charts

Misleading Infographics

I find few things more annoying than thoroughly misleading infographics. At the weekend, I was flicking through the latest copy of The New Statesman, and came across an advertorial published by Western Union addressing overseas trade.

The most startling part of the two-page spread was an infographic showing the top UK export destinations.

Now leaving aside the suggestion that WU Edge seems to present itself as the main route for this trade to be taking place, the most startling thing that instantly struck me was the scale of the US compared with everyone else. The size of the circle is significantly larger than any other circle on the page.

But hang on. If the US is worth $66.5bn, and Germany is worth $46.4bn (about 70% of the US), why does the German circle not look like it’s about 70% of the US one?

Let’s find out.

First of all, there are sometimes optical illusions, so I took a ruler out and roughly measured the diameters of the circles on the paper. (All more measurements and calculations from here on are a bit rough, with lots of rounding. However, the principles are correct.)

So the US circle is 28mm across, whereas Germany is 20mm, Switzerland 13mm and so on.

My suspicion is that they’ve sized these circles according to diameter or radius rather than area. Let’s see if I’m correct. Bearing in my mind I’m measuring roughly, here are my results:

If we assume a diameter of 28mm is equivalent to $66.5bn. then you can see that broadly speaking the other widths are in line with the printed numbers on the page give or take the odd billion.

But that’s a wrong way to do things!

If we were being presented with a bar chart, then the length of the bar would be fine. But we have circles here, and if we use radius (or diameter) as our measure, then the area increases exponentially. That’s because, as any schoolboy knows A = Πr2 (or Area = Π x radius2).

To show how this misleads, consider the US circle. The area of that 28mm circle (14mm diameter) is 616mm2.

That implies that $1bn = 9.3mm2.

But if we work back from that, then Germany’s circle should be 23.4mm rather than the 20mm it actually is.

That might seem a small difference, but with a circle it’s suddenly larger as this hand drawn (no compasses available) image shows.

More to the point, if you take a smaller example like China which in the printed chart has a width of 12mm, the calculations show that is should have a width of about 18mm.

An 18mm circle compared with a 12mm circle is significantly larger in appearance.

I’m not saying that anyone politically wanted to make the US look larger than the other countries, but misuse of circles, not taking into account radius, actively makes that impression.

Infographics are great, if they handle data responsibly.

This was a bad example and as a consequence presents a highly misleading picture.

Why Doesn’t The Chart Show Adopt An Opening Weekend Model?

This week came news that with the change in release dates of music – shifting from Mondays to Fridays on a global basis – the Radio 1 Official Chart Show will likewise shift. So that instead of going out at the end of the previous week (i.e. the previous Sunday to Saturday) it will air on Friday evenings and represent sales/streams from the previous seven days (i.e. the previous Friday to Thursday).

While this sort of makes sense, I think I’d have shaken up how UK charts are compiled.

The reason for the move to Fridays is apparently because so much more music is sold at weekends, and they can sync worldwide releases together on that day. I must admit that I used to enjoy a Monday lunchtime mooch around HMV – while there still were a reasonable number of HMVs – but I understand that Friday makes sense in an online world. Video games have been released on Fridays for years for the same reason.

But it’s all about immediacy these days, and I’d look to how cinema works. Instead of getting a full week of box offices, we actually get the opening weekend the following Monday/Tuesday. Yes, some distributors mess around by having “previews” the weekend before, or releasing a film on Wednesday to give them 5 rather than 3 days opening. But effectively we’re looking at three day totals when we see cinema top tens. Those who go to the cinema on Monday-Thursdays get added into the following week’s release.

So I’d do the same with the charts. I’d shift to an “opening weekend” model. You’d still get a Sunday teatime chart show, with the advantage that you could highlight the big new releases from Friday and see where they’ve got to while they’re completely fresh. Sales data can be generated pretty instantly these days, so the chart could be compiled at the last minute, giving two and a half days of data. And Sundays are just better for listening than Friday early evenings when there are more distractions.

Then in place of the current Wednesday “Chart Update”, I’d run something that would have a better name than “the consolidated chart” which actually had a full week’s worth of data. Ideally it’d run Thursday evening, but Friday lunchtime/afternoon would be fine too.

It’ll be interesting to see what happens with the commercial Big Top 40. I reckon they’ll go alone and stick with the Sunday show and do a three day model. That way, they can get in first with big new singles going to number one five days before they do the same on Radio 1. That’s got to give them a massive competitive advantage, even though I realise chart shows aren’t what they once were.

I’m old enough to be the sort of person who compiled my own “Now”-style cassette compilations – sitting there with my JVC radio cassette player, finger hovering over the Pause button to remove as much of Richard Skinner or Bruno Brookes as possible, then realising I didn’t want that song and carefully re-spooling the cassette with a pencil to get to the precise point for the next single. The chart show was also basically the background to my weekend homework – it was Sunday early evening and obviously I still hadn’t done it!.The next generation needs a chart to do its homework to!

[Updated to correct chart collection days. Cheers Sam]

Bad Visualisations

There’s a really interesting piece to write about Taylor Swift, Spotify and iTunes revenues, and the music industry in general.

But this isn’t it.

I’m writing this because TechCrunch has an interesting piece about European Spotify earnings overtaking iTunes earnings.

The piece reports that Kobalt, a publishing company that collects music royalties on behalf of its clients, is collecting more revenues from Spotify now than it is from iTunes – the first time this has happened. Whether this is specific to Kobalt’s clients or is indicative of the whole industry in Europe isn’t clear. But the story it itself is highly relevant.

The story’s point is illustrated by the following chart:

Kobalt1
Source: TechCrunch

This is a terrible chart.

Here’s why:

Kobalt has been trading since 2001, but this chart very much suggests that Kobalt was collecting precisely no revenues from iTunes or Spotify prior to Q3 2013. This is very unlikely. The chart also suggest that iTunes and Spotify revenue has fallen off a cliff post Q1 2014. This is also very unlikely.

Essentially this chart has six data points, and it shouldn’t have been presented using an area chart like this. Personally I think a simple bar chart would have sufficed. As it stands, this is grossly misleading.

Here’s a second chart in the piece:

Kobalt2

I think this might be even worse. It certainly suggests that in the period prior to Q1 2012, Kobalt’s share of publishing income was zero. I suspect that was not the case – it’s something that has probably been growing over time given the age of the company and the fact that Spotify has been around since 2008 and had 10m users in 2010.

And again, it also suggests that by Q2 2014, their revenue share had plummeted to zero. Whereas I suspect that they’re a growing business. This is just another really bad chart.

Seriously, if you can’t do charts properly, just publish a table. There are three numbers in this chart, and charting them like this is simply useless.

Just because your graphics package lets you publish some data in a funky visualisation, it does not mean it’s right for the data. Visuals should help tell your story, not distort the data and ultimately misrepresent it.

How Many Listened?

According to a story in Media Guardian, “an estimated 5 million” people tuned in to hear who would be the Christmas number one yesterday during the various chart shows.
There are two main chart shows these days: the Radio 1 version which is considered the official chart, and the Big Top 40 chart which runs on dozens of local commercial radio stations and during which you can actually affect the top ten chart placings by buying songs during the show.
Radio has a couple of problems with listening figures for one-off shows: RAJAR only measures audiences over three month periods, and then publishes those figures at something of a delay. So even if one show achieved four times the audience of the regular show, when averaged over a thirteen (or twelve) week period, that audience surge is flattened out. This is even more the case with the Big Top 40 chart which has 6 month weighting meaning that the numbers are derived from the previous 12 weeks’ performance.
So the 5 million figure is a complete (educated) guess.
To be fair, that’s what John Plunkett’s piece says, and the figure comes from Mark Goodier:
But Goodier estimated that the combined audience for the Radio 1 chart show – yesterday hosted by Scott Mills – and its commercial radio rival, the Big Top 40 , could have topped 5 million.
How did Goodier get to that figure? Well he might have looked at the audiences of Radio 1 and the Big Top 40 shows at that time. Between 1845 and 1900 on Sundays, Radio 1 is heard by 748,000 listeners, while the Big Top 40 chart is heard by 968,000 listeners across its network of 139 FM stations as well as various digital outlets*.
So something like 1.7m people usually hear the number one. Goodier is speculating that around 3 times as many people heard yesterday’s chart.
I think that he might actually be being a little conservative. The two songs battling for the number one sold around a million copies between them. Ordinarily a number one sells much less than this (perhaps by a factor of ten if this table from Wikipedia detailing download only sales is to be believed).
In summary – nobody knows how many people listened yesterday. This is a bad time of year to do any kind of research (RAJAR takes a break for a couple of weeks), and unless somebody like the BBC has commissioned some, we’ll never know.
I think that Goodier is actually being conservative given the many millions who saw Joe win X-Factor the week before, allied with the hundreds of thousands of Rage and Joe sales achieved. I’d put the figure a bit higher perhaps at around 6 or 7 million. But I have no real proof either way. So it’s a fair guestimate.
* Note that I’ve used 6 month weighting for the Big Top 40 figures, but only 3 month weighting for Radio 1, in line with their respective RAJAR reporting periods. Source: RAJAR/Ipsos MORI/RSMB period ending September 2009.