This apparently contradictory conclusion was made by researchers of the University of Toronto in the prestigious British Medical Journal [1]. The compulsory helmet was introduced in some Canadian provinces, not in others. Good comparative data should in theory result from this, but in practice things turn out to be more complicated. Between 1994 and 2003, researchers found a 54% decrease in the number of head injuries in young cyclists in provinces with compulsory helmets, compared to a 33% decrease in provinces without the compulsory helmet. The number of head injuries in older cyclists decreased by 26% in provinces with the compulsory helmet, and remain constant in provinces without the compulsory helmet.
According to the researchers, these results do not warrant the conclusion that the compulsory helmet works. A deeper analysis of the data revealed that these decreases could not be attributed to the compulsory helmet, since a decrease could already be observed before the requirement was implemented, and the trend simply continued after that.
But according to the researchers other research does sustain the hypothesis that a helmet can significantly reduce the incidence of head injuries. They advance a number of explanations for the lack of observable improvements following the introduction of the compulsory helmet. Firstly, other measures that improved cycling safety – such as improvements in road infrastructure plus awareness initiatives and campaigns to stimulate the voluntary use of helmets – probably had a greater effect. Secondly, the data is probably distorted by the fact the compulsory helmet was introduced in separate municipalities but was not mandated at the provincial level. Finally, the effectiveness of helmets could be greater for light and medium severity head injuries, but this was not apparent from hospital data.
All in all, researchers can only state that the compulsory helmet has had only a marginal effect on the number of head injury cases being treated at the hospital.

[1] Helmet legislation and admissions to hospital for cycling related head injuries in Canadian provinces and territories: interrupted time series analysis - Jessica Dennis e.a. (BMJ 2013;346:f2674).