From skimming and scanning to (the ultimate) reading, a new paper by Nir Grinberg looks at the ways we read online and introduces a novel measure for predicting how long readers will stick with an article.
Grinberg, a research fellow at the Harvard Institute for Quantitative Social Science jointly with the Northeastern’s Lazer Lab, looked at Chartbeat data for seven different publishers’ sites — a dataset of more than 7.7 million pageviews, on both mobile and desktop, of 66,821 news articles from the sites.
Grinberg was able to identify five types of reading behaviors: “Scan,” “Read,” “Read (long),” “Idle,” and “Shallow” (plus bounce backs, in the case that someone gets to a page and almost immediately leaves).
Not surprisingly, different kinds of news sites see different kinds of reading behavior. On the sports site, for instance, “we see there is a lot of scanning. I think what’s going on there is a lot of people go to sports sites in order to find a result, like the outcome of a game, and don’t read the full thing.” On the magazine site, meanwhile, people really seemed to be reading for extended periods of time.
In the second part of the paper, Grinberg identifies a measure that he calls “Semantic Information Gain” (SIG) – this can be useful for publishers, Grinberg says, because it ends up being highly predictive of how engaged someone will be with an article, and they should consider it along the other metrics tracked by companies like Chartbeat.