Data Visualized: Seattle Construction Craze
This project is an interactive storyboard visualization aimed to help builders and construction investors figure out the construction market trends in Seattle over the past few years, as well as help them learn more about its business logistics. It focuses on and compares residential and commercial construction as they are the two main types that constitute Seattle’s development.
I took the following roles in this team project: researcher, concept ideator, and interactive infographic designer. My three team members and I collaborated throughout the whole design cycle for 10 weeks and produced the following visualization in Tableau. All the elements of the visualization are clickable, moveable, and zoomable. For more information about the research and design process we went through, please read on.
About the Dataset
Having a broad audience of Seattle dwellers, home-buyers, builders, and investors in mind, we took two data sets from the government website that included all construction data. The first set had all building permits issued in Seattle within the past 5 years. The other set had data about construction permits issued more than 5 years ago. Since they had the same values in the columns, we combined the two sets in Excel, which gave us 10,106 rows. During our data cleaning stage, we renamed all single family and multifamily construction types to “residential” and removed buildings with blank construction type cells. We started working with that document that had the values for construction type (four categories), monetary value, issue date, finished date, contractor name, and location. After a series of usability tests and peer feedback, which are described later in the document, we decided to keep just the residential and commercial construction categories, as the data for the other two types (institutional and industrial) was scarce and did not contribute much to the big picture we were trying to represent. With that (also as an insight from our testing and feedback) came another plan of narrowing our initial target audience down to just builders and investors as the logistical data we had in the dataset and that we could visualize would be relevant to them more than the other two groups. Thus, with that idea in mind, we added another column to our Excel spreadsheet with a formula to calculate the days it took for each construction site to get an issued permit and to finish the project. That contributed to our further visualization of these logistical matters.
Our revised and final target audience includes Seattle builders and investors. I developed the following personas that we used as a guide:
Persona A (primary): James, 50, is a commercial construction company owner. His business has been pretty successful over the past 9 years, but with the recent city population growth and expansion of the tech giants, he has lost some project bids to competitors. Because of that, he would like to check out the current commercial competition in Seattle and how the pattern has been changing over the past years.
Persona B (secondary): David, 43, is a full-time senior project manager at Amazon, who enjoys looking for and investing his extra money into flipping houses around Seattle, which has become his venturesome hobby. His engineering mindset is number-curious, so he would like to evaluate the market in more detail before he starts investing in commercial construction.
The prototype that we set out to create had to allow these user tasks:
- Gain an overall understanding about the construction distribution in Seattle
- Sort and examine the data by construction type, value, and time frame on the map
- Be able to tell what trends related to the number of constructions and project value there are
- Examine the number of days it takes for a permit to be issued depending on construction type and time of the year
- Examine the number of days it takes for a construction to be completed by type and time of the year
- Examine how much unfinished construction there is
As a group, we used the Five Design Sheet (FDS) method to come up with a few design ideas for our visualization concept. We started out with a brainstorming page, where we went through the stages of filtering, categorizing, and combining/refining our best ideas. Then we generated the three following design ideas.
Based on our three design ideas and the tasks they intended to support, we produced the following paper prototypes so our users could interact with them and designed specific tasks for the users to follow.
The goal of our usability testing was to generate user insights and visualization design recommendations using our low fidelity prototype for the concept. Our study findings highlight what worked well and any usability issues the participants encountered. The user insights were summarized to propose changes to the prototype that could improve its usability, making it easier for users to accomplish their tasks.
- Do users know how to filter the data by construction type, value, and time frame?
- Do users know how to view the data by district and make cross-comparisons?
- Do users know how to pick specific years to compare the breakdowns of construction types?
Usability Testing Timeline
The approach of this usability study was four in-person moderated usability sessions with the participants who would be a good fit for our target group.
We conducted four interviews with all four participants on the same day of November 22nd, 2017 at the Husky Union Building. The participants were asked to perform the three tasks we prepared for them using the think-aloud protocol. Each session had a moderator sitting in the room with the participant, a note-taker, and a role-player to simulate the computer using the prototype.
Before conducting the usability sessions, we had created a data collection spreadsheet to record user insights and comments for each participant and each task.
Usability TESTING Findings
What Worked Well
Task 1: All four participants could easily tell the construction types were encoded by color and values were encoded by size. All of our participants liked that they could easily filter the data by construction type and value by clicking the corresponding items in the legend. They also liked that they could easily select a time range by using the two-point slider.
Task 2: All of our participants could easily read the construction trend from the lines. It was easy for them to filter the data to a specific construction type by clicking either the corresponding line or the radio button. They also easily understood that the small multiples were showing the data by districts and that the views were coordinated since they used the same scale. Participant 4 commented, “All elements are clear and all makes sense”.
Task 3: All of our participants could easily tell the breakdowns per construction type, using the pie chart visualizations.
Users’ Pain Points
Task 1: Two of our participants were confused about the size of the bubbles on the prototype because their size did not match the legend.
Task 2: One of the participants was confused when he “clicked” on the green line as he could not understand what years that trend line represented, as the graph was missing appropriate labels. He pointed his finger silently on the x axis of the green line graph, trying to see where the time range from 2004 to 2007 would fall, since he assumed the line was for the full time range from 2004 to 2017. He was confused to see the coordinated views were for the 2004-2007 period and was not sure why the trend line above did not match.
Task 3: Three participants had difficulties interpreting the wording of the task - they were not sure if the task was asking them to cross-compare the data from 2007 to 2010 (since the computer showed the pie charts for these four years on the same screen) or just for the years of 2007 and 2010. One of them commented on this, “How do I compare 2007 and 2010? Not just look at the other two?” One of the participants did not realize that the boxes that were intended for check marking were clickable and could be used as filters. Another participant did not find them helpful either, since all the pie charts for the years in question were already displayed, “I don’t understand why the years legend is there, the pie charts already have year labels.” One of the comments was on the lack of labeling for the pie charts as a whole - one participant did not understand what kind of breakdowns he needed to look for, and another participant did not understand what exactly the pie charts represented, since there was no title for all of them, “There is no label for these graphs - hard to tell what breakdowns the question refers to.”
improvements to the design
Based on the usability tests, we made the following improvements to our prototype that we incorporated in the final design:
Ruled out shapes to represent different construction types as it would be difficult to tell different shapes apart on the map, especially at points of occlusion. Instead, we chose to use color to represent construction types as color-coded nominal data is visually salient and easy to differentiate on the map.
Narrowed the target audience down to residential/commercial builders and investors.
Trimmed down construction types from commercial, residential, institutional, and industrial to commercial and residential only.
Changed bubble sizes.
In the prototype, we had three tiers of construction value ($0 - $300K, $300K to $1M, and $1M to $200M). The problems with this were that 1) the bin size for each tier was not even and 2) there were far more constructions falling within the $0 - $300K tier than the other tiers, which would result in occlusion when displaying those data points. Our users also found the construction value brackets to be very detailed and found the bubble sizes slightly difficult to differentiate on the maps. Based on this issue, we changed the bin size to separate data into distinctive tiers without creating too many bins.
Added clear labels as poor labeling of the graphs in our prototype caused some confusion among users regarding the purpose of specific graphs and their relationship to other graphs.
Added visualizations based on users’ needs.
In our user research, we learned that our target users are very interested in how long it takes to get a permit issued from the time of application and how long it takes to complete a construction on average, so we included these data cells in Excel and visualized them in our final concept to meet our target users’ needs. Our target users were also interested in learning how many ongoing constructions there are in Seattle, as well as their status, so we added this visualization as one of the stories on the board.
Added key insights to the visualization.
Our users also expected the visualization to give some context or information on the key insights. Thus, we created a story around the data so that it would be easier for the users to draw conclusions from the data. We added textual aids, using the UCD perspective, to ensure that the users know how to use the tools and what the data means.
Obstacles and Challenges
Although we wanted our users to view and compare the data by district in small multiples, Tableau’s would not allow us to separate the data into regions and present them individually, since zip codes were missing from our dataset. Thus, we had to remove that concept from our action plan.
Another issue we encountered was occlusion. Since our initial prototype did not have a lot of data points on the map, we did not anticipate this issue to be as bad as it turned out to be. A couple of our high-fidelity prototype reviewers mentioned occlusion as an issue, although they realized that it could be avoided by zooming in on the map, which our concept allowed. Regardless, we tried to solve it first by playing around with bubble sizes by decreasing the number of value ranges, so there would be fewer points on the map. Secondly, we tried to add trims around the bubbles to make them more distinctive, but that made the map look too busy, so we discarded that. And finally, we used color transparency, that allowed smaller bubbles to still be visible under bigger ones and made the map look more uniform. The map can be zoomed in for any “grey area” to uncover what’s underneath for better understanding.