The Effects of Interactive Latency on Exploratory Visual Analysis

In their paper, Zhicheng Liu and Jeffrey Heer explore “The Effects of Interactive Latency on Exploratory Visual Analysis” – that is, how user behavior changes with system response time. As the authors point out, while it seems intuitively ideal to minimize latency, effects vary by domain.

In strategy games, “latency as high as several seconds does not significantly affect user performance,” most likely because tasks which “take place at a larger time scale,” such as “understanding game situation and conceiving strategy” play a more important role in affecting the outcome of a game. In a puzzle game, imposed latency caused players to solve the puzzle in fewer moves – spending more time mentally planning their moves.

These examples illustrate perhaps the most interesting aspect of latency: while it’s often true that time delays will make users bored or frustrated, that is not the only dimension of effect. Latency can alter the way a user thinks about a problem; consciously or unconsciously shifting strategies to whatever seems more time effective.

Liu and Heer focus on latency effecting “knowledge discovery with visualizations,” a largely unexplored area. One thing which makes this domain unique is that “unlike problem-solving tasks or most computer games, exploratory visual analysis is open-ended and does not have a clear goal state.”

The authors design an experimental setup in which participants are asked to explore two different datasets and “report anything they found interesting, including salient patterns in the visualizations, their interpretations, and any hypotheses based on those patterns.” Each participant experienced an additional 500ms latency in one of the datasets. They recorded participant mouse clicks, as well as 9 additional “application events,” such as zoom and color slider, which capture user interaction with the visualization.

The authors also used a “think aloud protocol” to capture participant findings. As the name implies, a think aloud methodology asks users to continually describe what they are thinking as they work. A helpful  summary of the benefits and downsides of this methodology can be found here.

Liu and Heer find that latency does have significant effects: latency decreased user activity and coverage of the dataset, while also “reducing rates of observation, generalization and hypothesis.” Additionally, users who experienced the latency earlier in the study had “reduced rates of observation and generalization during subsequent analysis sessions in which full system performance was restored.”

This second finding lines up with earlier research which found that a delay of 300ms in web searches reduced the number of searches a user would perform – a reduction which would persist for days after latency was restored to previous levels.

Ultimately, the authors recommend “taking a user-centric approach to system optimization” rather than “uniformly focusing on reducing latency” for each individual visual operation.


Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.