Create a personal approach to making meaningful data vis.
Define your intent with sponsor users
Identify yourand get to know their problems. In order to convey accurate messages, try to understand the topic deeply, even becoming a domain expert of the field when possible. Then, map users mental models, the flows of information they deal with, and stories of jobs to be done. Rely on simple charts up front and don’t be intimidated by Excel sheets with huge amounts of data. Define a point of view or hypothesis according to your target and purpose.
Understand and clean data
Start to look at your data set for its structure and the typologies of data included (strings, numbers, dates). Analyze various rows and columns by observing their relationship and implications. Clarify requirements and how your data will be included within the tool or product. Ask yourself: Is the data going to be real time? Will there be a periodic update of data? Will future updates have different variations and which “ranges” should I consider?
To clean data: look for inconsistencies, remove duplicates, and check for character encoding. Leverage tools for data checking and validation to continue refinement of your data set. Pay particular attention to similarity between entities in your data set (e.g. a misunderstanding of Congo versus the Democratic Republic of Congo).
Model data and check for visual validity
Check your initial hypothesis: group elements into categories or create alternative data structures. Verify numerical possibilities to observe data behaviors. If there is time for advanced analysis, include secondary, external data sets in order to look for meaningful layers and additional points of view. Apply multiple data manipulations in order to remodel the data set into a structure better suited for your intentions.
Rely on simple visual representations of data to identify patterns and trends or to verify a hypothesis. Use basic visual models to actually see the data, especially when the data set is too big to be understood directly. Visualization can help you focus on the real topic and find an interesting entry point for a story. Do not consider visual validation as the final point of your data exploration. This process is meant to be iterative; model or clean data any time you find something to investigate and move on when you are confident in your analysis.
Structure and style
The essence of a powerful data vis relies on your decisions of what to emphasize and what to hide in order to convey a message. Avoid presets, automation, and short cuts at first to constrain your focus on content. Try to represent the information in new ways imagined from scratch. Consider the interaction behaviors that will allow your design to drive useful insights. Draft different versions of your design and try a variety of charts. Consider this as a cyclical step in your process until you find a solid base structure from which to reintroduce your data set.
Next, work with color palettes, typography and line weights to make your visualization look like IBM and be harmonious. Don’t limit your thinking to basic bar charts and line graphs if you feel stuck with an existing model, but aim for an appropriateness and reduce reinventing the wheel when possible. Pushing the boundaries is all about searching for a better truth to the data. Consider two key principles when exploring:
- : Say everything you want to say—no more, no less—and don’t mislead.
- : Use or create the best method available to show your data
Test and iterate
Set up heuristic evaluations and usability tests. Walk users through prototypes and run stress tests with small and large data sets to look for edge cases. Gather impressions and opinions from sponsor users, or for more general visualizations, people who have never seen the project. Don’t worry if you have to start over and refine your initial hypothesis. Visual representations of data are not always easy to understand on the first try, and even small usability changes can solve big problems. Always use research as a means to influence iterations and justify necessary changes.
Refine and implement
Fix last details in your style and animation. Look for bugs and functional errors. Collaborate with your developer tofor building the data vis. Once you review the designs in code for accessibility, globalization, etc., do a final spot check of your work with another designer. Check your vis for the following common :
- People don’t go in order. Our pacing happens in “fast bursts” based on stimulation.
- We don’t see first what stands out. Instead, we look for differences.
- We only see a few things at once.
- We seek meaning and make connections.
- We rely on connections, metaphors and visual signals to depend on what is expected.
Designing a data vis goes beyond aesthetic exercise. If you’re getting a request to make a data visualization or believe it might be a valuable approach, try to answer a few of these questions up front. Each topic is designed to help you and your stakeholders keep in mind what’s involved in the process.
- Where are you starting from—a user need, a data set, a request from a manager or exec?
- What problem are you trying to address and why will data visualization help to solve it?
- What goals do you hope to accomplish with the vis?
- What is the nature of your intention—to make a point, tell a story, provide deep exploration?
- Who is the target user for your data vis?
- What does your user want to do with their data?
- What cultural, domain, or industry-specific needs does your user have for the visualization?
- What user outcomes will indicate you’ve been successful?
- Do you have a usable data set?
- Are you designing mock-ups with real data?
- Will the visualization need to get periodically updated?
- What is your plan to make the visualization accessible?
- What is your strategy for language support?
- Where will the data vis live — in software or a website, a report or presentation, an article or blog post?
- Where will your user be when viewing or exploring the data vis?
- Is it going to be static or dynamic, passively consumed or interactive?