I wish they had used the full data from Academic Torrents. Not only would this eliminate selection bias, but it would also allow us to see whether comments with em dashes are more popular than those without.