Maternal smoking during pregnancy and autism: using causal inference methods in a birth cohort study