Redundant Encoding
Collaborators: Christine Nothelfer1, Michael Gleicher2, Steven Franconeri1
1Department of Psychology, Northwestern University, 2Department of Computer Sciences, University of Wisconsin - Madison
The redundant use of multiple visual features (redundant encoding) is a common data visualization technique. However, there is surprisingly very little evidence that there is a benefit to adding such visual complexity. Here we tested whether selection performance is better for simultaneous selection of multiple dimensions (color and shape) that specify the same set, relative to selection of either dimension alone. That is, is redundant feature selection helpful? Below is a summary of our studies.
Take-away 1: Visual selection is better with redundant encoding
When participants were asked to simultaneously visually select a large set of objects and indicate the global shape of a set of target objects, accuracy was much higher when those objects were redundantly encoded by color and shape (88%, across 3 experiments) than when those objects were encoded by color alone (66%) and shape alone (58%). This suggests that visual selection is better when selecting redundantly encoded objects. Further, this task was remarkably accomplished with a very brief (staircased) display time (M=89ms, SD=32ms).
Take-away 2: Visual selection is faster with redundant encoding
Participants were 7-10% faster in detecting the global shape of a collection of objects when they were redundantly encoded by color and shape, than when they were encoded by color alone or shape alone.
Take-away 3: Redundant encoding is beneficial in more realistic tasks as well
Does the redundancy benefit extend to more realistic tasks? When participants were asked to attend to a large set of objects, they were faster to indicate which quadrant had the smallest number of target objects (i.e., attending to the general distribution of objects, rather than global shape) when viewing redundantly encoded displays, than to displays encoded by color alone or shape alone.
Take-away 4: Grouping is stronger with redundant encoding
To investigate whether the redundancy benefit generalizes to other tasks, we used the repetition discrimination task (Palmer & Beck, 2007), which assesses the strength of a grouping cue. We tested grouping by luminance similarity, shape similarity, and (redundant) luminance combined with shape similarity. Grouping strength is assessed as the difference in response time to detecting the repeated letter (A or H) when it occurs within a group vs between groups. Our results revealed that grouping is much stronger when objects are redundantly encoded by luminance & shape, than when encoded by either dimension alone.
Take-away 5: Redundancy benefits do not appear to depend on task (attentional mode)
When looking at a set of objects, there are cases where you are attending to the entire set of objects (e.g., What is their distribution? Their global shape? Their variance?), and other cases where you are foraging through the set (e.g., Which object is the outlier? Which object is closest to you? Farthest away?). These types of questions require different types of attention -- spreading it out globally or focally, respectively. Is the redundancy benefit different with different attentional modes?
Participants again attended to a set of target objects embedded among a large set of distracting objects. However, half the participants were asked to attend to the global shape of the target objects (indicating toward which quadrant the global shape of the target objects was angled), while the other half were asked to attend to the target objects more focally (indicating simply which quadrant did not have any target objects). Both groups of participants were similarly faster in their responses to redundantly encoded displays (versus color encoded displays and shape encoded displays). This suggests that the redundancy benefit does not depend on attentional mode (task).
Implications
These findings have implications for the way that we encode features in data visualization (e.g., when graphing software such as Microsoft Excel defaults to redundant shape/color conjunctions in graph glyphs). More broadly, this result applies to how we attend to objects in our daily environment – it is much more likely that an object will differ from its surrounding objects in multiple feature dimensions than a single feature dimension.
Try it out yourself
Relevant Publications & Presentations
Nothelfer, C., Gleicher, M., & Franconeri, S. L. (2017). Redundant encoding strengthens segmentation and grouping in visual displays of data. Journal of Experimental Psychology: Human Perception and Performance, 43(9), 1667. [PDF]
Nothelfer, C., Gleicher, M. & Franconeri S. L. Redundant Coding Can Speed Up Segmentation in Multiclass Displays. Poster presentation at the IEEE VIS, Baltimore, Maryland (October, 2016). [Poster, Abstract, Video]
Nothelfer, C. & Franconeri S. L. Feature Redundancy Benefits in Different Attentional Modes. Poster presentation at the Vision Science Society Annual Meeting, St. Pete Beach, Florida (May, 2016). [PDF]
Nothelfer, C., Gleicher, M. & Franconeri S. L. Redundant Coding Can Improve Segmentation in Multiclass Displays. Poster presentation at the IEEE VIS, Chicago, Illinois (October, 2015). [Poster, Abstract, Handout, Video]
Nothelfer, C., Gleicher, M. & Franconeri S. L. Rapid feature-selection benefits from feature redundancy. Poster presentation at the Vision Science Society Annual Meeting, St. Pete Beach, Florida (May, 2014). [PDF]