Trumping (Most) All on Twitter

Initially I wanted today’s piece to be coverage of the apparent coup d’état in Zimbabwe over night. But while I have found some coverage of the event, I have not yet seen a single graphic trying to explain what happened. Maybe if I have time…

In the meantime, we have the Economist with a short little piece about Trump on Twitter and how he has bested his rivals. Well, most of them at least.

Trumping one's rivals
Trumping one’s rivals

The piece uses a nice set of small multiples to compare Trump’s number of followers to those of his rivals. The multiples come into play as the rivals are segmented into three groups: political, sport, and media. (Or is that fake media?)

Small multiples of course prevent spaghetti charts from developing, and you can easily see how that would have occurred had this been one chart. But I like the use of the reddish-orange line for Trump being the consistent line throughout each. And because the colour was consistent, the labelling could disappear after identifying the data series in the first chart.

And worth calling out too the attention to detail. Look at the line breaks in the chart for the labelling of Fox News and NBA. It prevents the line from interfering with and hindering the legibility of the type. Again, a very small point, but one that goes a long way towards helping the reader.

I think the only thing that could have made this a really standout, stellar piece of work is the inclusion of another referenced data series: the followers of Barack Obama. At 97 million followers, Obama dwarfs Trump’s 42.2 million. Would it not be fantastic to see that line soaring upwards, but cutting away towards the side of the graphic would be the text block of the article continuing on? Probably easier for them to do in their print edition.

Regardless, this is another example of doing solid work at small scale. (Because small multiples, get it?)

Credit for the piece goes to the Economist Data Team.

Phillip’s Curves are Flatlining

I’ve worked on a few scatter plots of late and so this piece from the Economist grabbed my attention. It examines the correlation between unemployment rates and inflation rates. Broadly speaking, the theory has been that low unemployment rates lead to high inflation rates. But the United States has had low unemployment rates now for a few years, but inflation is around that ideal 2% realm. This theory is called the Phillips Curve.

Straightening out the curve…
Straightening out the curve…

The graphic does a nice job of showing three data series all in one plot. Normally, I would argue for splitting the chart into three smaller plots, a la the small multiples. But here, the data aligns just well enough that the overlapping is minimal. And smart colour choices mean that each data range appears clearly separate from the rest. A nice thoughtful addition is the annotations to the time period are set in the same colour as the dots themselves.

My only two quibbles: One, I would probably increase the height of the chart to better show the trend line. I find that for scatter plots, a more squarish profile works better than the long rectangle. Overall, though, a really well done chart. Second, I would consider adding a zero line to the x-axis to show 0% cyclical unemployment. But that might also not be terribly useful, because you can see how the curve should move regardless of that natural line.

Full disclosure: the Economist article cites a paper from the Philadelphia Fed Research Department, which employs me.

Credit for the piece goes to the Economist Data Team.

An Ailing Graphic on the Healthcare Labour Force

I know I have said it before, but I like the increasing number of graphics-led articles published by Politico. Many policy and politics stories are driven—or should be driven—by data. But, myself included, we cannot hit it out of the park at every plate appearance. And that is what we have from Politico today, actually last week.

The graphic focuses on the healthcare industry and its need for a larger labour force in coming years as the baby boomers continue to age and start to retire. If their own doctors retire along with them, who will be their new doctors?

But there are two components of the graphic on which I want to focus. The first is the projection of the number of registered nurses (RNs) in 2024 compared to a 2014 baseline.

We need more. Just more.
We need more. Just more.

The story focuses on the future condition, but that colour is set to the lighter green thus drawing the reader’s eyes to the 2014 data point. Flipping those two colours would shift the focus of the chart to the 2024 timeframe, which would better match the text above.

Then we have the design decision to include a line chart for the growth rate, presumably total, for each category of RN from 2014 to 2024. The problem is that the chart itself does not sit on any baseline. While I do not care for the dual axis chart, that format at least keeps an axis legend on the right side of the chart. (You still have the problem of implying certain things based on what scale you choose to use relative to the first data series.) Here, because there is no chart lines associated with the growth data, I wonder if a table below the x-axis labels would be more efficient? Home health care, a very small category, will have the highest growth (a small change from a small base will beat the same small change or even slightly bigger changes from a far larger base) but the eye has the furthest to travel to reach the 61% number from the top of the bars or the labelling.

The other component I wanted to discuss is the scatter plot that compares the number of jobs to their average salary.

Bursting these bubbles…
Bursting these bubbles…

But this is a bubble chart, not a scatter plot, and so we have a third variable encoded in the size of the dot/bubble. The first thing I looked for was a scale for the size of the circles. What magnitude is the RN circle vs. the Personal Care Aides circle? There is none, but unfortunately that seems to be a common practice with bubble chart. But after failing to find that, I noticed that the circles decrease in size from right to left. That was when I looked to the legend and saw the y-axis in numbers of jobs and the x-axis in average salary. But then the circles are sized in proportion to the average salary of each profession to the other. In other words, the circles are basically re-plotting the x-axis. The physical therapist circle should be roughly twice as large, by area, than the vocational nurses. But we can also just see by the x-axis coordinates. The bubble chart-ness of the chart is unnecessary and the data could be told more clearly by stripping that away and making a straight-up scatter plot where all the circles are sized the same.

Credit for the piece goes to Christina Animashaun.

The Economic Impact of Hurricanes

Yesterday Hurricane Ophelia hit Ireland and the United Kingdom. Yes, the two islands get hit with ferocious storms from time to time, but rarely do they enjoy the hurricanes like we do on the eastern seaboards of Canada, Mexico, and the United States.

Earlier this hurricane season the US had to deal with Harvey, Irma, and Maria. And in early October the Wall Street Journal published a piece that looked at the economic impact of the former two hurricanes as exhibited in economic data.

Overall the piece does a nice job explaining how hurricanes impact different sectors of the economy, well, differently. And wouldn’t you know it that leisure and hospitality is the hardest hit? But then they put together this stacked bar chart showing the impact of the hurricanes on both Florida/Georgia and Texas for Irma and Harvey, respectively.

I just want a common baseline…
I just want a common baseline…

The problem is that the stacked bar chart does not allow us to examine each hurricane as a specific data set. Because the Harvey data set is first, we have the common baseline and can compare the lengths of the magenta-ish bars. But what about the blue sets for Irma? How large is natural resources and mining compared to professional and business services? It is incredibly difficult to tell because neither bar starts at the same point. You must mentally move the bars to the same baseline and then hope your brain can accurately capture the length.

Instead, a split bar chart with each sector having two bars would have been preferable. Or, barring that, two plots under the same title. Then you could even sort the data sets and make it even easier to see which sectors were more important in the impacted areas.

Stacked bar charts work when you are trying to show total magnitude and the breakdowns are incidental to the point. But as soon as the comparison of the breakdowns becomes important, it’s time to make another chart.

Credit for the piece goes to Andrew Van Dam.

How Bad is the Rohingya Crisis?

Pretty bad.

Less than a week after posting about the satellite views showing entire villages razed to the ground, we have a piece from the Economist looking at refugee outflows. And they are worse than the outflow of refugees during the Rwandan genocide back in 1994.

To be clear, they are not saying that nearly a million people have been killed—though there is quite a bit of evidence to say the Burmese security forces are cleansing the state of Rakhine of one of its primary ethnic groups.

That is a lot of people fleeing Burma
That is a lot of people fleeing Burma

But when it comes to the chart, I am not quite sure what I feel about it. It uses both the x and y axis to show the impact of the refugee outflow. But the problem is that we are generally rubbish at comparing areas. Compounding that, we have the total number of refugees represented by circles, another notorious way of displaying areas. (Often people will confuse the circle’s area with its radius or diameter and get the scale wrong.)

I wonder, would a more straight forward display that broke the dataset into two charts would be clearer? What if the designers had kept the Marimekko-like outflow display, but represented each crisis and its total outflow as a straight bar chart to the right of the timeline? (I do think the timeline is particularly good context, especially since it highlights the earlier persecution of the Rohingya.)

Credit for the piece goes to the Economist’s Data Team.

Irma’s Impending Arrival

Your author is on holiday today and is actually writing today’s post on a Thursday night train to Boston. But by the time he returns late Sunday night—a Monday morning post is not guaranteed—Hurricane Irma will have likely made landfall somewhere along the Florida coast.

Thursday the Guardian published a nice article looking at the potential tracks for Irma. And while the specific routes will certainly be amended and updated over the weekend, the article is worth looking at prior to Irma’s arrival at Florida. As of my writing the track has shifted ever slightly westward and the current predicted path looks for Irma to land south and west of Miami. Ergo this screenshot is already a little outdated.

The three little wolves will huff and puff…
The three little wolves will huff and puff…

The remarkable thing about this graphic, which is just a cleaner version of the standard meteorological maps through more a more considered palette, is that there is not just one path of winds, but three. Following quickly on the heels of Irma are Katia and Jose, the latter the one taking the nearly same path as Irma while Katia spins towards Mexico.

But the graphic I really wanted to look at is the one ending the piece.

A very wide range of countries
A very wide range of countries

This looks at the countries in Irma’s path as of Thursday morning. What I do not understand is the vertical axis of the bars. What does the height represent? To simply show the rank of countries able to cope with natural disasters, a more straight-forward table could have been used. A dot plot would also make some sense, but again, it would require an understanding of the underlying metrics driving the chart.

The graphic is saved by the annotations, in particular the more/less vulnerable directional arrows. Because I do not understand why countries are grouped into the particular buckets, I find the coloured bins out of place.

I think the concept of showing the most vulnerable countries is terribly important, however, the graphic itself needed a little more thought to be a little more clear in presenting the concept.

Credit for the piece goes to the Guardian graphics department.

Rising Tides, Rising Disasters?

One more day of Harvey-related content. At least I hope. (Who knows? Maybe someone will design a fantastic retrospective graphic?) Today, however, we look at a piece from the Economist about the rising number of weather-related disasters, but thankfully falling numbers of deaths. The piece has all the full suite of graphics: choropleths, line charts, and bar charts (oh my!). But I want to look at the bar chart.

A timeline of disaster causes around the world
A timeline of disaster causes around the world

I cannot tell from this chart whether there has been any change in the individual elements, the meteorological, hydrological, or climatological disasters. And unfortunately stacked bar charts do not let us see that kind of detail. They only really allow us to see total magnitude and the changes in the element at the bottom of the stack, i.e. aligned with the baseline. So I took their chart and drew the shapes as lines and realigned everything to get this.

My take
My take

You can begin to see that meteorological might be overtaking hydrological, but it is too early to tell. And that right now, climatological causes are still far behind the other two.

Credit for the piece goes to the Economist Data Team.

Credit for mine goes to me.

Harvey’s Rainfall Part Two

Let’s consider today a follow-up to yesterday’s piece. (No, I do not believe I have ever done a follow-up piece, but why not start now all these years later.)

Yesterday we looked at the Post, Journal, and Times for their coverage of the fallen rain amounts in southeast Texas. But at the time, we only had actual totals from the Post and Journal. The Times had only produced a projection map. The Times piece yesterday was perhaps the most underwhelming of the three, though it certainly did some things correctly, namely it was small, simple, and quick to get the reader to the point that Houston was likely to be flooded by storm’s end.

Well that had changed by the time I got home last night.

The Times' graphic
The Times’ graphic

What is different about this piece? Well this one is an animated .gif showing the cumulative rainfall. In other words, Texas starts dry and every hour just makes the map bluer and bluer. An additional feature that I find particularly useful is the dot map, which indicates where the heaviest rain was falling in each hour. Especially early on in the event, you can see the bands of rain sweeping in from the Gulf.

The bins also work better here, though I wonder if more segregation or a different palette would have worked a bit better. But, my biggest critique is the same I have with many animated .gifs: the looping. And unfortunately I do not have an easy solution. You certainly need to see it loop through more than once to understand the totality of the rainfall. But then I really do want to be able to examine the final map, or at least final as of 03.00 today.

Anyway, this was a really nice piece that should have been showcased alongside the others yesterday.

Credit for the piece goes to Gregor Aisch, Sarah Almukhtar, Jerey Ashkenas, Matthew Bloch, Joe Burgess, Audrey Carlsen, Ford Fessenden, Troy Griggs, K.K. Rebecca Lai, Jasmine C. Lee, Jugal K. Patel, Adam Pearce, Bedel Saget, Anjali Singhvi, Joe Ward, and Josh Williams.

Harvey’s Rainfall Totals

Hurricane Harvey landed north of Corpus Christi, Texas late Friday night. By Monday morning, Houston has been flooded as nearly two feet of water have fallen upon the city, built on and around wetlands long ago paved over with concrete. Naturally the news has covered this story in depth all weekend. Even leading up to it, when I was still posting eclipse things, various outlets had projections and why we should care graphics. But as the storm begins to move back into the Gulf—only to move back inland tomorrow—I wanted to compare some of the graphics I have been seeing.

Of course, not all graphics are the same, let alone cover the same things. So this morning we are looking at just the rainfall total maps of a few different outlets.

From the Washington Post, we have the following graphic.

The Post's rainfall graphic
The Post’s rainfall graphic

The palette chosen performs well at quickly scaling up to the record level of rainfall, i.e. the 20+ inches realm, but quickly shifting from the green–blue palette into dark purples.

Then we have the Wall Street Journal’s graphic.

The Journal's graphic
The Journal’s graphic

Here we have a more familiar blue–red diverging spectrum. The point of divergence set to 20 inches.

Lastly, we have the New York Times graphic. Though in this case, it’s not an exact like-for-like comparison. I could not find a graphic mapping total rainfall, instead this is for projected rainfall totals. But the design is for the same type of map, i.e. how much rain falls in a location.

The Times' graphic
The Times’ graphic

The Post takes the closest approach to the true continuous spectrum palette, where the shift from dry to drenched is gradual. It makes for a smoother, more blended looking map. Somewhere around that 20 inch point, however, the palette shifts from the green to blue range to purple. It emphasises the record-hitting point, but otherwise the totals are presented as more fluid. Perhaps correctly since rain does not neatly fall evenly into pixels.

By comparison, the Journal segments the rainfall totals into bins of blues. The scale is not even, the lighter blues incorporate two inches, the darkers upwards of five. And then again, like the Post, separate 20+ as a different colour, here switching to reds.

Lastly the Times keeps to a simple segmented bin palette of all blues. 20+ inches is rendered is just a dark blue.

Each map has pluses, each has minuses. The Times map, for example, is simple and quick to understand. Southeastern Texas will be wet by the middle of next week. If your goal is only to communicate that point, well this map has done its job. It is worth noting, again, that this is a map of projections. Because the other thing missing from this map is the storm’s path. So if the goal were to showcase the rainfall along the storm’s path, well this graphic does not accomplish that nearly as well as the other two.

The Post and the Journal both show the track of the storm. The Journal takes it one step further and plots its projected course through Thursday. This helps us really see if not understand the east side problem of hurricanes. That is, the eastern quadrants of hurricanes typically experience the heaviest amounts of rain. And as the darker portions of the map all fall to the north and east of those lines, it reaffirms this for us.

So what really differentiates the two? The colour palette and its application. The Post’s palette is more natural as, again, rain does not fall neatly into bins and instead makes for blurred and messy totals across a map. Separating the heaviest rains into the purples, however, makes a lot of sense as that amount of rainfall, as we are seeing this morning, makes for a mess in Houston.

But the point of a graphic is to translate nature and the observed into a digestible and pointed statement of the observed. What should I learn? Why should I care? The Journal, like the Post, does a fantastic job of splitting out the 20+ inch totals by using a divergent palette. But instead of blending into that colour, the distinction is sharp. And then below that threshold, we get rainfall totals segmented into just a few bins. These help the reader see, also more starkly because of the selection of the specific blues, just where the bands of heavy rain will fall.

I do want to point out, however, that all of these maps occur in articles with lots of other fantastic graphics that visually explore lots of details about the story. And in particular, I want to highlight that the normal bit where I state the credits includes a lot of people. Creating a whole host of graphics to support a story takes a lot of work.

Credit for the Washington Post piece goes to Darla Cameron, Samuel Granados, Chris Alcantara, and Gabriel Florit.

Credit for the Wall Street Journal piece goes to Bradley Olson, Arian Campo-Flores, Miguel Bustillo, Dan Frosch, Erin Ailworth, Christopher M. Matthews, and Russell Gold.

Credit for the New York Times piece goes to Gregor Aisch, Sarah Almukhtar, Jeremy Ashkenas, Matthew Bloch, Joe Burgess, Audrey Carlsen, Ford Fessenden, Troy Griggs, K.K. Rebecca Lai, Jasmine C. Lee, Jugal K. Patel, Adam Pearce, Bedel Saget, Anjali Singhvi, Joe Ward, and Josh Williams.

Alaskan (im)Permafrost

I woke up this morning and before breakfast I opened the door to bring in today’s edition of the New York Times. I enjoy reading the paper, or at least a few articles, over breakfast (and more often than not preparing a post for here at Some of the best days are when I open the door and find a giant piece of data visualisation there above the fold. Other images, for example the other day’s eclipse coverage, also strike me, but as someone who visualises data as part of his career, I particularly enjoy things like maps. (I should point out I also do editorial design, so things like this layout are even closer to the intersection of my interests.)

Lo and behold, this morning I opened the door and we had the shrinking permafrost of Alaska this morning.

Now that is basically it. I have a crop of the map at the end here, but the map was the extent of the data visualisation in the article. Indeed, other articles in today’s edition carried more interesting graphics—I took photos to hopefully circle back—but the nerd I am, I really do get a kick finding a paper like this in the morning.

The graphic itself occupies half the space above the fold and the bright cyan and magenta steal the user’s attention. Even the headlines of the other articles recede behind the Alaska maps.

White space around the maps subtly helps focus attention on the piece. To be fair, the shape of Alaska with its archipelagos and bays along with the southeast extension help to create that space. A more squarish shape, say Colorado, would not quite have the same effect.

If I had to critique anything, I might have placed the city labels, especially Fairbanks, and the state label elsewhere to enhance their legibility. But at that point, I’m really just quibbling around the edges.

Red means it's warming up
Red means it’s warming up

Credit for the piece goes to Jeremy White.