Can the "specificity" of an association be rehabilitated as a basis for supporting a causal hypothesis?