A controlled classroom test, not a claim about all screen time
Educational apps are frequently marketed as a shortcut to better learning. But a tablet alone cannot teach a child how to solve a problem.
A peer-reviewed study published in PLOS One on 9 October 2026 provides a more specific finding: when carefully selected STEM apps were used as part of a structured, teacher-supported kindergarten programme for 12 weeks, children made greater gains on a computational-thinking assessment than children following ordinary lessons.
The benefit was measurable but modest, and the research does not show that more screen time is always better.
What computational thinking means for a five-year-old
Computational thinking is not simply coding or knowing how to use a device. It includes breaking a complex problem into smaller steps, spotting patterns, following a sequence, understanding rules and finding mistakes. A child might arrange picture cards to show how to complete a task, recognise a repeating pattern, or discover why a sequence produces the wrong result.
These foundations may be useful beyond computing, but whether gains on a targeted test transfer to general academic achievement is a separate question. That distinction matters when schools consider spending scarce resources on educational technology.
How the study was designed
Researchers Yumei Wang and Shiying Liu conducted a cluster-randomised controlled trial in five kindergartens in Northern China. Each kindergarten contributed two classrooms. One classroom at each site was randomly allocated to the STEM app programme and the other to a waitlist control condition, producing five intervention classrooms and five comparison classrooms.
The final analysis included 125 children per group, all aged four to six. The sites were recruited through existing institutional relationships rather than randomly selected from all schools, which limits generalisation.
Randomisation occurred at the classroom level rather than independently for each child, an important feature because classmates share teachers and learning environments.
What children actually did
The intervention consisted of three sessions each week, approximately 20 minutes per session, over 12 weeks. That is 36 planned sessions or 720 minutes of guided app activities, rather than unrestricted tablet access. Researchers supplied Android tablets running three educational apps with Mandarin-language interfaces, icons and audio prompts.
One app focused on sequencing and algorithms, another on pattern recognition and mathematical reasoning, and a third on debugging and problem-solving. The apps were selected to be developmentally appropriate, work offline, and avoid advertising and in-app purchases. Teachers received a two-hour training session and a structured implementation guide.
Research staff monitored implementation through app logs, teacher records and observation.
The curriculum became progressively more demanding
The first three weeks emphasised recognising sequences and understanding simple procedures. Weeks four to six introduced pattern recognition alongside sequencing practice. Weeks seven to nine focused increasingly on finding and fixing errors. During the final three weeks, children were expected to combine the skills in more integrated tasks.
The sequence is an important part of the treatment: the experiment tested a package of app content, guided teaching, progression and regular practice. It cannot isolate the effect of the tablet itself from the quality of instruction or the structured activities.
How learning was measured
All children completed a pretest and posttest using TechCheck-K, an assessment designed for young children that does not require previous programming experience. The instrument comprises 15 picture-based multiple-choice questions covering six computational-thinking domains, including algorithms, representation and debugging.
Researchers converted raw scores to a percentage-correct scale. Trained assessors administered the assessments individually, and posttest assessors were blinded to children’s assigned groups. Parents also supplied demographic information, including age, parental education, household income and home language.
Statistical models adjusted for baseline ability and demographic characteristics.
The main result: a modest advantage
After adjustment for pretest performance and measured demographic differences, the children assigned to the app programme scored an average of 3.75 points higher than controls at posttest. The reported p-value was below 0.001 and Cohen’s d was 0.39, conventionally interpreted as a small-to-moderate standardised effect.
This is evidence of a difference in the specific computational-thinking outcome tested. It does not mean that children became 39% better thinkers, nor that every participant gained 3.75 points. The latter figure is an adjusted difference between groups, not a guarantee about any individual child.
Why classroom randomisation matters
The study had ten classrooms, but 250 children. That does not mean there were 250 fully independent randomised units. Children in the same classroom can resemble one another because they share a teacher, school and activities. The researchers therefore examined clustering and conducted mixed-effects sensitivity analyses at the kindergarten and classroom levels.
They reported that the main intervention association remained evident under those checks. Nevertheless, the authors’ own planning calculations indicated limited statistical power once clustering was considered, around 0.49 rather than the commonly targeted 0.80. Only ten clusters make the design less robust than its headline participant count might suggest.
Did more engaged children learn more?
The researchers also examined behavioural logs within the intervention group. Their engagement index included completion of tasks, time spent, persistence, efficient help-seeking, independent navigation, self-correction and attendance. Children with higher engagement tended to achieve better posttest performance after accounting for baseline ability and demographics.
But engagement contributed only about 1.5% of additional explained variation. This was a statistically detectable association, not a large standalone predictor. Children who begin with greater motivation or confidence may engage more, and the analysis did not randomly assign different levels of engagement.
What happened to children in the comparison group
The control classrooms continued their usual kindergarten educational activities during the 12-week comparison period. They did not receive the specialised computational-thinking app programme at that time. After posttesting, control classrooms were offered access to the apps and materials for a shorter four-week period.
This waitlist design addressed access to the intervention, but it also means the study compared a highly structured targeted programme with ordinary instruction. It did not compare the apps with an equally intensive non-digital programme targeting exactly the same skills.
Why this is not a general endorsement of tablets
The findings concern three selected, advertisement-free apps, short scheduled sessions, trained teachers and a staged curriculum. They should not be extrapolated to entertainment apps, unsupervised browsing, or unlimited screen exposure. Nor do the results show that the gains lasted after the intervention or improved later mathematics, reading or academic attainment.
The study measured computational thinking immediately after 12 weeks, not long-term development. Physical play, conversation, rest and other forms of learning remain important parts of early childhood education.
What educators and parents can take from the findings
The strongest practical message concerns design and implementation. If a school uses educational technology, it should choose activities tied to specific learning objectives, introduce skills progressively, offer opportunities to correct mistakes and provide teacher support.
Short, planned sessions are different from handing a child a device and hoping learning will occur. Schools should also consider costs, access, privacy and whether non-digital alternatives could deliver comparable outcomes.
Parents need not buy a particular app on the strength of this study, because the research tested a coordinated classroom programme rather than a consumer product in isolation.
Limits on applying the findings elsewhere
Five kindergartens in one part of China cannot represent every early-years classroom. The children were four to six years old and the apps were configured for Mandarin-speaking classrooms. The sample excluded some children with developmental conditions that would prevent participation in standard activities, limiting conclusions about inclusive education.
Six children were lost before posttesting, and the primary analysis used complete cases rather than a strict intention-to-treat approach. There was no longer-term follow-up to test whether gains persisted. App engagement was observed rather than experimentally manipulated, so the relationship between engagement and performance should not be interpreted causally.
The wider research context
Other research has also examined teaching computational thinking through classroom curricula, hands-on activities and programming games.
A 2026 randomised trial in Head Start classrooms evaluated a curriculum combining unplugged lessons with app-based games, while a family-based randomised trial reported improvements from both tangible programming and screen-free activities.
Related Research Today coverage includes AI and peer feedback in student writing, a low-cost environmental learning platform, and technology-related teacher stress. These examine different populations and outcomes, so none directly validates the kindergarten trial.
Those findings reinforce an important distinction: the educational objective is to teach problem-solving concepts, not to maximise device use. The present trial adds evidence about one particular structured app programme, not proof that screens are indispensable.
What future studies should test
A useful next step would compare app-based lessons with equally structured non-digital instruction, using the same teacher contact time and learning goals. Larger trials with more schools could test whether results hold across different socioeconomic settings and languages.
Longer follow-up would show whether children retain their skills, while independent assessment of mathematics and other outcomes would reveal whether gains transfer beyond the test. Research should also measure accessibility, costs and the amount of adult support required to reproduce the programme at scale.
The central finding
In a 12-week classroom trial involving 250 kindergarten children, a carefully structured STEM app programme produced a modest improvement on a targeted computational-thinking assessment compared with usual instruction. More active app engagement was associated with slightly better outcomes, but explained little additional variation. The evidence supports thoughtful, supervised use of selected educational tools, not the idea that more screen time automatically makes young children better problem-solvers.
Source Information
Study: Wang, Y. and Liu, S. (2026). Enhancing computational thinking in early childhood: A cluster-randomized controlled trial of interactive STEM app intervention and the role of engagement. PLOS One, 21(10), e0353980.
Published: 9 October 2026.
Study type: Peer-reviewed cluster-randomised controlled trial in ten classrooms at five kindergartens, involving 250 children aged four to six and a 12-week intervention.
DOI: https://doi.org/10.1371/journal.pone.0353980.
Data availability: Child-level behavioural and demographic records are not publicly released for privacy reasons; qualified researchers may request de-identified data from the corresponding author.
