Anybody with a working knowledge of Microsoft Excel and PowerPoint can create data visualizations today if they know how to present the tabular data in pictorial format. However, it does not mean that the data visualization has all the right design elements. Like any skill we apply in our day to day lives there is a notable difference between a novice user and a trained expert.

At Algorics, data visualization and discovery is at the core of what we do. Our trained data analysts and data scientists employ a user centric design (UCD) review process to ensure that what we present as a data visualization is not just a bar or pie chart but has all the correct design elements embedded within it to ensure the visualization is quickly and easily understood. For this approach we owe a big thanks to industry leaders in the field including Edward Tufte, Stephen Few and others.

If you are interested in learning more about the UCD approach, here is an outline of how we approach visualization design at Algorics.

  1. Does the visualization convey the most important goal/ purpose for which it is built?
    1. Use of a variety of graph formats in a dashboard should be avoided on the same screen. The end goal is to convey the message consistently and not show different variety of graphs.
  2. Does the visualization have adequate context for data?
    1. Measures without context (reference) do not give any indication whether the measure is bad or good.
  3. Does the visualization display the measure indirectly?
    1. If the intent is to show the variance of actual vs target, then the visualization should show the variance rather than displaying actual and target lines leaving the end user to calculate the variance.
  4. Does the visualization represent the most important data rather than projection of noise?
    1. The rule of thumb is to keep the most important data on the X or Y Axis and represent less important data with color, size or shape.
  5. Does the visualization representation employ proper design elements?
    1. Vertically oriented labels are hard to read.
    2. Limit the number of colors and shapes in one view to maximum 6 or 7 so that the viewer can distinguish them and see patterns.
    3. Limit the number of views in your dashboard to three or four.
    4. If a legend applies to one or more views, place it as close to those views as possible.
    5. Values in built-in filters are ordered in a way that makes sense for your data. For example, instead of listing classes alphabetically, you might order them by popularity.
    6. While scroll bars can be acceptable in some list views, in general you should avoid them in the major views of your visualization.
    7. Is there a consistency in the color pattern for variables across visualizations?
    8. Do the axis labels have appropriate units when necessary?

We could add more to this list but you get the idea of what goes in as part of our rigorous review process. It all goes back to the end goal, as Stephen Few put it “An effective dashboard is, above all else, about communication.”1

[1] Few, Stephen, Information Dashboard Design, 2nd Edition. P2