There Is No Single View...

By Jock D. Mackinlay and Chris Stolte

Graphical views of data can be effective at answering questions or presenting findings. Many effective views have been developed for specific tasks. However, people can be seduced by the power of graphical views of data into searching for the single perfect view that addresses all tasks, which is akin to the search for the Holy Grail -- no one has returned with a proven result. Treemaps and Dashboards are two examples of effective views that have encouraged Holy Grail searches. We argue that these Holy Grail searches have distracted us from the larger job of learning how to use multiple graphical views to see and understand our data. Rather than focusing on a single perfect view, we need to focus on a process of visual analysis, which explores a wide range of views to answer questions or present findings.

Treemaps

Figure 1: Treemaps

Courtesy of Martin Wattenburg.

Treemaps were invented by Professor Ben Shneiderman in 1990 to create a compact view of file directory tree structure that helped people to find large files. The Treemap design has proven to be very effective for certain types of questions. One of the most successful is the Map of the Market (see Figure 1), which can be found on the SmartMoney.com web site. Rectangles represent stocks with the size of the rectangle encoding the market capitalization of the stock. The rectangles are organized into market sectors and a red-green coloring shows the daily change in the stock value. The resulting view is very effective for questions that involve the general direction of the stock market. The Map of the Market encourages visitors to return to the SmartMoney web site every day to see what the stock market is doing.

However, the Map of the Market is not the single perfect view for all tasks. For example, imagine that the Map of the Market indicates that the communications sector is up, which leads to a host of questions. Is that sector volatile? What has been the performance for the last six months? What was the low for AT&T last week? Where are the communications companies located? Is there a correlation between their market capitalization and the volatility? You might be able to answer some of these questions with the Map of the Market but there are better views for many of them. In fact, SmartMoney provides additional charting tools for such questions.

Although Treemaps are not effective for all tasks, some people have tried to make them into the single perfect view. For example, a research paper in the 2006 IEEE conference on Information Visualization describes a Treemap layout algorithm that can make them look like pie and bar charts, which are standard views of data that should be accessible without having to adjust a Treemap layout algorithm.

Dashboards

Figure 2: Dashboards

Tableau Software entry in b.eye network 2006 data visualization competition

Dashboards are defined by Stephen Few as:

“A dashboard is a visual display of the most important information needed to achieve one or more objectives; consolidated and arranged on a single screen so the information can be monitored at a glance” (from Information Dashboard Design, O’Reily 2006)

A single view of your business is an enticing concept. However, searching for this single perfect view is a Holy Grail search. A Dashboard designed for one task will not be the perfect view for another task. Consider your CRM system. At one moment, you may want to monitor the locations of your customers to make sure you’ve defined equitable sales territories – a map is ideal for these tasks. But then you want to monitor how certain accounts have grown over time and a time series is a much more perceptually effective view. Finally, you might want to monitor how sales to those customers are segmented into different demographics categories – a question answered effectively by a matrix of bar charts. You will quickly run out of space if you try to include all the views you might need to run your business in a single Dashboard.

Although a Dashboard might be useful for monitoring your business, it cannot be the single perfect view of your business. Imagine seeing something unusual on your Dashboard. You will probably need additional views to find out what is going on. Stephen Few suggests that such analysis should be done with a “faceted analytical display”, a Dashboard that provides interactive views of your data. The focus shifts from the single perfect view to the interactive process of using views to understand your business.

Visual Analysis

Figure 3: The Visual Analysis Loop

Although Treemaps and Dashboards are effective for some tasks, people need many different views to see and understand their data. Rather than focusing on a single perfect view, we should focus on the process of using a wide range of views to answer questions or present findings. We describe this process with the visual analysis loop shown in Figure 3, which starts with a task. The first step is to forage for data that might address the task. Since raw data is hard for people to understand, the next step is to search for a visual structure that can be instantiated with the data. Working with the view, the next step is to develop an insight about the data. The final step is to act on the insight, which often involves communicating findings to other people with graphical views.

The key property of this process is that it is rarely sequential. Most views of data generate more questions than answers. A graphical view of data might suggest the need to forage for more data. An insight might suggest a new task. The act of communicating a finding to someone may require new views.

Since there is no single perfect view, people need the freedom to be able to explore many views of their data. A wide range of view types need to be provided and it should be easy to switch between views and view types, at any time. For example, data about a hypothetical coffee chain is described in four different views in Figures 4 through 7 (see below). Each is appropriate for a different type of question. Since the data is also an important aspect of an effective view, a visual analysis application needs to be connected directly to databases so that a person can easily change their data at any time.

People also need to explore views incrementally. They might need to adjust the data in the view by drilling down or introducing another dimension. They might need to adjust the view incrementally to focus on a particular aspect of the data. For example, the human visual system is very effective at comparing the lengths of aligned bars. The stacked bars in Figure 7 are effective for comparing the product type values. On the other hand, the user might be interested in comparing the values for each mark, which leads to the view in Figure 8. In fact, there are many different bar charts for this data, each with a different alignment. Furthermore, the user should be able to smoothly transition to the line, map, and scatter views shown in Figures 4 to 7. And sometimes you need to create unconventional views that are perfect for the question at hand. Figure 9 is a heatmap that clearly shows products that are having trouble in a particular market.

Thinking with data is hard. Most people want to focus on their task and data rather than the visual presentation of their data. A visual analysis application should default to views of data that embody the best practices of visual presentation. The user interface should also be easy to use and maintain the flow of visual analysis. Given the proper support, people will be able to generate many effective views of their data and avoid the seduction of the “single view”.

Figure 4: Lines are effective for trends







Figure 5: Maps are effective for geographic relationships







Figure 6 Scatters are effective for comparing two measures







Figure 7: Bars are effective for comparing values







Figure 8: Profit for each market by product type







Figure 9: A heatmap of profit and sales