When does seeing motion help a learner? Three meta-analyses compared instructional animations or other dynamic visuals with static graphics. Their pooled estimates favour motion on learning tests, while estimates for hands-on procedures vary across reviews.
Five estimates from the research
Höffler and Leutner (2007) report d = 0.37. Learning from animations versus still pictures. 26 studies, 76 comparisons, mostly students. Sections 4.1–4.2, pp. 726–730. 95% CI [0.25, 0.49]. Learning and Instruction, 17(6), 722–738.
Video transcript
A drawing shows where a gear sits. An animation also shows which way it turns.
Across twenty-six studies, Höffler and Leutner found an average difference of d equals zero point three seven in favour of animation.
That number compares learning scores in standard deviation units. It tells us nothing about percent improvement.
Berney and Bétrancourt (2016) report g = 0.226. Learning from animation versus static graphics. 7,036 learners, 61 experiments, 140 comparisons. Sections 3.2 and 4.1. 95% CI [0.12, 0.33]. Computers & Education, 101, 150–167.
Video transcript
You can trace a route on a map, or watch a dot move along it. Each format can be useful.
Berney and Bétrancourt combined sixty-one experiments. The average favoured animation, at g equals zero point two two six.
The results varied widely. A small benefit across experiments does not tell you how well one particular animation will work.
Castro-Alonso et al. (2019) report g+ = 0.23. Learning from dynamic versus static visuals. 5,474 participants, 46 studies, 82 effect sizes in STEM and hands-on tasks. Table 3, Overall effect. 95% CI [0.17, 0.28]. Educational Psychology Review, 31(2), 361–387.
Video transcript
A diagram shows the parts of a circuit. Animation can show the path a signal takes through those parts.
Castro-Alonso and colleagues combined forty-six studies. Dynamic visuals did a little better on average, at g plus equals zero point two three.
The review combined science and hands-on tasks. Its average describes learning test scores across those tasks.
Höffler and Leutner (2007) report d = 1.06. Learning physical procedures, subgroup. Learners in five procedural-motor comparisons drawn from the 2007 review. Table 2, p. 732, Procedural-motor, and section 4.3.2. 95% CI [0.72, 1.40]. Learning and Instruction, 17(6), 722–738.
Video transcript
A picture shows a part in place. Watching it slide into a slot shows the movement needed to fit it.
Höffler and Leutner reported d equals one point zero six for learning physical procedures.
That estimate comes from five comparisons within the review. It describes physical actions, so it cannot predict results for a software demo.
Castro-Alonso et al. (2019) report g+ = 0.18. Learning hands-on procedures, subgroup. 1,094 participants and 22 effect sizes within the 2019 review. Table 4, Manipulative–procedural. 95% CI [0.06, 0.31]. Educational Psychology Review, 31(2), 361–387.
Video transcript
A diagram can show where to fold a sheet. An animation can show the fold happening.
For hands-on tasks, Castro-Alonso and colleagues found a small average advantage for dynamic visuals, at g plus equals zero point one eight.
The review did not find clear evidence that animation helped these tasks more than science and technology tasks.
How to read these estimates
The three reviews share some underlying studies. The two procedural rows are subsets of their respective overall results. Keep the reported estimators and intervals separate. Each d, g or g+ is a standardised difference between groups, in standard-deviation units. The magnitude cannot be read as a percentage. Differences between review estimates may reflect the sample, the task, the medium and how the static comparison was designed. The results measure learning, and offer no estimate of SaaS sales or conversions.
Motion and screenshots in a demo
For a demo, decide what the viewer needs to see change. A short clip can show a drag-and-drop action; a labelled screenshot can keep the final settings visible. Test comprehension with the same task and information in each version. This is a design example, separate from the reported studies. See the demo video research and visual cueing research.
Sources and methodology
DemoBro checked the public full text of three peer-reviewed meta-analyses on November 9, 2026. This is a selection of published findings. We did not recalculate their estimates or pool the reviews. Publication years are those of the papers.
- Tim N. Höffler, Detlev Leutner (2007). Instructional animation versus static pictures: A meta-analysis. Learning and Instruction, 17(6), 722–738. DOI · Public full text. Included studies published 1973–2003. Peer-reviewed random-effects meta-analysis of 26 studies, with adjusted weights for shared samples. Samples included school and university students, adults and military recruits. Media, visual realism and procedural-motor tasks were partly confounded. The procedural-motor subgroup has five comparisons.
- Sandra Berney, Mireille Bétrancourt (2016). Does animation enhance learning? A meta-analysis. Computers & Education, 101, 150–167. DOI · Public full text. Search through December 2013. Peer-reviewed random-effects synthesis of 61 between-group experiments from 50 papers. School, university and adult samples learned from animations or static graphics. The search reused studies from Höffler and Leutner. Effects varied substantially across comparisons (I² = 78.38%). The published version was checked in the University of Geneva repository.
- Juan C. Castro-Alonso, Mona Wong, Olusola O. Adesope, Paul Ayres, Fred Paas (2019). Gender Imbalance in Instructional Dynamic Versus Static Visualizations: a Meta-analysis. Educational Psychology Review, 31(2), 361–387. DOI · Public full text. Earlier experiments synthesised in 2019. Peer-reviewed meta-analysis of randomised experiments on STEM and hands-on procedures. The analyses included 46 studies and 82 effect sizes. The overall estimate is heterogeneous (I² = 67.94%). The hands-on subgroup was not detectably different from STEM in the between-subgroup test (p = .45). The full text was checked in the Erasmus University Rotterdam repository.
Reference downloads
Credit the original papers when reusing estimates. These files contain references and published summary statistics.