A graph redesign

I came across this recent post on Cole Nussbaumer Knaflic’s popular and highly worthwhile blog, wherein Cole proposes three redesigns of a less-than-optimal graph that a client of hers had created. I wanted to propose an another, alternative redesign but, as with most blogs (including this one), I couldn’t embed an image of my redesign in a comment on her blog, so I’m posting it here instead. To understand the context and purpose of my redesign, and for the rest of this post to make any sense at all, you’ll need to read her post first. Don’t worry, I’ll wait here.

Back? Great. So, my redesign looks like this:

 
gray 3-column graph copy.png
 

The main reason for this redesign was that I saw an opportunity to make the key message of the chart more visually salient and obvious. As per Cole’s description, the main message that we want to convey is that our company has two problems:

  1. We’ve fallen short of our target for one attribute (Attribute 5).
  2. Customers are rating us as worse than our competitor for one attribute (Attribute 1).

With this redesign, I think that these key messages will be understood more quickly and with less cognitive effort since much of the initial work required to interpret this graph will be done by users' unconscious visual systems, which generally begin by interpreting graphical elements before text. If we suppress the text, we can see that many aspects of the basic nature of the data and key messages are communicated by the graphical elements alone:

 
gray 3-column graph copy - blurred text.png
 

Within the first second or so, and before the reader has read any text or started to consciously think about what they’re seeing, they’ll probably grasp the following basic information in roughly the following order:

  1. There are two problems, i.e., the two red objects (which is good, since that’s our main message).
  2. Those two problems are problems because they extend to the left (i.e., are negative) while all of the other values in their respective groups extend to the right (i.e., are positive) (an important aspect of the message).
  3. The data is showing values for five items, and each row consists of three values associated with one item (the basic nature of the data being shown).
  4. The items are sorted from largest to smallest, based on whatever variable is represented in the left-hand column of bars.

As the viewer starts to read the text labels, the additional understanding that those labels provide (i.e., that the data are measures of how customers feel about our company's attributes, that those measures are being compared with internal targets and responses to a competitor, etc.) falls nicely into the basic --but accurate-- framework of understanding that’s been set up by the graphical elements. This, in turn, enables what I’m guessing would be rapid and cognitively easy visual consumption and quick understanding of the key messages, though only a well-designed user test would be able to determine this with confidence.

As is almost always the case, of course, there’s a trade-off with this design, which is that the actual “% of customers agree” values for the competitor aren’t shown, only the differences between those values and those for our company. That was a judgment call that I made based on the assumption that the audience would care more about the differences between our company and our competitor as opposed to the actual values for the competitor, but that assumption may not be valid.

What do you think? Can you see any other problems with this redesign or can you think of a better one? If so, please pipe up in the comments.

New book coming in 2018: "Beyond Dashboards" – feedback needed!

When dashboards fail to yield traction and satisfaction among users, dashboard creators often blame the visual appearance of the dashboard (colors, layout, fonts, etc.). Based on my experiences designing dashboards for many organizations, however, I now believe that the way that metrics, other data and interactive analytical features are organized onto an organization's dashboards, reports, self-serve analysis tools, and other types of information displays is another, possibly even more important cause of dashboard failure. A new book on which I'm working, tentatively titled Beyond Dashboards, proposes a framework for more logically organizing an organization's information and interactive analytical features onto its various types of information displays, thereby eliminating the most common user satisfaction and productivity problems with those displays.

Read More

Deep dive: Visualizing wide-variation data

In March 2016, I guest-wrote Stephen Few's popular quarterly VIsual Business Intelligence Newsletter. The topic was one that came up often enough in training workshops to merit a longer write-up (i.e., a "deep dive"): how to visualize data sets that include a combination of very small values (i.e., close to zero) and very large values (i.e., far from zero). Creating a standard line or bar chart based on such data sets yields bars or lines that are too small to visually estimate or accurately compare with one another, so the newsletter suggests some creative solutions to address this common challenge.

Read More

Mindware: Tools for Smart Thinking (book review)

Richard Nisbett's Mindware: Tools for Smart Thinking should be required reading for every university student (or anyone else who wants to make fewer reasoning errors). The book consists of an eclectic but extremely practical collection of "tools for smart thinking", covering concepts as varied as the sunk cost fallacy, confirmation bias, the law of large numbers, the endowment effect, and multiple regression analysis, among many others.

Read More

Do tooltips reduce the need for precision in graphs?

In many modern data visualization software applications, users can hover their cursor or finger over any bar, dot, box, line, etc. to see the exact, textual value(s) of each element. Since this allows users to see exact values whenever they need to know them, does this mean that graph designers no longer need to worry about how precisely values in their graphs can be estimated visually (i.e., without seeing a tooltip)?

Read More

Avoiding quantitative scales that make graphs hard to read

Every so often, I come across a graph with a quantitative scale that's confusing or unnecessarily difficult to use when visually estimating values in the graph. In this post, I propose simple guidelines for data visualization software developers to follow to ensure that the default quantitative scales that their products generate make it easy for audiences to "eyeball" values in the graph easily and accurately.

Read More