Or is it a statistical artifact from crap control of confounders, who knows?