Team/Roles: Stuart Giles: Design lead, research
Laura Hunter: Project management lead, writing
Irina Litvin: Development lead, usability
Eric Soderlund: Research lead, data management
Date: April - May, 2015
Skills used: Information visualization, slide design, usability testing, interviewing, persona development
An exhaustive, 40-page documentation of the project is available if you would like to see more detail than is provided here.
This six-week project spanned the majority of my graduate Information Visualization class at UW, and provided a fantastic opportunity for me to work alongside experienced professionals from the industry as my teammates. The assignment, specifically, was to create an information visualization of one or more novel datasets of at least 10,000 points total, using the user-centered design process, and making sure to follow Shneiderman's information seeking tasks for information visualizations. The project flow started with initial planning to write out the proposal, followed by interviews with key stakeholders (concurrent with refinement and cleaning of the datasets), persona development, creation of the initial visualization prototype, usability testing with stakeholders, and refinement of the prototype, all over the course of six weeks.
We decided pretty quickly on a topic: Eric, one of my team members, knew of a number of bike counters around the city of Seattle which kept data on how many cyclists passed them each day, and at what times. We wondered if we could analyze that data to find trends, perhaps comparing the dates against various holidays or other datasets. That seemed like it would be helpful, and on inspection, the bike counter data was suitable for our purposes (with some cleaning up), so we decided to proceed with the idea.
To begin with, we spoke with a number of experts and enthusasts on bicycling, data, and the combination of the two. Most notably, Craig Moore of the SDOT bicycling program was very excited to see us taking the project on, and gave us his blessing. According to Craig, SDOT would love to do more with the data, but there is very little time to engage with it on the level we were going to, with all of their other responsibilities, so any insights gained from our project, not to mention the final product, would be extremely helpful to them. While we conducted interviews, Irina and Eric worked with the data, cleaning it for use, and as a team we discussed the possibility of including weather data from Weather Underground.
Our prototypes were a work in progress from the moment we began ideating, simply because we were never making a mockup of the visualization, but rather actually building it in Tableau, capitalizing upon our group's strength of having an experienced developer and data scientist. However, they were also constantly iterated on as they were built. Before we presented our prototype to the class, we met and talked about our scope, and decided to focus more tightly on expert users rather than casual ones, as our interviews had indicated that while our visualization was moderately interesting to cyclists with a casual interest in data analysis, it focused too much on depth of information to cater too much to that group, and we wanted to go all in on making something useful for experts. With that in mind, I created a persona to guide our work, which can be found in the gallery below.
Once we were reasonably confident in our prototype (and had presented it to the class, received feedback, and implemented it), we took it to each of our original interviewees and conducted think-aloud usability tests. No guidance was given during these tests, even when the answer seemed painfully obvious to us as the designers of the visualization, in order to receive the most appropriate feedback to the product's intended context of use. The results of this testing, while sometimes contradictory, were incredibly helpful for us. Some feedback was, unfortunately, impossible to address: we were using Tableau Public as the platform for our visualization, and were thus constrained to its faults, such as some input lag. However, much of the other feedback was both immediately implementable and addressed concerns we had not even been aware were concerns, which was highly valuable. You can see and play with both of our visualizations, one of just bike data and one comparing it with weather data, below.