I thought the paper could do more is in moving from statistical non-significance to ruling out substantively meaningful associations. For example by showing that confidence intervals exclude *meaningful* reductions in autism or ADHD diagnosis rates. Definition of meaningful matters a lot.
