Tag Archives: infovis

The chart-junk of Steve Jobs

On June 9th Apple CEO Steve Jobs will take to the stage in San Francisco to give the keynote address at his company’s 2008 WWDC. Rumours are strong that he’ll be unveiling the second generation (or at least 3G capable) iPhone. He usually does about two big keynotes a year, and they’re almost always worth watching. Undeniably the best showman in the technology industry, he packs his own reality distortion field, a charismatic glow that can convince you of the urgent need to buy Apple’s latest product if you are to live a fulfilled iLife™. The older keynotes for significant product launches like the Macintosh launch, the unveiling of OS X, and the iPhone launch are particularly good examples[1]  worth watching to see how he brings the audience into the palm of his hand. He brings them up to speed with what the company’s been doing. He then starts naturally turning the topic into a story of a missing product in the market. He’ll talk about competitors’ products (if they have any) in that space, and talk about their short-comings, shaping the descriptions so that it becomes increasingly obvious what the perfect example of that product should look like. And then he unveils it, pulling it out of his pocket, pulling a sheet off it, or pulling it out of a manilla envelope. He’s spell-binding, and it’s only obvious what a talent he has when you see other CEOs try to emulate it and fail decidedly (warning: video autoplays).

One of the props he often relies on are graphical charts to show things like how marketshare of an product has been doing, or how a new one compares with competition for performance. Data presentation guru Edward Tufte has an expression for Job’s style: chart-junk. One of the themes Tufte keeps coming back to in his book Visual Explanations (which I’m reading and really quite enjoying at the moment) is that the magician is worth studying, as an example of a performer who acts out a lie. Learn from what he does, says Tufte, and then seek to do the opposite when presenting information. So it’s appropriate when you consider Job’s style how often he’s described as a magician. And the elements of a Stevenote are in some ways closer to a magic show than they are an honest data presentation. It’s sadly to be expected from what is in essence a marketing pitch. I’m always slightly annoyed when good marketing crosses the line into subtle dishonesty, and Jobs really shouldn’t have to engage in it. His products usually advertise on their own merits well enough.

On Tufte’s forum I saw a comment in a discussion on the sins of pie graphs pointing out this photo.

It’s actually horribly blatant. The Apple segment gets the closer perspective, and somehow 19.5% ends up with a bigger surface area than 21.2%.

Wondering if he did this often, I delved off into the Engadget archives of their coverage of other Stevenotes. And found plenty of chart-junk. For each of these, spot the bias introduced by the 3D perspective, and how it always seems to fall in Apple’s favour when there’s a direct comparison to be made.

Above: It’s a performance comparison, so smaller numbers mean the faster software. Safari is Apple’s browser.

These two above aren’t so bad. But the parallax bias is there, and it is in Apple’s favour. Apple’s numbers fall the left of the vertical vanishing line, so get foreshortened.

He’s normally so good about have some numbers on a chart. In their absence, and the presence of a 3D perspective, it’s impossible to regard this chart as anything useful.

Tiger is the latest version of OS X in this case. Panther’s the previous one, and older is everything before that.

In a 2D graph you’ve got two opportunities to skew the data. Growth is good, so if you want to make the numbers at the right hand side of the graph look better, bring that end closer. But additionally in the vertical axis, we see the same parallax bias as the web browser performance comparison charts from above, so that the “Industry” figures get skewed to look even lower than the Mac’s.

This one’s angled and segmented slightly so that the Leopard (the latest version of OS X) area gets magnified, but there’s not much you can do about numbers that different.

This is the blurry successor to the first one. Apple’s the green segment. Again the upward perspective that the Apple segment benefits the most from.

Again, two perspective skews here for bias, and a style bias. The first is the obvious make-the-right-hand-side-closer angle to magnify the latest figures. Which is interesting, since the vertical bias (the vanishing point is in line with the top of the graph) is shrinking those numbers compared to the older ones. Which would appear to be counter-productive, until you read the title and realise that this is a cumulative sales graph. Which is an interesting choice in itself, since cumulative graphs are often hard to get meaningful data out of. They don’t go down after all. Which is why they chose this form for this slide. Growth is slowing. It’s inevitable with a product like the iPod, but there’s ways to soften the news. Combined with the choice of a cumulative graph, the vertical perspective now makes sense. The vertical perspective has the effect of playing down the early growth and maximising the latest, bringing the graph slightly closer to looking linear, than an S-shaped graph that’s beginning to hit the final plateau.

So come June 9 we’ll once again be invited to be spell-bound by Cupertino’s wizard, and expect more chart-junk. It’s fun to be sold a story like this, provided you’re aware that you are being sold one. It’s not a scientific presentation. Jobs is known for prizing aesthetic form, and in a way 3D charts look nicer than 2D. But I would question whether it’s honest to take this consistently a biased approach in presenting your data. And as Tufte asks in Visual Explanations, “Why lie?”

[1] Worth watching for the contrast to these celebrations of Appledom is the infamous Macworld 1997 keynote where he announced the company’s partnership with Microsoft to a congregration (any religious implications from that word are fully intended) that doesn’t buy into his forced enthusiasm. It’s rare to hear actual booing during a “Steve-note”, and the scene was recreated in the 1999 TV movie Pirates of Silicon Valley.


Using that second display: 4 news visualisations of questionable utility

For both yours and my ever decreasing attention spans, in the race to distinguish and spice up the daily news product, here’s more news, shallower, and faster.

MSNBC Spectra screenshot

Spectra from MSNBC is a pretty terrible romp into 3D. Pretty, but completely unusable and just rather useless. You select what channels of news you want, and as you do a selection of stories from each channel floats into the display as a rotating ring. It wouldn’t be so bad if you could actually click on the floating news items. But no, that does something completely unexpected, it ejects that entire ring of stores. To get to a story you want, you have to navigate via a ridiculous horizontal scrollbar. I thought we had learnt in the 90s that 3D interfaces like this just don’t work. From Information Aesthetics via Data Mining.


Moving from the realms of insanity to just the slightly overwhelming comes Newsmap, based off Google News.

Digg\'s \"Big Spy\" visualization

Digg\'s \"Stack\" visualization

From the very epitome of fickle and populist news rivers comes a selection of cool-looking, fast moving and not really that value-additive visualizations at their Labs section.

Mapped Up screenshot

Finally comes a low-key (and the most embeddable of the lot) Flash widget that just rotates geo-coded stories on a world map.

WikiStat, a Greasemonkey extension for viewing Wikipedia edit distributions

In my previous post I wanted to make a pie graph quickly, and so for the first time used Google’s relatively new charting API. It’s a pretty neat little concept, taking in all the data and parameters for the chart in a single URL and then giving you the resulting image. I thought there had to be a better use for that than a static joke pie chart. Then in a meeting of my quiet interest in information visualization and some previous experience at writing a Greasemonkey script (if you know Javascript but haven’t tried GM, do so. It’s actually quite fun and easy), I decided to have a stab at using the API in a slightly more dynamic and useful way.

It’s called WikiStat and shows the time distribution of up to 250 of the most recent edits on an article, giving you a quick insight on how recently and intensively edited a page has been edited.

WikiStat screenshot

You’ll need Firefox and Greasemonkey to use it.  If you’ve got those, then click here to install it.

I haven’t tested it with anything other than GM 0.7 and FF3b3. I do recall that Opera can do user scripts these days, so it might be able to do it too. No promises though.

I’m most interested in suggestions for improvements.


WikiStat pie chart screenshot

Added a pie chart to show approximately how many of the edits were reversions. Makes for somewhat depressing viewing when you then consider there’s an invisible but equally large and ultimately non-productive segment of the pie.