Virtual Labs Beat Physical Ones for Inquiry Skills, Major Meta-Analysis Finds

For two decades, science educators have argued about whether students learn to think like scientists better by manipulating real glassware and circuits or by clicking through computer simulations. A new meta-analysis published in BMC Psychology offers the most statistically rigorous answer yet: virtual experiments, on average, give students a meaningful edge in scientific inquiry skills over physical experiments, but the advantage is neither universal nor automatic. The study, conducted by Hongwei Wu of the College of Educational Science and Zhengdong Zhang of the College of Mathematics and Statistics at Northwest Normal University in Lanzhou, China, synthesized experimental and quasi-experimental research published between 2000 and 2025, and its findings are already being discussed as a turning point in the debate over simulation-based science instruction.

The researchers followed the PRISMA 2020 reporting guidelines and systematically searched the empirical literature, ultimately including 21 experimental and quasi-experimental studies that yielded 41 effect sizes drawn from a combined sample of 3,931 students. Rather than using a conventional meta-analytic approach, the team employed a three-level meta-analysis model, a statistical framework designed to correct for dependency bias that arises when a single study contributes multiple nested effect sizes. Because many intervention studies measure several inquiry outcomes or compare several treatment groups, effect sizes drawn from the same paper are not statistically independent, and ignoring that structure can artificially inflate the precision of pooled estimates. The three-level model partitions variance into within-study and between-study components, giving a more honest picture of how much the true effect varies across contexts.

The headline result is striking in its clarity. Compared with physical experiments, virtual experiments produced a significant small-to-moderate positive effect on students’ scientific inquiry skills, with a pooled Hedges’ g of 0.454 that reached statistical significance. In practical terms, an effect of this magnitude means that a student at the 50th percentile of inquiry skill in a physical-experiment condition would, on average, move to roughly the 67th percentile after a virtual-experiment intervention. For a field where educational interventions often produce effects hovering near zero, that is a substantial signal, and it held up even after the authors accounted for the statistical dependence among effect sizes sharing the same source studies.

Yet the analysis refuses to tell a simple story. When the researchers examined four potential moderators, educational level, discipline type, intervention duration, and the dimension of inquiry skill being measured, they found no significant overall moderating effects. In other words, the data did not support the claim that virtual experiments work dramatically better for one age group or subject than another in a statistically definitive sense. The heterogeneity hiding beneath that null result, however, became visible in the exploratory subgroup analyses, which revealed where the advantage concentrates and, just as importantly, where it disappears.

The subgroup findings sketch a revealing map of the effect. Virtual experiments showed a significant positive effect at the senior high school level, with a Hedges’ g of 0.690, and in physics disciplines, where the pooled effect reached 0.603. Interventions lasting four to eight weeks also produced a significant advantage, with a Hedges’ g of 0.590, suggesting that the benefit emerges when students have enough time to move past the novelty of the interface and engage repeatedly with the simulated phenomena. Shorter interventions did not show the same reliable advantage, hinting that virtual laboratories may need a settling-in period before their pedagogical payoff materializes.

The most consequential exception concerns the design of controlled experiments, one of the core dimensions of scientific inquiry. For that specific skill, the analysis found no significant advantage for virtual experiments, with a Hedges’ g of just 0.056, statistically indistinguishable from zero. Designing a controlled experiment requires reasoning about variables, confounds, and fair tests, and the data suggest that this particular cognitive achievement is not something simulations confer more readily than hands-on apparatus. It is the one place in the analysis where the traditional laboratory appears to hold its ground completely, and the authors treat this asymmetry as a clue to the underlying mechanism rather than an anomaly.

To explain the pattern, Wu and Zhang turn to what they call a cognitive-embodied dynamic balance framework, which assigns complementary functions to the two modes of experimentation. Virtual experiments, on this account, excel at optimizing cognitive load. A well-designed simulation strips away the procedural noise of real laboratories, the fumbling with equipment, the measurement errors, the safety constraints, and directs the learner’s limited working memory toward the conceptual structure of the phenomenon. Physics, with its abstract and often invisible constructs like fields, forces, and charge, benefits disproportionately from this streamlining, which may explain the discipline-specific effect. Senior high school students, who possess enough prior knowledge to exploit a simulation’s abstraction but still benefit from its scaffolding, sit at a developmental sweet spot for the same reason.

Physical experiments, by contrast, derive their power from embodied experience. Handling real materials grounds abstract concepts in sensorimotor activity, and the unpredictable friction of the physical world forces students to confront measurement uncertainty in a way no simulation can replicate. The finding that designing controlled experiments shows no virtual advantage fits this framework neatly: planning a fair test may depend on embodied, material reasoning about how variables actually behave, which physical apparatus supports directly. The two approaches, the authors argue, are not rivals but complements, each optimizing a different half of what it means to do science, and the practical implication is that curricula should sequence them deliberately rather than treat the choice as either-or.

The study’s methodological choices strengthen its claim to settle a contested question. By restricting inclusion to experimental and quasi-experimental designs, the authors filtered out the correlational studies that have muddied earlier syntheses, and the three-level model’s explicit handling of nested effect sizes addresses a known weakness in previous meta-analyses of educational technology. The 25-year publication window captures the full arc of simulation development, from early desktop applets to the sophisticated interactive environments in use today, though the authors acknowledge that the modest number of included studies, 21 in total, means the subgroup results should be read as exploratory rather than definitive. The research received no external funding, and the authors declare no competing interests.

For educators and policymakers watching the rapid digitization of science classrooms, the message is both encouraging and cautionary. Virtual experiments can genuinely improve students’ capacity to formulate questions, analyze data, and draw evidence-based conclusions, particularly for adolescents tackling physics over multi-week units. But the near-zero effect on designing controlled experiments is a reminder that the embodied, material dimension of laboratory work carries instructional value that pixels cannot fully replace. The most defensible reading of the evidence is not that simulations should displace physical laboratories, but that the two should be deployed in deliberate combination, with simulations shouldering the cognitive-load-heavy work of visualization and repeated experimentation, and physical apparatus anchoring the embodied understanding of how real variables behave. As schools worldwide invest heavily in digital science education, this analysis provides the quantitative footing that debate has lacked, and it reframes the question from whether to go virtual to how to balance the virtual and the real in a single, coherent inquiry curriculum.

Subject of Research: The effect of virtual versus physical experiments on students' scientific inquiry skills

Article Title: The impact of virtual experiments on students’ scientific inquiry skills: evidence from a three-level meta-analysis

Article References: Wu, H., & Zhang, Z. (2026). The impact of virtual experiments on students’ scientific inquiry skills: evidence from a three-level meta-analysis. BMC Psychology. https://doi.org/10.1186/s40359-026-05568-1

Image Credits: AI Generated

DOI: 10.1186/s40359-026-05568-1

Keywords: virtual experiments, physical experiments, scientific inquiry skills, meta-analysis, science education, cognitive load, embodied cognition, physics education, educational technology, simulations, secondary school, BMC Psychology