Incremental Improvements is a series on small changes that can be done in order to make a visual design better. In each post I analyse the data visualization to see what works, what doesn’t, and what to do to improve it.
This time I’ll focus on another way of presenting Likert scale. The original chart by PEW Research Center inspired me to explore the diverging chart, which, in my opinion, is a better way of showing polarization in opinion. The chart shows the change in feeling about Trump over the course of his presidency among three groups — all Americans, republicans, and democrats.
There are two main issues with this chart. The first one is using uneven time intervals, which is probably caused by the data availability, but I would argue that it only makes comparison harder and doesn’t enrich the insight. The second one is using stack bars which due to different base (“no answer” responses are not shown) makes it impossible to properly compare the change in positive feelings.
So how would I address these issues?
What works? ✔️
- Palette selection — diverging palette is the best choice for Likert scale. Additionally, the color choice is very fitting, it both fulfils the cold-warm opposition, and avoids the confusion of natural association we have with colors. Imagine if instead of dark yellow there would be a red that is associated with negative / bad.
- Clustering data by different groups—this chart was primarily designed for within group comparison, which allows us to see the change in their opinion over the time. Because the differences are well visible, we can also compare those groups with one another.
And what doesn’t? ❌
- Categories order —due to different bases (excluding “no answer” responses) it’s hard to properly compare the change in warm feelings. For example the change among democrats between December 2016 and March 2018 looks bigger than its really is.
- Using column chart — the more natural way is presenting time on x-axis, that’s why column chart seems more intuitive.
- Using uneven intervals — time intervals are not even, sometimes there is 15 months difference and sometimes only 6 months, which, I guess, varies due to different data collection time. Adding more points only obscures the overall message, for the story is more important how the attitude changed since the beginning of Trump’s presidency.
How to make it better❔
- Switch to diverging stacked column chart —and add the most polarizing opinion closer to the x-axis. In that way it is possible to compare change over time in both very warm and very cold feelings, and positive (very warm + warm) and negative (very cold + cold) feelings.
- Remove additional data points —which allows to focus the attention on the overall change that took place between the beginning (2016) and the end (2021). Another positive effect of this change is avoiding the distorted time intervals which would be hard to show in accurate and commensurate way.
- Format axis and legend — adding the axes next to each chart group allows easier comparison of degree of warm / cold feeling among each group because it shows the local minimum and maximum. In addition, it allows for limiting chart junk and removing the data labels. It also creates the visual separation between the categories and invites to focus more on the in-group comparison (but the cross comparison is also possible). Such axis placement makes some space for the legend.
- Provide context for easier comparison — final touches like coloring legend, adding data summary, and formatting the time labels deliver better and clearer message.
Do you think it delivers the message better? Would you improve something else? If you have some suggestion for the next makeover feel free to drop them in comments.
If you are looking for some dataviz inspirations check the essential resources every data visualization designer needs to know