My makeover focused only on the most recent year, 2015. I did look at the trends, but didn’t find anything particularly of note to add to or support the story. I wanted to keep things simple. I played around with versions without maps, but this data set seems challenging to show in other formats. Any time I took the map out, I found myself wanting to know the detail of each mark: WHAT country is it and WHERE is it?
This week’s original chart came from CNBC.
Here’s what I like:
I happen to know this one was specifically designed for mobile. CNBC had a very specific brief that required the dashboard to render correctly on mobile devices. With that in mind, the dash is quite small and simple.
It has a nice “About” tooltip in the top left, where I’m likely to see it.
The lookup table does allow me to see individual numbers, should I wish to find them.
What I’d improve:
The colour palette seems arbitrary to me. I think there are nicer palettes available which could portray the categories more clearly
The circles in the lookup table make it hard to read the numbers.
I don’t like the vertically oriented labels in the upper table. There is space enough to have them horizontal
Here are a couple of ideas I rejected:
The timelines didn’t reveal too much and the dots left me feeling too curious about the actual country each dot represented. Hence I went back to a map.
I had a crazy idea that listing out all the countries next to a stacked bar would work. It didn’t. Or did it? Could you see that idea going anywhere?
This week my aim is to emphasize where the dataisn’t. I made the y-axis scale show the full range from 0% to 100% female representation. Why? In an ideal world, states should have approximately 50% female representation.
When I first explored the data, Tableau automatically tops out the y-axis just above the highest percentage state. This is reasonable but it means you don’t immediately see how low the representation is in some states. In the image below, you can see bunch of states are high on the y-axis. You have to stop yourself and wonder about truncating the UPPER end of the y-axis when showing percentages. Colorado and Vermont are the closest to 50% so well done them. But it’s still NOT 50% female representation.
I decided to ignore the values relating to gender split in the population: all the states are close enough to 50% to make any differences mere noise.
A map helps for quick lookup if the viewer is looking for a specific state
Things I would have changed:
Was this data best shown as a map? If the story is about underrepresentation, then show that, rather than just geographic information. Bar charts, or distributions could show that story more powerfully.
The colour scheme is, to me, a little muddy. High/low doesn’t pop out. The grey is especially unclear. Without looking at the key, would you be able to tell if grey was good, bad or middling? I wouldn’t know.
This week I decided to see what everyone else did before I did my own, to force me to come up with a new perspective.
Most people seem to have focussed on visualising the ages that these entrepreneurs made their millions/billions, reusing the same main fields as the original.
Since that’s what most others do, how could I show the same story in a different way?
I decided to focus on the other measures: the gap between millionaire and billionaire and their net worth. It’s not quite the same story, but I needed something original!
Did I succeed? I’m not sure.
As I write and reflect on this one, I don’t think my story is as compelling as the original. Not only that, I don’t think the scatterplot makes it clear! I do think the difference between Alan Sugar and Bill Gates is pretty clear, but as a scatterplot it’s just too hard to decipher for a story that’s better told in the manner of the original.
I do propose four changes to help focus on what I think the key story is: the growing obesity crisis.
Change the colour scheme. I found it a little hard to decode the colour scheme. Which colour is good and which is bad? I’ve changed to a red/grey scheme: now red is bad and pops out more clearly. My palette choice isn’t perfect as it’s not a truly continuous scale.
Change the order of the BMI categories. I’ve put the highest levels of BMI at the bottom, so it’s easier for us to see the growth of obesity.
Change the default view to be sorted showing the worst countries first.
What do you think? Would you have kept the original? What would you have changed? One thing I wonder about is whether the original colour scheme, harder as it was to decode, was more appealing because of its novelty and brightness? If you spent time looking at Ramon’s, did the colour scheme draw you in? Would my red/grey one have piqued your interest?
Tableau labels the zero on axes. There’s nothing wrong with that, unless you’re showing Ranks, when zero is meaningless. What I did here was to drop a text object over the top of the 0.
I also added the axis label (“Rank”) above the 1 as it’s more likely to be seen and read by a viewer at the top than halfway down the left hand side of the axis, oriented on its side.
How do you label both ends of a line of a slope chart? Simple: you just turn on labels and choose Start/End of line? Well, no, because then you end up with your labels misaligned:
One option is to just label one end of the line. This is ok, but sometimes reduces the speed to insight. If you want to label both ends, you need to duplicate your measure onto a dual axis, and set each label differently. One measure is set to label the start of the line, and the other is set to label the end of the line.
There is a downside to this: the marks are all duplicated. This can lead to the edges of the lines looking jagged.
This week reiterated that when you engage with data rather than just consume it, you learn more. This week’s data is fascinating. FiveThirtyEight’s original article is very good at explaining the perils of choosing just one measure to explain something as complicated as a city’s racial profile.
Understanding the nuances of statistical measures is vital at all times. This is especially true this year for people in the US and the UK. In the US, the election will be rife with statistical claims, and in the UK the EU Referendum is also awash with claim and counter-claim. Even when statistics being used are true, we need to be educated enough to judge if they are appropriate.
Here’s their great chart:
What do I like about their scatterplot?
Two trend lines. One is the 45 line, showing what a perfect correlation would be. The trend line for the data is in red. Having both helps you see more clearly how one measure skews the other.
It’s well labelled. Scatterplots do need time to digest as you work out what all the different areas of the chart mean. The annotations and title help greatly.
What would I improve?
I do find FiveThirtyEight’s grey background a tiny bit frustrating: I don’t know why they consciously reduce the contrast between foreground and background in this way. It’s only a small gripe.
The article discusses three measures but the chart only displays two of them.
I wanted my makeover to focus on comparing all three of the different measures in one view. I chose to highlight some of the cities they talk about in the article in order to see how the 3 measures change the rank of the city.
I imagined 3 rows, one for each measure. On the left of each row would be a description of the measure, and on the right, a tidy sorted bar chart. All bars would be grey except for the cities I wanted to highlight.
But it just didn’t quite work. The bars for the highlighted cities were too indistinct:
Since horizontal didn’t work (a shame, because horizontal charts fit better in a tweet!), maybe vertical would? Nope.
Note: writing up this post, I now think I could have succeeded in this design by dropping the 5 citites and using highlighting to show cities. Problem is, that would require people interacting with the viz to get anything out of it, and I’m trying to focus on static charts in my Makeovers.
The problem with both of these is the Integration/Segregation index. Because that crosses the zero line, many of the bars are tiny, so it’s hard to see individually highlighted bars.
My solution to this was to change the measure: instead of showing the actual value, I’d use Rank. Also, at the back of mind all along had been the words “Slope Chart”. I had initially wanted to avoid the slope because I seem to be using them for lots of my Makeovers.
Could I tell a slope story in a new way? Can slopes work horizontally?
I don’t think that really worked. Slopes seem like time lines in that we are totally accustomed to seeing them oriented horizontally. Is this is an aspect of our visual system or because convention has trained us to read them this way?
Of course, your data is also in the cloud. Our archipelago needs clouds, too, raining data from the devices to the local storage represented by the islands.
In the archipelago data flows around like water. Water is like the air around us. It contains the information. It flows freely. Polluted water corrupts your data and Pirates, literally, can land on your islands and steal your stuff.
Modern data architecture is an archipelago.
Or maybe it’s a Data Swimming Pool.
Your data’s the water. Data needs to be kept clean. Chlorine and Vacuums represent the ETL layer. Lifeguards are your disaster recovery processes. And kids who pee? Well that’s the SQL Injection attack, of course.