tracked ninety-one students through a combined sequence of virtual and hands-on titration practice and found that knowledge-test scores rose after the first activity but changed little after the second, with no statistically detectable difference between the two starting sequences.

The study was embedded in the foundational module Practical Basic Laboratory Techniques at the University of Applied Sciences Northwestern Switzerland, a course that all first-semester students must complete. Ninety-two students initially took part, but one participant discontinued the virtual reality condition after experiencing discomfort and was removed from the sample, leaving ninety-one students who completed every stage. Of these, fifty-seven identified as female, thirty-three as male, and one as non-binary, with a mean age of 21.82 years. Most participants had little or no prior experience with VR headsets: thirty-four had never used one before, forty-two reported very limited experience, thirteen had used a headset multiple times with some support, and only two described themselves as experienced users.

Students worked in self-selected pairs and were assigned to a starting condition by their instructors rather than by the research team, a design choice the authors emphasize makes the comparison observational rather than causal. One group began with the virtual titration and continued in the physical laboratory; the other reversed that order. The virtual environment, developed in Unity for the Meta Quest 2 headset, replicated a real laboratory fume hood complete with a burette, measuring cylinder, pipette, beakers, funnel, various solutions, a magnetic stirrer, and a pipette bulb. Tutorial videos were integrated into the virtual space, and a SharePoint interface allowed students to record measured volumes in an Excel spreadsheet. The simulation unfolded in three phases: an orientation period with no time constraints, a guided titration in which text instructions on the fume hood wall and visual cues such as yellow outlines and flashing indicators highlighted the relevant objects, and a test phase in which students repeated the same procedure with an unknown solution and no additional guidance.

The physical laboratory condition mirrored the virtual one closely. Students received printed step-by-step instructions, worked in pairs at benches arranged inside a fume hood in the same configuration as the virtual environment, recorded measured volumes on a protocol sheet, and calculated the concentration of the unknown substance under instructor supervision. Instructors intervened only for safety issues or when students requested help. Before the main study, a separate usability test with twelve participants had yielded a System Usability Scale score of 68, prompting revisions to the virtual environment, including a controller-based guidance system in which grip and trigger buttons flashed to indicate graspable objects, a visual cue for opening lidded containers, a reset button for unreachable objects, short tutorial videos, an enlarged display of the burette fill level, and an extended desktop application allowing instructors to mute voice chat and trigger the reset function remotely.

Knowledge was measured with a thirteen-item multiple-choice test developed by the course instructors and aligned with the module’s learning objectives. Items covered declarative knowledge such as required chemicals and endpoint recognition, procedural knowledge such as the correct sequence of titration steps and proper burette filling technique, and problem-solving items requiring stoichiometric calculations. Scoring awarded one point for each correctly selected option and one point for each correctly unselected distractor, yielding zero to four points per item and a maximum total of fifty-two. At baseline, the ninety-one students averaged 39.80 points (SD = 4.02), or 76.5 percent of the maximum, indicating substantial prior knowledge on this option-level scoring metric. After the first activity, the mean rose to 42.87 (SD = 4.47), an increase of 3.07 points, and after the second activity it reached 43.18 (SD = 4.35), an additional gain of only 0.31 points.

The statistical analysis used a baseline-adjusted linear mixed-effects model with participant random intercepts, treating each student’s two post-activity observations as linked. In the first period, before any crossover, the adjusted difference between the VR-first and laboratory-first groups was −1.26 points (95% CI [−2.74, 0.22], p = 0.094), meaning the VR-first group scored slightly lower on average but the interval spans zero. The time-by-sequence interaction, which tested whether the two sequences differed in their additional change from the first to the second post-test, was 0.76 points (95% CI [−0.42, 1.94], p = 0.204), also imprecisely estimated. After both activities, the adjusted VR-first minus laboratory-first contrast was −0.50 points (95% CI [−1.98, 0.98], p = 0.502). A participant-level Hotelling test of the joint null that both mean changes from baseline equal zero was significant (F(2, 89) = 36.91, p < 0.001), and paired comparisons showed a t2 minus t1 increase of 3.07 points (95% CI [2.31, 3.83], Holm-adjusted p < 0.001) and a t3 minus t1 increase of 3.37 points (95% CI [2.54, 4.21], Holm-adjusted p < 0.001), but the additional t3 minus t2 change of 0.31 points was not statistically detectable (95% CI [−0.28, 0.90], Holm-adjusted p = 0.303).

The authors stress that these descriptive score increases combine exposure to the activities with repeated testing. Without an untreated comparison group, the gains cannot be separated from retest effects or other time-related influences, and adjusting for period and sequence does not by itself identify an isolated modality effect. Because starting modality coincides with sequence at the first post-test and switches at the second, the coefficients cannot be interpreted as independently randomized, context-free modality effects. The first-period comparison precedes crossover and is therefore not contaminated by exposure to the second activity, but it still shares the limitations of the initial non-random allocation. No equivalence or non-inferiority margin or test was specified, so the wide confidence intervals, which include differences in either direction, cannot establish equivalence or rule out educationally meaningful differences.

A further constraint on interpretation comes from the knowledge test’s limited reliability. Raw-score Cronbach’s alpha was 0.349 at baseline, 0.579 after the first activity, and 0.587 after the second, with corresponding bootstrap intervals of [0.134, 0.499], [0.435, 0.678], and [0.418, 0.693]. The authors note that limited reliability constrains precision and can attenuate associations with other variables, meaning non-significant comparisons cannot rule out educationally meaningful differences. They also caution that the alpha coefficients do not by themselves establish unidimensionality or validity, and that differences between point estimates across administrations should not be read as a demonstrated developmental change in reliability. The test measures option-level decision accuracy rather than the number of entirely correct items, and it cannot establish whether practical laboratory skills improved or transferred between environments.

The study also explored relationships within the Cognitive Affective Model of Immersive Learning using an exploratory reduced path model with bootstrap confidence intervals. In this model, presence was negatively associated with cognitive load (β = −0.469, bootstrap 95% CI [−0.628, −0.344]), the only path whose interval excluded zero, explaining approximately 22 percent of the in-sample variance in the cognitive-load construct. The remaining paths—technology affinity to presence, presence to motivation, cognitive load to knowledge, and motivation to knowledge—all had intervals that included zero. In-sample R² was 0.220 for cognitive load, 0.045 for motivation, 0.033 for presence, and just 0.022 for the knowledge score, describing limited explanatory power, particularly for the learning outcome. The authors emphasize that these observational questionnaire associations do not establish that increasing presence reduces load or improves learning, and that both constructs were measured by self-report after the VR experience, so the statistical association does not establish temporal direction.

Usability of the revised virtual environment averaged 71.75 on the System Usability Scale (SD = 13.88, n = 90 complete questionnaires), a level the authors describe as consistent with acceptable or good perceived usability. The authors frame the work as a pilot implementation, the first iteration of the VR titration activity within the curriculum, and note several limitations: differential waiting times before the post-VR knowledge assessment may have introduced memory decay effects, instructor support was not uniform across VR sessions, group assignments were not fully randomized, and the second-period assessment follows both modalities without a washout, so carryover cannot be separated from the second activity’s contribution. They conclude that the findings describe knowledge-score increases during a combined training sequence but do not establish equivalent learning outcomes, a causal modality effect, or procedural skill transfer, and that future evaluations should specify educationally meaningful differences in advance and include validated procedural-performance outcomes alongside knowledge assessments.

Subject of Research: Sequencing immersive virtual reality and physical laboratory work for titration training in a first-semester Life Sciences course

Article Title: Sequencing immersive virtual reality and physical laboratory work for titration training: effects on learning outcomes

Article References: Chalupny, U., Meier, P., Schärer, C., Varòn Silva, D., & Christ, O. (2026). Sequencing immersive virtual reality and physical laboratory work for titration training: effects on learning outcomes. Frontiers in Virtual Reality, 7, Article 1958188. https://doi.org/10.3389/frvir.2026.1958188

Image Credits: AI Generated

DOI: 10.3389/frvir.2026.1958188

Keywords: immersive virtual reality, titration, laboratory training, learning outcomes, cognitive load, presence, CAMIL, pilot study, chemistry education, knowledge test, sequence effect, usability