This blog post is part of a series called Asked and Answered, about writing great survey questions and visualizing the results with high impact graphs. Dr. Sheila B. Robinson is authoring the Asked series, on writing great questions. Dr. Stephanie Evergreen is authoring the Answered series, on data visualization. View the Asked counterpart to this post on Dr. Robinson’s website.
When we ask people to rate something, it is usually on a numbered scale, though sometimes that scale will have word associations, like “Strongly Agree.”
A common, simple way to visualize this sort of data is with a column chart.
However, a regular column chart does not quickly communicate that we are talking about 100% of our survey respondents. So people like to stack the data together – making stacked bars.
Stacked Bars *seem* like a good idea – we show 100%, we can fit more questions and data into a similar amount of space – advantages, right? Except that stacked bars are difficult for people to read. How well can you compare the values of the orange segments? Not so much.
If you are going to use stacked bars, make some helpful formatting tweaks, like smarter color coding and an order from greatest to least.
Better, right? But what will these same tweaks work when we have many more response options, like when our scales run 0-10?
Working with shades of two colors is possible when we only have 4 response options but we’ll end up with too many shades and colors here. And while order may help us interpret the 0 data, it won’t boost the readability of this chart very much. (And please don’t @ me re: net promoter. If you are getting stuck on the metric, yer missin the point, friend. See Sheila B Robinson’s blog for a perspective on the use of net promoter survey questions.)
Really, it’s hard for anyone to distinguish between a 4 and a 5 on a 0-10 scale, so let’s assign the same color to multiple response options, like this:
And if we’ve gone this far, we might as well aggregate all categories within one color so that there are fewer segment breaks and legend entries.
Fewer segments in the stacked bar makes order more meaningful, too. Now we can see some stories emerging in this data. If we wanted to highlight certain aspects of this story, we could consider only adding an action color to a single segment.
A single color on a different sort order tells yet another story.
In my past couple of examples, I used red to indicate the not-so-great stuff, blue to indicate the great stuff, and gray to mark the neutral or the passive. But when interpreting this data, you might tell a different story – one that identifies the passive crowd as a place to focus some outreach efforts. In which case, passive isn’t neutral.
If so, you wouldn’t want to hide it in gray. And you’d probably want to look at every group in its own sunshine, comparing values for each. When that’s the case, use a small multiple bar chart with action colors on each segment.
We see yet more stories when we recast the data this way.
Of course, you could selectively group two or more response options and position them against the others, in a diverging stacked bar.
This time we sorted by the sum of passive and detractor, where one more story pops out.
Of course, rating scales can always be boiled down to a single number or otherwise essentialized. If you want to keep all of the data in your visual, these are some strong options.
Wondering what to do with neutral? That’s coming in its very own blog post, but you can pick up some hints here, for sure.
In the other posts in our Asked and Answered series, we’ll provide options for Check All That Apply, Ranking Data, and Demographics. See you soon.