Usually, non-randomized studies can provide evidence only of correlations, not actual causation. But the type of analysis that the Chaisemartins applied is different. "The regression discontinuity method we used is considered almost as reliable as a randomized controlled trial in terms of teasing out correlation from causation," said Clément de Chaisemartin.
The researchers took advantage of the fact that the Swedish policy essentially created a randomized controlled trial. People born in March and April 1975 are extremely similar in terms of their susceptibility to COVID-19. Meanwhile, those born in March got the BCG vaccine, while those born in April did not. It's almost as if the individuals were randomly placed in the two different groups.
The researchers compared the COVID-19 outcomes between the two groups and found that cases per capita, hospitalizations per capita, and deaths per capita were very similar for people born just before and just after the April 1st cutoff.