What research really says

A 2012 systematic review by Benitti concluded that robotics generally contributes to learning, especially in science and mathematics. Some studies, however, show no gain. The difference lies in the design: a clear subject goal and a guided activity, rather than robotics for its own sake.

More recent meta-analyses put a figure on the effect. In 2021, Zhang and colleagues brought together 17 studies from compulsory schooling. The average effect is about half a standard deviation, mainly on computational thinking. The effect on attitudes towards science and technology is almost zero. In 2024, Ouyang and Xu brought together 21 studies. They found medium effects on achievement and attitudes, but no clear effect on computational thinking. The effect also depends on the subject. So two serious syntheses disagree, each based on few studies.

One question comes up often: does programming make you better at other things? In 2019, Scherer and colleagues analysed 105 studies on learning to program, with or without robots. They found an average transfer of about half a standard deviation, towards creative thinking, mathematics and metacognition. It is weaker towards school achievement and reading. Effects are larger when the comparison group does nothing than when it follows another activity. They are also larger in published studies than in others. These are two signs that the average overestimates the real effect.

• What is it compared with? A class that does nothing, or a class that learns the same concept another way?
• Who designed the test? A test written by the activity’s authors often favours it.
• How many students, for how long, and does the effect hold a few weeks later?
• Who ran the activity? The usual teacher, or an enthusiastic researcher?