Pair Analytics

Problem
Understanding the cognitive process of a participant using a visualization tool is difficult. Conducting an interview after an experimental session may reveal some major points, but the results will be mostly summative and will not capture details encountered on the fly. A think-aloud protocol may help capture such information, but this approach may affect the behavior of the participant, and the collected data is often somewhat random and difficult to understand. Participants often cannot articulate what they think, or provide too much information that is not necessarily helpful. Finally, the tool itself often serves as a barrier against effective sensemaking since most InfoVis evaluations do not provide proper, long-term training for the tool, making the participant less than fluent in using it.

Solution
The basic idea of Pair Analytics (Arias-Hernandez et al. 2011) is to form a team consisting of an experimenter (often a visualization expert) and a participant (often a subject matter expert) to analyze the dataset. The pair complements each other since the experimenter is well-versed with the tool and will "drive" it, and the participant is well-versed with the problem domain for the dataset to analyze. Furthermore, in solving the task together, the driver and the domain expert will be verbally externalizing their cognitive processes when they communicate to investigate the data. This verbal communication provides natural insight into the sensemaking process, compared to think-aloud protocols where the verbal communication is easily perceived as artificial by the participant.==Consequences== The Pair Analytics pattern requires that the experimenter who is driving the tool remains objective and does not influence the domain expert in any way.

Examples
The VAST 2007 competition (Plaisant et al. 2008) included a special session where the winners were invited to use their visual analytics tools on a smaller dataset and working together with a professional analyst. Additional examples can be found at http://tinyurl.com/pair-analytics.