“Look both ways before you cross the street!” “Look left, right and left again!” These classic childhood safety lessons span generations and cultures. Yet traffic accidents remain one of the most common sources of injuries and fatalities for children around the world. In the European Union, children younger than 14 years account for a higher proportion of pedestrian mortalities than any other age group except the elderly; in the U.S., among children killed by cars, nearly a quarter were on foot. The numbers are particularly chilling in Israel, where kids account for 20 percent of pedestrian deaths.

Past studies have found that youngsters are less adept at identifying road hazards than adults, but Anat Meir, a lecturer in industrial engineering and management at Ben-Gurion University of the Negev and the Holon Institute of Technology in Israel, wanted to pinpoint exactly which behaviors lead to accidents, with the goal of finding ways to correct them.

To do that without putting anyone in danger, she turned to virtual reality. In 2013 Meir and her colleagues simulated 18 prototypical streets in Israel and used an eye-tracking device to study how 46 adults and children (ranging in age from seven to 13) evaluated when it was safe to cross. Children aged seven to nine, they found, exhibited the least caution when crossing, typically deciding to step into the virtual road with little or no hesitation, even when their field of vision was restricted. “We had parents looking on who were like, 'Wow, I cannot believe my child just crossed there!'” Meir says. “It caused them to reassess their child's road-crossing abilities.” The older children did not perform much better, though for different reasons. They often lingered on the curb for an inordinate amount of time—an indication that they are less able to distinguish between safe and hazardous situations than adults—and in interviews did not express an understanding of how factors such as car speed and field of vision affect crossing safety.

Interventions do seem to improve crossing success. In Meir's most recent study, described in Accident Analysis & Prevention, two dozen seven- to nine-year-olds underwent 40 minutes of hazard-detection training. Afterward, Meir and her colleagues compared trainees' and control kids' performances in the virtual road-crossing task. The children who received safety instructions were significantly better at crossing than the control subjects—to the point that their crossing skills resembled those of adults.

Next, Meir and policy makers aim to figure out how to translate these findings into the real world. “These kind of results are important because you cannot build interventions without an understanding of the problem,” says Joseph Kearney, a professor of computer science and associate dean of research and infrastructure at the University of Iowa, who was not involved in the work. “Now it's up to people with their feet on the ground to determine how they can develop training programs for children and for parents about good road-crossing habits.”