I opened Twitter on Saturday night to see indignant tweets everywhere, most of which seemed to be directed towards a plot (or, more accurately, a series of plots) that were shown as part of the UK Government Coronavirus briefing on Saturday evening.

‘Excel’ was actually trending, and you’re probably thinking, ‘Hold on, we’ve done this already’, and you’re right, we have.

But, what drew me in was seeing that the plot that was garnering the most ire was unquestionably a ggplot2 graphic.

And that disturbed me.

How can anyone not like a ggplot image?
What on earth had happened?

“She’s turned the weans against us”
Me seeing ggplot2 getting hammered on the timeline

I’m not saying it’s impossible to make a bad chart in ggplot2, but it’s easier to make a good one.

Here is the plot in question, at least, here is how it looked, at a point in time, on someone’s TV :

People were complaining they couldn’t see the axis.
There were BBC graphics superimposed.
There was a man in the bottom right doing sign language for deaf viewers. And many more complaints.

And on the face of it, these were entirely justified, if some of the screenshots were reality.
What use is a plot if it has no scales, axis labels or legend?

But hang on, that isn’t what happened – here is how the plot actually looked:

I feel out of solidarity with the analyst who made it, that it should be said that it is not a bad piece of work.

Quite the opposite, when you see it as it should have been displayed.

How was the analyst, possibly working from home, to know how this plot would look when shown at the wrong aspect ratio, on television, with loads of superimposed extras?

How do you test for that? Because I’m WFH at the moment, and I have no way of knowing how my stuff would look on national TV.

Are there any improvements that could be made?

Well, Twitter seemed to think so.

Initially, the only thing I would pick up on is the ‘bumpf’ down the bottom of the plot. It’s almost certainly mandated, and so would be outwith ¹ an individual analyst’s discretion to remove. Visually, it looks good, and would be very helpful in a meeting as the sources are documented.
Initially I thought it would reduce vital screen real-estate, by squishing the space available.
For example, the dates could be rotated on the X axis if there was more space below. That way, you could see more of the week labels.
That’s just a personal thing that I like, some folk hate rotated labels and they are probably correct.
In the end, it didn’t matter, because of how the plot was projected on screen – rotated labels would not have been visible.
TL / DR : nothing to see here.

Critiquing the critique

The following headers represent the broad categories of complaint I saw on my timeline. They are not MY criticisms.

The legend position – legend not fully visible

That turned out to be unfortunate. If it had been placed along the top of the plot, it might have been more visible, and it would have helped the general public at home decipher what they were seeing.
But on the slide, as a stand-alone graphic, it’s fine.

The palette – too colourful

People didn’t like it.

It’s one of several available in the viridis palette package.

I’m not a huge user of this one, personally, I prefer ‘viridis’ or ‘plasma’.

.

I don’t use the other palettes because I don’t like the black at the extreme end of the palettes. (And yes, I know, I did produce a ten tone black palette in metallicaRt, but that was ironic). I don’t think I’ve ever seen “cividis” being used anywhere. ²

The criticism seemed to be on the low values being yellow, and high values being black.
I don’t think that matters, as long as the scale is visible.

Maybe a completely different palette would have worked…something much simpler such as blue to orange? This is based on people understanding red, amber and green.
But these are bad choices for colour blind viewers.
Blue to orange would go some way to negate that. But that would have meant needing to plot rates of change, rather than rates (assuming the transition was blue ..white..orange).
Would people understand rates of change? Would the lighter tones transmit well on TV? I guess not.

A straight blue to orange would almost certainly involve brown, and that would be horrific.
And it’s these sort of things that the viridis palettes seek to overcome.
So I don’t necessarily agree with changing the palette, or, at least, I don’t have any better ideas.

The plot title

“Heat Maps” seemed to annoy people.
I think this title could have been the subtitle, perhaps. The title could have been punchier, and could also have told the story. “Cases are rising in all regions amongst all age groups” would have got the message across.
A title change is still a very minor nit-pick.

The plot itself

I’m not going to criticise time based heatmaps. I often use them, and I think they are very powerful.

This is an information dense graphic.

Too much for a lay audience who were impatiently waiting for a celebrity dance programme to begin?
But information density is good. Max data to ink ratio, and all that.
You’d be hard pushed to accomplish what this plot achieves by other means.

Maybe, one facet per slide, allowing the audience to see each region, and then finishing off with the overall view, might have worked?

This was a suggestion put to me, that I think is fair

But there would have been no time for any other slides.

Maps would have told the story, but not over time (unless animated), and would have required many more facets, or slides.

I don’t think they are viable here.

Line charts would have ended up being even more confusing.

I don’t think this is a bad graphic.

To be very clear, I think it’s fine.

It shows the data, it tells the story, it’s labelled and captioned.

We can see rates getting worse over time in all regions and age groups.

Notice how the font colour varies on the points depending on the palette background colour.
This is not a straightforward plot, you have to separate out the layers to get this to work.

Was this criticism fair?

It seems there is a disconnect between data visualisation best practices (which this graphic follows), and what the public expects.

This plot stuck out amongst the others with the viewing public for some reason.
I’ve seen it shared in an unflattering way on other platforms.
Maybe it was just too unusual.
After all, most analysts wouldn’t dream of using pie charts, but they are still popular generally.

There was a variety of graphics with different palettes, and so overall, the slides may not have looked coordinated, but that’s often what happens when different pieces of work from different departments or organisations are pulled together.

To cap it all off, there was an Excel RAG status disguised as a heat-map, and because it was last, it ended up taking the blame for the overall poor show.

ggplot2 be like “Got away wi’it. Just pure got away wi’the lot o’it”

This is not intended as a data visualisation critique.
I don’t see much that needs changing on that graphic, when you see it as intended.

In fact, I’m almost at the point of saying, “if this is wrong, then I don’t want to be right”.
Instead, I’ll say that if that graphic is wrong, then a lot of us are wrong on a daily basis.

So I’m pondering why something that most analysts would say was a fine graphic, went down so badly.

Maybe I am too isolated in a data analyst bubble.
It’s a bit disconcerting to see something that ticks the boxes of what would be expected amongst other analysts, bumping up so hard with the reality of meeting Joe Public.

But then I wonder if this was simply a case of something that was never intended for TV, ending up on TV.

Man goes for job interview and ends up on TV

I know in the past I’ve had slides that were not intended to be presented (e.g. additional plots that were supposed to help the presenter understand the data better) suddenly flash up on screen.
It’s not great.
It feels like the same thing happened here.

Stay safe, and all the best for whatever comes next..

¹ Outwith is a word, as any Scot will tell you.

² By anywhere, I mean Twitter / Tidy Tuesday. I’m sure it gets used. It just doesn’t seem to be widely used.