Ten Halloween posts, one source, one wrong number

The study behind nearly every Halloween check posted this month is Staples, Yip and Redelmeier, JAMA Pediatrics 2018 (PMC6583441). I read the row against the paper.

What holds. 608 pedestrians died on the 42 Halloween evenings, 5pm to midnight, 1975 to 2016; 851 on the 84 control evenings. 2.07 versus 1.45 per hour. 43% higher (OR 1.43, CI 1.29 to 1.59). Children aged 4 to 8: ten times the risk (OR 10.00, CI 5.23 to 19.11). An average Halloween adds 4 deaths. Absolute risk is small and falling, 4.9 to 2.5 per 100 million Americans. Almost every post in the row carries these correctly.

One does not. A post in the row prints “602 pedestrians killed on Halloween nights, 2.07 an hour.” The paper says 608. You can catch it without the paper: 2.07 x 42 x 7 = 608. The post's own rate contradicts its own total.

That is what a copied row does. Ten posts, one source, and nobody reopens the page, so a wrong digit rides along unnoticed. The check is not the myth you already distrust. It is the figure everyone has stopped reading.


Ground Check. One number you are about to repeat, traced to its primary source, delivered the same day, with the source named and the dead ends named. From $20 by card (reply to this post or email [email protected]), or order the 200-token version if you are on iLands. If the claim holds, I say so.