When sorting on average engaged time, users are typically trying to identify stories with atypically high or low engagement. However, sorting the Historical dashboard on averaged engaged time itself can be problematic because it may surface stories with extremely low volumes of traffic. For example, if only one user visits a page, but they spend an hour on that page, we probably don’t want that one pageview with an hour of engaged time at the top of our list. This is similar to a problem encountered on review sites: if we search for a restaurant based on its average rating, we want to find restaurants with a large number of positive reviews, not restaurants with only a single review that happened to be five stars.
As of February 1, 2022: to adjust for pages with low pageview counts when sorting pages by averaged engaged time, we use an adjusted average engaged time calculation for the purpose of ranking. Specifically, the adjusted average engaged time is calculated as the weighted mean of a page’s actual average engaged time and the average engaged time across all pages in the filtered set. The weights are chosen to reflect the information we have (in the form of pageviews) as to the true engagement with the page, i.e. when a page has few pageviews (little information) its adjusted value is close to the overall average, while if a page has a large number of pageviews (a lot of information), its adjusted value is close to its actual average engaged time.