Most Active Stories
- Le Show For July 20, 2014
- Jazz Composer Jerome Theriot Celebrates New Release; Cat On A Hot Tin Roof; Hurray For The Riff Raff
- Women Stage Protest At Hobby Lobby In Elmwood
- 'Pink Slime' Is Making A Comeback. Do You Have A Beef With That?
- State Representative In New Orleans East Sounds Call Over Coastal Erosion
Fri June 15, 2012
Putting a Friendly Face on Statistics
Originally published on Fri June 15, 2012 12:26 pm
IRA FLATOW, HOST:
Imagine that you're staring at a heap of numbers, stacks of statistic, piles of pie charts, bunches of bar graphs, trying to make sense of it. It's exhausting, like running a marathon with your brain, no? But think how many faces do you see every day, right? You see lots and lots of faces, and your brain doesn't struggle to read the sizes of their eyes or the shapes of their mouths, the styles of their hair. So why not look at statistics the same way you'd look at a face? Marc Abrahams, editor of the Annals of Improbable Research, is tell - is here to tell us why turning tough numbers into friendly faces makes data smile. Welcome back to SCIENCE FRIDAY, Marc.
MARC ABRAHAMS: Thanks, Ira. Nice to be here. This is about Herman Chernoff, who almost 40 years ago came up with a tool, scientific tool that really has been almost completely overlooked. He, at the time, was a statistics professor at Stanford, and he since moved to Harvard and MIT, and he's now 88 years old. And even he, a statistics professor, had trouble looking at a sea of numbers and picking out which ones are different, because the ones that are different are usually the ones that are important and have something of interest.
So what he realized is I can take some numbers and pick any feature of the face, say, the width of the mouth or the roundness of the eyes or how dark the hair is, and I can make that a variable. And I can use that to plot a face, a very cartoony face and use a computer to do it. So that's what he did to measure things like if you got some numbers to describe what's the condition of all the patients in the hospital today, their temperature, how much pain they're in, height, weight. And if you look at just a chart of numbers of 40 people, it's tough to see which one is strange.
FLATOW: Yeah. But if you turn them into faces, you can look at them. Let me just remind everybody that this is SCIENCE FRIDAY from NPR. I'm Ira Flatow, talking with Marc Abrahams, editor of the Annals of Improbable Research. So he's able to take the statistics and turn them into features on the faces, so will the faces look different?
ABRAHAMS: And you have - yes, you have complete choice if you're doing this about which features represent what. He did - somebody did it with forged bank notes from a country.
ABRAHAMS: Because in forgeries, apparently, forgers don't always take care to make sure that the heights are uniform or the left side is the same size as the right side. And there are all kinds of things you can measure. But again, it's a sea of numbers if you're just looking at numbers. But if it's a collection of faces and one of them is really weird, it's making a strange expression, it's got a lot more hair, it's got a longer chin, it's got rounder ears, you're going to spot that.
The thing is almost nobody noticed at the time. And there have been over the years a bunch of people who've done little experiments, and almost all of them get excited and publish a paper and then nothing happens. And the reason - I think it's the reason - why it didn't get noticed too much at the start or people didn't pay too much attention was this was back in 1973. And at the end of the paper where Herman Chernoff describes how to do this and how simple it is and the promise of this thing, at the very end of it, he says, at this time, the cost of drawing these faces is about 20 to 25 cents. Now, this is a long time ago. It's about 20 to 25 cents a face using the IBM computer at Stanford University.
ABRAHAMS: And he says, most of this cost is in the computing, and I believe that it should be possible to reduce it considerably. So 40 years ago, it was so expensive just to print out cartoon faces that nobody would consider using a technique where you had to print out a few faces. Now, of course, computing is pretty cheap.
FLATOW: Have we seen it in use anywhere now where people making the faces?
ABRAHAMS: As far as I know - and I talked to Herman Chernoff a few years ago and as far as he knows, nobody is using it regularly, but there's a blossoming, if you go on the Internet and search, of people who've done little demonstration things. There's even a cute one about four or five years ago, professor named Wang(ph) at Swarthmore, I believe, did something that The New York Times wrote about where he took statistics to describe the managers of all the Major League baseball teams - how often do they send in pinch hitters, how often do they make this kind of decision, that kind of decision. And he turned those into Chernoff faces. And you can see a sea of faces, each one represents a manager, you know, Joe Torre for the Yankees at that time and whoever else. And you can, at a glance, see which of them do something that the others don't, which of them yank their pitchers early, that kind of thing.
FLATOW: Baseball statistics would be perfect for those...
ABRAHAMS: Yeah, but any kind of statistics, even the driest, you know, even insurance data or the, you know, something about the size of cells in a particular organism, anything, weather data. There's one that somebody plotted a bunch of data for different cities - Pittsburgh, San Francisco, Wichita. You see all these different faces. One - some things jump out at you. New Orleans is very dark-haired compared to the others. When you look up and you see the hair and the darkness corresponds to how much rainfall do they get.
FLATOW: Well, Marc, we got to go, but thank you for - we'll be looking for those faces, and thanks for cluing us in on this.
ABRAHAMS: Thanks, Ira.
FLATOW: Marc Abrahams, editor and co-founder of the Annals of Improbable Research as well as founder and master of ceremonies of Ig Nobel Prize Awards. Transcript provided by NPR, Copyright National Public Radio.