# Improve my bar graph

April 1, 2006 11:16 AM Subscribe

How can I better present my information than a bar graph with both very tall and very small bars?

I have been given the task of presenting some information visually. The idea is that there are five or ten problems, and each had somewhere between 10,000 and 1,000,000 possibilities, but each has been reduced to around 100 possibilities.

Right now I have a bar graph with 5 bars going as high as 1,000,000, each next to a bar around 100 high, but it doesn't convey much information -- the lower bar is just a pixel line at the bottom of the graph. Example here. There is a grey bar next to each black bar but it may be so small you can't even see it.

Is there a better way to present information like this? Maybe something other than a bar graph?

I have been given the task of presenting some information visually. The idea is that there are five or ten problems, and each had somewhere between 10,000 and 1,000,000 possibilities, but each has been reduced to around 100 possibilities.

Right now I have a bar graph with 5 bars going as high as 1,000,000, each next to a bar around 100 high, but it doesn't convey much information -- the lower bar is just a pixel line at the bottom of the graph. Example here. There is a grey bar next to each black bar but it may be so small you can't even see it.

Is there a better way to present information like this? Maybe something other than a bar graph?

If you represent the data as squares or cubes with areas/volumes proportional to the number of possibilities, then you will be able to present things omre easily, and still keep the image more intuitive than using an exponential scale.

posted by nowonmai at 11:26 AM on April 1, 2006

posted by nowonmai at 11:26 AM on April 1, 2006

i think that it would help us to know what kind of data you're trying to represent.

posted by exacta_perfecta at 11:31 AM on April 1, 2006

posted by exacta_perfecta at 11:31 AM on April 1, 2006

The exact nature of the problem is kind of an ugly computer network thing. The basic idea is that there is a set of problems, where each problem has hundreds of thousands or millions of possible answers, and the only way to figure out the real, actual answer is to try every possibility. We've found a way to reduce that down to a few hundred possible answers.

I want to emphasize how great the improvement is, so I'm a little leary of something like a log scale graph. nowonmai's idea is just the sort of thing I'm looking for; I'm trying that out now.

posted by event at 11:43 AM on April 1, 2006

I want to emphasize how great the improvement is, so I'm a little leary of something like a log scale graph. nowonmai's idea is just the sort of thing I'm looking for; I'm trying that out now.

posted by event at 11:43 AM on April 1, 2006

if 2d isn't enough you could go to 3d - a near-isometric image of a cube with a tiny cut-out cube in the near corner showing the relative size.

posted by andrew cooke at 11:51 AM on April 1, 2006

posted by andrew cooke at 11:51 AM on April 1, 2006

How about a cut?

Scale goes from

0 - 250 //cut// 100,000 - 2,500,000

posted by PurplePorpoise at 12:00 PM on April 1, 2006

Scale goes from

0 - 250 //cut// 100,000 - 2,500,000

posted by PurplePorpoise at 12:00 PM on April 1, 2006

Loto's right.

Change your y-axis to a log scale. Here's a quick comparison in Excel. Both charts use the same data.

posted by SAC at 12:13 PM on April 1, 2006

Change your y-axis to a log scale. Here's a quick comparison in Excel. Both charts use the same data.

posted by SAC at 12:13 PM on April 1, 2006

Or, present a graph where the y-axis is orders of magnitude. That is usually far more important for CS applications that the actual value. And depending on your audience might be easier to communicate than a log scale.

posted by Tallguy at 12:16 PM on April 1, 2006

Tallguy, unless I misunderstood what you wrote, that's what log-scale is. Excel automatically uses log base 10.

posted by SAC at 12:37 PM on April 1, 2006

posted by SAC at 12:37 PM on April 1, 2006

rats. was hoping you'd have to go to higher dimensions. ;o>

so is this some cool new advance in integer programming? any chance of some linky goodness?

posted by andrew cooke at 1:55 PM on April 1, 2006

so is this some cool new advance in integer programming? any chance of some linky goodness?

posted by andrew cooke at 1:55 PM on April 1, 2006

It's a network security thing -- doing forensic work on intrusions. If and when the paper gets published, I will certainly put up a link.

posted by event at 3:49 PM on April 2, 2006

posted by event at 3:49 PM on April 2, 2006

This thread is closed to new comments.

posted by Loto at 11:20 AM on April 1, 2006