Skip

Ooh, look at all the pretty data.
July 31, 2008 4:30 PM   Subscribe

Stream graphs, or stacked graphs, are a new form of (sometimes interactive) visualization that present data in a fluid timescale format. For example, the NY Times website has a graph showing the box office receipts from 1996-2008. There's a Twitter streamgraph based on keywords. Here's one of all the musicians a Last.fm user has listened to over time. Track the popularity of baby names back to the 1880s. Possibly the most striking, if not necessarily intuitive, is this visualization of US population by county, 1790-2000. There's already an academic study of the technique.
posted by desjardins (27 comments total) 20 users marked this as a favorite

 
Huh. That's pretty nice. I'd seen the baby name thing before, but the box office example is really pretty striking.
posted by cortex at 4:35 PM on July 31, 2008


Eh? Stacked graphs are new? The box office one seems quite pretty and visually very different from others I've seen though.
posted by Artw at 4:39 PM on July 31, 2008


Here's another, Tufte-inspired version of box-office visualization I just came across today. It reminded me of having seen the NYT link a few months back. That one is definitely more striking, but this one is cool in a more minimalist, non-Flash kind of way.
posted by Mr. Palomar at 4:40 PM on July 31, 2008 [2 favorites]


The census data ones are quite pretty too, though I suspect they are useless for anything.
posted by Artw at 4:41 PM on July 31, 2008 [1 favorite]


Yeah, I think it's more the fluid look that most of the examples present that feels like a Clever New Thing—with the box office example, it does a really nice job of capturing the "envelope" of the films' revenue in context, with the tails slipping off into the center like that.

If there's any good freely-available software for producing something that slick, I might have to have a go at like mefi comment activity over time.
posted by cortex at 4:42 PM on July 31, 2008


The first person to make a box office-esque stacked graph for MetaFilter posts wins one (1) internet.
posted by turgid dahlia at 4:43 PM on July 31, 2008


Related to the last.fm graph: make your own.
posted by chrominance at 4:45 PM on July 31, 2008 [1 favorite]


(I actually made a very rudimentary not-flowy-looking stacked graph of comment activity across subsites a while back, but it doesn't capture anything as interesting as per-thread activity, just subsite distribution as a whole, and it's also hella ugly.)
posted by cortex at 4:48 PM on July 31, 2008


On the box office receipts one, is there significance to whether the film shows above or below the center line? If not, why is there an up and down differentiation -- stacking everything upward would give a better picture of total receipts of all films combined.

And, I don't think this is really new, except for the interactive elements made possible by an online version. I recall studying stacked graphs in science and geography books in the 1950s and 60s, and I'm sure they were around before that.
posted by beagle at 5:06 PM on July 31, 2008


I read that as steam graphs, and immediately thought "I bet they're just regular graphs with some brass doodads tacked on..."
posted by UbuRoivas at 5:07 PM on July 31, 2008 [1 favorite]


On the box office receipts one, is there significance to whether the film shows above or below the center line?

For the pretty, would be my guess. I suspect Tufte would call it noise.
posted by Artw at 5:08 PM on July 31, 2008 [1 favorite]


It reminds me immediately of this graph of popular music geneology, which I had a hard time tracking down until I turned up this MetaFilter post which I somehow missed a few days ago.
posted by hattifattener at 5:18 PM on July 31, 2008


Check out June of 1993 and look at Jurassic Park. Is this a (for lack of a better word) 'typo' on the part of the graph makers, or was Jurassic Park really kept in theaters from June of 1993 until October of 1994? The following year, check out Lion King and Forest Gump. And Toy Story in 1995. I mean, many big movies I can see lasting from summer to Christmas, but over a year?

Compare that behavior to more recent years. Pirate of the Carribean 2 for example, in 2006. Thanks to DVD and other variables, over a year of exposure in movie theaters just doesn't happen much anymore.
posted by ZachsMind at 5:26 PM on July 31, 2008


That genealogy of pop/rock music that hattifattener linked to, while also neat, reminds me immediately of the kids' book Cloudy With a Chance of Meatballs.
posted by Mr. Palomar at 5:31 PM on July 31, 2008


desjardins, the academic link is interesting for me on two fronts. One is the visualisation technique itself, the other is the association with Martin Wattenberg. Martin is also involved in a collaborative visualisation initiative known as Many Eyes.
posted by michswiss at 5:36 PM on July 31, 2008


Check out June of 1993 and look at Jurassic Park. Is this a (for lack of a better word) 'typo' on the part of the graph makers, or was Jurassic Park really kept in theaters from June of 1993 until October of 1994?

Keep in mind they're using a logarithmic scale. So the answer is "yes, but..." It wasn't making much, only about $200k per weekend, but I guess it was profitable enough for a few hundred theaters to keep it running. I don't think most blockbusters actually peter out properly nowadays. Like your example of POTC2 was pulled from theaters and it was still making $200k.

Bug report: Waiting to Exhale is mislabeled as Mr. Holland's Opus (in December 1995).
posted by smackfu at 5:43 PM on July 31, 2008


ZachsMind: that was pre DVD, so yes, I can believe the films continued to see some small box office stream that long.
posted by beagle at 5:45 PM on July 31, 2008


On the box office receipts one, is there significance to whether the film shows above or below the center line? If not, why is there an up and down differentiation -- stacking everything upward would give a better picture of total receipts of all films combined.

My guess is that you get double the horizontal density if you use both sides of the x-axis. If you only used the top half you'd have to at least double the horizontal scale to be able to see everything.

Plus, the pretty.
posted by GuyZero at 5:54 PM on July 31, 2008


There are a very few movies with long tails, but it does happen. Check out Titanic, or My Big Fat Greek Wedding.
posted by Chuckles at 6:15 PM on July 31, 2008


Jurassic Park had long tails.
posted by UbuRoivas at 6:33 PM on July 31, 2008 [3 favorites]


From Byron & co's white paper: "...arranging the layers in a distinctively organic form."

They also reference Tufte's "macro/micro" principle.
posted by ZenMasterThis at 6:40 PM on July 31, 2008


Tufte would be rolling in his closet. Mr. Palomar points to a better solution. At first glance it looks like this National Treasure thing triumphed over I am legend. It has a knee on the damn thing yelling 'yaaarhaangaa I tha winnah'. To actually get any information out of the graph besides the relative total box office receipts for a time period requires clicking around and a ruler. The NWT graph like an awesome bowel movement - sure it's pretty, but it's mostly poop.
posted by xorry at 9:14 PM on July 31, 2008 [1 favorite]


Now I can better visualize what long tail really means.
posted by infinitewindow at 12:06 AM on August 1, 2008


I make these posts to bring out all the data geeks.
posted by desjardins at 8:13 AM on August 1, 2008


The NWT graph like an awesome bowel movement - sure it's pretty, but it's mostly poop.

I wouldn't put it quite so colourfully, but I agree.

The human eye is good at comparing distances, but really terrible at comparing areas. A long and skinny object looks bigger than a more compact, but larger object. A related problem is the fact that a circle of double the area will only appear to be 1.4 times bigger; eyes tend to judge the differences of the diameters, not that of the areas.

I've never seen a graph work which depends on area. The human brain just doesn't see things that way.
posted by bonehead at 8:22 AM on August 1, 2008


Those are pretty nice, Palomar, and a great recommendation for their creator's Vecto vector-drawing library, if there are any Common Lisp people lurking here.

A while back I did a similar stack graph of news coverage over time that started out looking quite a bit like the NYT one linked here. Eventually, for all of the reasons mentioned above and after actually reading Tufte, I decided it was really more pretty than enlightening, and the final version was much more bland (in part, too, because of performance problems with lots of SVG curves). But the Vecto graphs do a great job of being pretty without being misleading.
posted by enn at 12:00 PM on August 1, 2008


Metafilter streams.
posted by rtha at 6:08 PM on August 3, 2008


« Older Watching the Growth of Walmart Across America   |   A long time coming. Delicious. Newer »


This thread has been archived and is closed to new comments



Post