Duggan and Payne show that readers used structural cues -- paragraph and section beginnings -- to guide them in deciding what to read, and what to leave out. Readers would continue reading until information gain fell below some threshold, at both the small (paragraph) and large (section) scales.
Duggan and Payne suggest that the reason prior research could not show an advantage for skimming was because that earlier research used small documents -- 500 words, as opposed to 3000 words -- so information judgments could not be made at a large scale. However, clearly readers in earlier studies could make judgments at small scales (paragraph level, within roughly two pages), so why couldn't they tell what was worth reading and what wasn't? Is making such judgments at large scales more important than making them at small scales? We would have like to see more discussion of this by the authors.
Also, we note that skimming is something often done by experts, who will clearly have information gain thresholds that are different than those of novices, as were used in this experiment. It would be nice to learn how skimming differs between experts and novices.
Skim reading by satisficing: evidence from eye tracking
Geoffrey B. Duggan and Stephen J. Payne