Edward Tufte's Slopegraphs

July 11, 2011 8:51 PM

What’s interesting is that over 20 years before sparklines came on the scene, Tufte developed a different type of data visualization that didn’t fare nearly as well. To date, in fact, I’ve only been able to find three examples of it, and even they aren’t completely in line with his vision.

Edward Tufte's slopegraphs.
posted by Foci for Analysis (16 comments total)

I've seen charts like this occasionally. Sometimes the cult of tufte seems to over work the coolness of the chart, sometimes with not so much actual information.

Tufte has a cool gallery in NYC of his art, some is quite clever. No charts or graph art though.


=========
-----
*************
======
posted by sammyo at 9:06 PM on July 11, 2011

Tufte developed a different type of data visualization that didn’t fare nearly as well.

I think the reason is fairly obvious. There aren't many data sets that this type of visualization works well for. Most of the examples are muddled and nearly incomprehensible.

The baseball team example is just an eyesore, and the person who made it should be forced to apologize.
posted by chimaera at 9:48 PM on July 11, 2011 [3 favorites]

This is kind of confusing and I wonder if Tufte is engaging in a rebranding exercise (perhaps in an attempt to create some buzz?). I think this style of chart is also called a bumps chart (here’s an example from the New York Times) and Tufte has written about these before. Maybe I’m missing something, but how does a slopegraph differ from a bumps chart? Tufte appears to be the only one who’s using the term “slopegraph”. Fundamentally, it’s a line chart with the all the borders and axes lines stripped out.
posted by Jasper Friendly Bear at 10:00 PM on July 11, 2011 [1 favorite]

Yeah, I've heard of this before. They're called Line Graphs. The "slopegraph" examples are just line graphs with limited data points on the X axis (like only 2) and extra labels.
posted by charlie don't surf at 10:05 PM on July 11, 2011 [1 favorite]

Well I'll be damned. The moment after I commented, a "slopegraph" came up in my RSS feed. This is not a particularly novel idea. It looks like its application is obvious if your data set only has 2 points per domain.
posted by charlie don't surf at 10:10 PM on July 11, 2011

The baseball team example is just an eyesore, and the person who made it should be forced to apologize.

Maybe yes, maybe no. I just thought I'd point out that example was made by Ben Fry, who—your wording suggests that you might not know—is pretty famous as far as data visualization people go. He is co-creator of Processing. Here's something recent he made. It doesn't contradict what you're saying, but he's no novice.
posted by -jf- at 10:16 PM on July 11, 2011

I really don't see how these are better than scatterplots. (But then again, I know how to read scatterplots, and apparently some people don't.)
posted by madcaptenor at 10:16 PM on July 11, 2011

sammyo: On the Facebook link, Bing thinks the address is in Brooklyn. Piece of crap
posted by Pruitt-Igoe at 10:17 PM on July 11, 2011

Ben Fry, who—your wording suggests that you might not know—is pretty famous as far as data visualization people go. He is co-creator of Processing.

I was unaware of that. He clearly does know what he's doing, generally. But I still think the baseball graph is ugly as all get out, and not very illuminating. Even Michael Jordan threw bricks once in a while.
posted by chimaera at 10:22 PM on July 11, 2011

But there’s still a technical problem: How do you make these damn things?

Um... he's kidding, right? He's basically described a line-chart with two samples per series. The only real addition is to add labels to each data-point, which can be done in Excel*, the one piece of charting software not mentioned. I'm sure it can be done in a zillion other charting tools too.

He mentions that the main problem is label-positioning, which is understandable, because (in Excel at least) they're not quite as pretty. What he doesn't mention is that most people would be laughed out of the room if they presented charts based on two samples and asked their audience to draw conclusions from it.

Two samples do not make a trend-line, which is what we're asked to believe here.


* Format data labels --> Label contains: Y-value, Series Name.
posted by rh at 10:23 PM on July 11, 2011 [2 favorites]

I still think the baseball graph is ugly as all get out, and not very illuminating.

It's certainly not the prettiest graph ever.

Maybe I’m missing something, but how does a slopegraph differ from a bumps chart?

From the blog you link to, the author says "The bumps chart is invented for ... ranking over time data." If he means this as a definition, then I guess this would be how they differ, since "slopegraphs" are meant for continuous data. I'm not if it is meant as a definition, though, because he calls the last graph here a bumps chart.

Yeah, I've heard of this before. They're called Line Graphs.

If I ever get to put a name on a new-ish type of graph, I'm going to call it CAPS-GIANN. Change A Plot Slightly, Give It A New Name.
posted by -jf- at 10:42 PM on July 11, 2011

Here's an example of this style of graph being done in Excel 2003.
posted by Jasper Friendly Bear at 10:50 PM on July 11, 2011

Hey all, thanks for the comments on the post. I was hoping it'd get picked up on MeFi, as the critiques would be sharper than in most other places.

Jasper Friendly Bear, thanks for that term. I hadn't heard it before, and it's great to see a whole new set of charts that fall into the bucket. As -jf- notes above, it seems that the initial intent for bumps charts is to do the forced-ranking, but the Junk Charts authors began expanding on the term. Note how they explain at first that the non-forced-rank ones are "sort of" like a bumps chart. But over time, they explain the distinction less, and then in their more recent posts, they don't even (um) draw a line between them. But the Ben Fry example and the Speed Per Dollar example are both totally "bumps charts", and I'll be sure to update the post with that info, and with the other examples from the Junk Charts site.

As an aside: Someone who read the post just sent me a note on it and mentioned bumps charts. Here's what he said:

For another example see - 'Bumps charts'.

They show results of college rowing races each May at Cambridge University (England). The river Cam at Cambridge is too narrow for traditional races - no room for one boat (rowing eight) to pass another. So instead, boats start lined up one behind the other. All start at the same time, trying to 'bump' the boat ahead, and avoid being bumped by the boat behind.

The next day bumpers exchange starting place place with their bumpees, and everyone races again. The process is repeated over several days. The bumps chart shows the final result - who is 'head of river', who 'rowed over', who achieved 'double bump' etc.

The charts were certainly around in 1958 (my year at Cambridge) and I think in 1929 (my father's year at Cambridge). They used to be published each day in the Times newspaper. I dont know if they still are

Example chart at http://www.cucbc.org/charts?year=2010&event=M&day=Fi&sex=M

Fascinating. Bumps charts.

Since making the initial post, I've heard about a few other software implementations. The one Jasper Friendly Bear mentioned in Excel, as well as software for the Mac and PC (same spot as the Excel example you linked). Also, Processing.

I've also heard from someone at Mozilla who's planning on building out something in javascript / svg. I'll be updating the post as I hear of other implementations.

Two samples do not make a trend-line, which is what we're asked to believe here.

You're not expected to make trend lines out of the slopegraph representation of your data ... at least, not continuing the same lines into the future. Line charts would be better for that. It depends on what questions you're trying to answer, and what data you're using, but slopegraphs are more for comparisons within the data set than for applying beyond the data set. And, among those comparisons, you're dealing with more than two data.

Thanks, all, for your comments on the piece. Would love to hear any other feedback / criticisms / etc.
posted by Alt F4 at 2:22 AM on July 12, 2011 [3 favorites]

For those interested in R, there is a good tutorial on how to make bump charts in R here (from the excellent Learning R blog).

I've made a bump chart once (using R), and I don't think it is useful for manyinstances. In my case it was to show who was winning at each stage of a game, and it was sort of useful, but the people I showed it too had a hard time understanding it. I think the difficulty with this type of chart, is that there are few instances where that is useful, aside from circuit races of some type with marked intervals (e.g., who is winning after lap 1 or marker 1, etc).
posted by a womble is an active kind of sloth at 6:40 AM on July 12, 2011 [1 favorite]

But I still think the baseball graph is ugly as all get out, and not very illuminating.

It's funny, as soon as I read the first graf I thought of the baseball salary chart. I think it suffers as a static image. The thing that doesn't show you is the chart is interactive, allowing you to view change over time.
posted by yerfatma at 6:49 AM on July 12, 2011

Interesting posts. I've never heard of bump charts or slopegraphs before. I'd like to put a bump chart in front of some students and see how well they can extract information from it. Goodness knows that have a hard enough time with a normal graph of quantities vs. time; would this format be harder or easier for them?

I found the spending / life expectancy chart really interesting—once I'd puzzled it out. Pairing two different types of data is a was unexpected and definitely confusing at the outset. However, in this case it contextualizes the US datapoints (or, um, line, I guess) in a much more dramatic way than the scatter plot does.
posted by BrashTech at 10:45 AM on July 12, 2011