In my last post I tied to fix an overloaded line chart Jorge presented in a recent post about loss aversion:
Jorge asked "does it make any sense to add those nine series to a single chart?“
My attempt to fix the chart by using some color coding, has its shortcomings that caused quite some discussion.
So again, how can you give the users all the data they expect while keeping the chart clean and readable?
D Kelly O’Day pointed out "More data or better colors won’t help a poor chart type selection" and presented a dot plot
Lets try to select the right chart type. In Chart Rules, As Simple as Possible, But Not Any Simpler! I presented an easy to learn set of rules to determine the best chart type .
1. Determine the relationship you want to display –
In our case a we have a Distribution Relationship, we want to show the Distribution of the Income Levels
2. Determine if you want to emphasize individual values or the overall pattern and
emphasize individual values or the overall pattern and Determine the chart type
As we want to emphasize individual values a column chart works best.
This chart already gives us a good feel for the income distribution in 1967- Looks like a almost perfect bell distribution with a belly for the mid income levels. But how did things change from 1967 to 2005? Lets create a set of small multiples to show the situation in 1967, 2005 and the increase from 1967 to 2005.
¡Great article! Very enlightening for my line of work
As I say in my post on Information Ocean, simply presenting the change between the start and end of the period tells a story of improvement in household fortunes between 1967 and 2005, but masks the more complex story, of gains made between 1967 and 1979, and losses between 1979 and 2005.
i like it Andreas. I couldn’t make sense of line charts or the dot plot, but the small multiple columns make sense right away. Although I don’t know why you changed the scale on the third one.
I think you’ve drawn a histogram, but with two mistakes: the bars aren’t abutting, and the bar lengths aren’t proportional to the size of the classes. You could also label the bars better – label the breakpoints, not the range of each individual bar. Once you do this, you might start to wonder about how the divisions were chosen, and if they are the most revealing for this data.
You also need to be careful about labelling the axis on the final graph – the values need to be interpreted as differences of percentages, not percentages themselves. i.e. the number of people in the >$100 bracket did not increase by 15%.
Tim,
Thanks for pointing out the problem with the common scale.
I fixed that in the blog post.
Thanks,
Andreas
Tim,
I had a second look on the scale and think the first version with the -15%- 15% scale is right. This way all charts scales span from min to max 30%, and a 5% change has the same height in all charts.
Andreas
Andreas –
Very good post. I’m still thinking about the loss aversion issue. Looking at your’s and Derek’s post shows me that each of us has focused on different questions, leading to different charts.
As I commented on Derek’s blog, it’s like a photographer who uses different lenses to shoot a scene. While the scene doesn’t change, the view of the scene does.
While some scene views (photos or charts for us) may be more interesting – pleasing than others, each has a role to play in understanding the actual scene that the photographer shot.
Chart type selection is a very important topic. Jorge’s income data provides a good example of the role of question to chart type.
I will try to synthesize Derek’s, yours and my charts and show how they address different questions about the same data. Since a lot of readers work with time series data, it would be nice to put together some time series chart type guidelines that help sort out potential questions and corresponding chart types.
Kelly