Desirable Difficulties in Graphical Displays

• In Event: Cognitive Processes in Comprehension of Visual Representations: Art, Diagrams, Graphs, and Models

Abstract

Although the type of data may differ—the Earth’s temperature, the amount of CO2 in the atmosphere, the number of e. coli cells in a petri dish, the rate of lung cancer in rural and urban communities—graphical depictions of data are central to most scientific endeavors. As a result, there has been much recent interest in educational sciences on students’ comprehension of data presented graphically and the factors that make comprehension easy and difficult. A common conclusion is that data should be presented as clearly as possible, and the best displays are those that require the fewest processing steps and maximally rely on relatively effortless perceptual processes. In this presentation, we will argue that rather than presenting students with information depicted in the putatively best, computationally easy formats, it may sometimes be beneficial to introduce ‘desirable difficulties’ in graphical displays.
We take as a starting point research on learning and memory which finds that conditions that make learning difficult can actually lead to deeper, more durable learning (Bjork & Bjork, 2011). We argue that difficult displays, that require more computational steps to comprehend than simpler displays, have three possible benefits for students: difficult displays require active rather than passive processing to comprehend, difficult displays may give individuals a subjective sense of disfluency and thus lead to be less hasty in making conclusions about data, and difficult displays with novel or unfamiliar characteristics may increase viewer interests. We will briefly review literature that supports each of these claims (Hullman, Adar, & Shah, in press).
In the second part of the talk, we will present data from a serious of exemplar studies (Shah, Freedman, & Miyake, in preparation) in which we compared comprehension of line graphs that had labeled lines to line graphs in which a legend was used. The conventional wisdom is that graphs should always have labeled lines because labels are assumed to require fewer steps for comprehension than legends (Carpenter & Shah, 1998; Lohse, 1993; Milroy & Poulton, 1978).
In Experiment 1, participants made true-false judgments about relations depicted in the graphs. Consistent with previous research (Milroy & Poulton, 1978), graphs with labels were interpreted faster than those with legends. However, in Experiment 2, we gave participants a surprise memory test and found a 23% accuracy advantage for graphs with legends. The third experiment required participants to describe graphs, and we coded descriptions according to which main effects and interactions were described.
The most important finding from this experiment is that viewers were 11% more likely to make inferences about main effects when viewing graphs with legends relative to those with labels.

Authors

• Priti Shah, University of Michigan

• Jessica Hullman, University of Michigan

• Eytan Adar, University of Michigan

• Akira Miyake, University of Colorado

• Eric Freedman, University of Michigan - Flint