Graphics in Science

BishopBerkeley writes "Nature has an interesting nugget about the second meeting of the Image and Meaning Initiative which was held at the Getty Museum in Los Angeles. It is about the use of graphics in presenting scientific data. I am also a big advocate of using nice graphics in scientific presentations, but I also agree with Felice Franel, the founder of I-M, that not all images are meaningful scientifically. In fact, one encounters (and I am ashamed to admit that I have published) images that look nice but have no scientific import at all. One very cool Harvard physics professor, Eric Heller, produces wickedly beautiful (and meaningful) images of quantum mechanical models. These images have made the covers of Science and Nature, and are featured in his online art gallery, which was reviewed in the New York Times in 2002." And of course, any mention of graphic information should not go by without a big shout out to Edward Tufte.

Comments

Graphics in Science

[Winckle]

Fianlly, an excuse to buy that 7800 GTX!

Don't Forget The Cool Factor

[DanielMarkham]

I've struggled with the same question as a computer consultant -- do images always convey anything useful just because they are based on scientific data? I've created a lot of really cool graphs and 3-D animations, but as far as analyzing the data, most times the computer is a lot better at processing multi-dimensional data than our old Mark-1 eyeball.
But there is a cool factor involved with a lot of imaging. You can't deny that.
Probably more disturbing is when images appear to convey data when they really don't. The use of false color is a great tool to bring out detail in astonomical images, but many times is misleading to the casual observer who may not understand that the images are "doped"

Is BitTorrent Next?

Re:Don't Forget The Cool Factor

[luvirini]

Well.. complex data has to be broken down for a human to understand it.

But still.. a human eye is an extremly good tool for spotting things.. a computer can only look for the specific things you tell it to look sofr whereas an eye and a mind of someone knowledgable, will often sense something in a way that no computer can.

In most cases representing something gpahically makes that easier to grasp.

Re:Don't Forget The Cool Factor

[lars_stefan_axelsson]

I've struggled with the same question as a computer consultant -- do images always convey anything useful just because they are based on scientific data? I've created a lot of really cool graphs and 3-D animations, but as far as analyzing the data, most times the computer is a lot better at processing multi-dimensional data than our old Mark-1 eyeball

Well, I've taken a slightly different tack in my research. While the computer might be better at actually analysing the data, visualisation can be a great tool in getting the results of that analysis to the user. In my case I've visualised the states of self learning intrusion detection systems so that the user can 'see for himself' why the system operates the way it does. Making under and overtraining and false alarms visible to an extent they weren't before.

But I agree. Even though I started out (PDF) doing straight up visualisation, I've come to believe that it's the combination of computer analysis and visualisation to better match the capabilities of the human operator and the machine that's the interesting field to explore.

Re:Don't Forget The Cool Factor

[swelke]

Check out the neat pictures at The Gallery of Fluid Mechanics.

Just because I don't know what meaning a picture conveys doesn't make it nonscientific, does it?

Please make the story clear.

[Black Parrot]

> not all images are meaningful scientifically. In fact, one encounters [...] images that look nice but have no scientific import at all

Could you show that with a diagram or something?

Re:Please make the story clear.

[arose]

Sure, here you go.

Graphical information representation...

[Krankheit]

Sometimes, it is easier to demonstrate with graphics, but a powerpoint presentation (or OOo presentation) with only a few words is not good either when demonstrating to more than a few people. Your information should be represented in many ways (graphical, text) because individuals learn things differently.

Site slow

[HG Slashdot]

Here are the mirrordot links:
Nature has an interesting nugget
Image and Meaning Initiative
Eric Heller
online art gallery
New York Times
shout out to Edward Tufte

Design Museum London

[porlw]

The Design Museum in London has a whole section devoted to the presentation of information and the way bias can be introduced depending on the method selected.

They have everything from pie-charts prepared by Florence Nightingale comparing the death rates in battle vs. the field hospitals to a graphical representation of the Linux Kernel.

Well worth a look.

obligatory Soviet Russia

[Krankheit]

In Soviet Russia scientific data presents powerpoint.

favourite toolkit?

[spectrokid]

What is your favourite open source cross-platform toolkit for making scientific graphs?

Re:favourite toolkit?

[ghoti]

Excel

Re:favourite toolkit?

[splatterboy]

Adobe Illustrator - for line charts use the scatter chart instead - for the same reason you would use the scatter in excel - you have more control over the x and y axis plotting over time.

You also have an easier time saving it as a picture file of high quality, either as an .eps or using photoshop. That way the people you send it to (or documents you place it into) cant screw it up. If you are stuck with excel - always "paste special-as a picture", never simply copy and paste.

Its a bit tricky to use Adobe insead but professional medical imaging is what I do - from a graphic design standpoint - professional medical advertising. Consumer med is full of pictures of happy people cured of what ever ails them but professional med is all science, FDA approval and legal/scientific review. Every thing I do has to be checked and double checked to make sure the data points plot correctly. Adobe, especially for cross platform (I work on both mac and pc for that reason - it HAS to work) has never let me down where windows is like a 50-50 chance something will come out wrong. A bad font, line point size and always bad color.

It may take some time to learn - but you can manipulate the chart and how it looks to a much greater degree than in M$ - but you have to learn the whole app to do so -not just the chart tool.

Re:favourite toolkit?

[lunadog]

RLPlot is really nice, working towards being the opensource SigmaPlot..

... It even does error bars on coloured bar charts! (not seen that in any other graphing program on Linux, not even gnuplot). It exports nice vector graphics charts that import into Lyx nicely.

Re:favourite toolkit?

[imsabbel]

gnuplot.

Really. It took me a LONG time to come to this conclusion, mainly because it scared my away with the whole "file parsing" concept, but it has tons of features, high quality output, good TeX integration.

Coloring the Universe

[colonist]

This reminds me of this issue:

From Hubble Space Telescope pictures to the vocabulary used to describe the stars, astronomers and the media are coloring our universe, and they've been doing it for decades. While not intended to deceive, the efforts can range from the overly subjective to the absurd.

Slate explains that the raw images from space telescopes are colored with Photoshop before they are released to the public. The 'Pillars of Creation' shows the difference that color makes. You can download the free Photoshop plug-in to color your own images.

Synthesis.

[torpor]

I have long lamented the lack of visual effort in interface design, specifically in the realm in which I currently work, musical synthesizers.

One of the problems with synthesis today is that it is too scientific .. and I have concluded that one of the reasons we see waves of synth revivial occurring every few years is because that is how long it takes someone to 'grok' their synthesizer, and while we wait for that grok to occur, no use occurs.

I recently made a commitment as a synth builder to attempt to enforce a few rules on myself; one of them is the "No Label Philosophy", which basically means that if a knob needs a label in order for the user to work out what it does when they turn it, then its a poor interface design, but if it doesn't, its a strong one.

The question I have is, where are other examples of 'illustration pushing concept' in the slashdott'ers world today? Have you recently seen some examples of graphical/icon-based design being used to clearly communicate very high-order concepts to the end user? What are they? Anyone got any pointers to examples of superlative graphical interface function, where you know instinctively what is going to happen because the picture tells you so?

Re:who made this tuft guy czar

[Knuckles]

" Why does this tufte guy get so much credit (...)"

Might have to do with the fact that he was a professor of statistics, graphic design, and political economy at Yale.

so little

Did you read his 3 main books on scientific graphics (The Visual Display of Quantitative Information; Envisioning Information; Visual Explanations)? They are very insightful books with a wealth of examples that are very inspiring.

opinions on design (...) by definition are subjective matters

Bull. This might be true if you talk about art, but we are not. You can easily do experiments that show that viewers have an easier time extracting information in a specific graphic design than in others.

once we stop kow towing to the tuftewrongs, we might get somewhere

Sure, but please show specific examples where he is wrong

Understanding = images + contextual info

[G4from128k]

Graphics are especially prone to the problem of obscurity through insufficient context or shared knowledge. What is self-evident to the author, because they have worked for so long on the project, is often opaque to the viewer.

The problem is most felt in dealing with non-specialists. For example, all microscopists will instantly recognize the implications of a given visual patterns of an osmium tetroxide stain in an image. In contrast, other scientists, lay people, voters, politicians, PHBs, etc. need some grounding in what the image shows, how it differs from "normal" and what the image means. A few suggestions for improving the understandability of an image include:

1. textual summary: text creates reinforcement/redundancy
2. annotate the images: arrows, circled regions and call-outs help the viewer know what's important and what it is.
3. legends: color images, especially, need a legend or textual explanation of the color scheme.
4. supporting metadata: information such as subject, scale, time (relative to some event), etc. helps create meaningful context.
5. contrasting image pairs: Image pairs or sequences help cue viewers to the significant features or establish a pattern. Showing before & after, normal vs. abnormal, enhanced vs. non-enhanced, overview vs. detail, plain vs. peanut, etc. helps explain what's what.

A picture may be worth a thousand words, but if an image presenter wants the viewer to get the intended thousand words then a little extra annotation, metadata, and context can help.

Graphics for Processing

[cspring007]

I am a master's student in the geo-sciences and my thesis requires that i process and handle a tremendous amount of data
Most of the data that i use is spatial: topography, bathymetry, salinity concentrations..
Anyway, my point is that after i write some code to process the data
(I am developing an ecological model that tracks subsidence of marshland based on a whole bunch of environmental and geophysical parameters)
the best and easiest way for me to verify the output is reasonable is to draw a picture of it. I have spent probably 60% of my time writing software that displays the data in a graphical format.
However, this is only to verify that my data is close to accurate.. like say everything looks like it should.
You still can't beat some statistics for really checking the quality of the output.

Also, after watching a large number of presentations on theses, scientific studies, etc.. i would say that 0.05% of those presenters know nothing more of scientifc graphing than pushing buttons in xcel and seeing the nice graphics that pop up.
I mean, most of them dont even change the default graphic colors, so they are up there, talking about something and behind them is that crappy Xcel purple color.

Think simple and elegant.

[Assassin bug]

In my experience in science keeping graphics very simple is best. I usually hope to have the audience leave my presentations with three adjectives in mind when they critique it: simple, clean, and creative. Assuming that you have followed rules of grammar and your scientific method is sound, a simple yet innovative presentation can make a good memory. Your data will be well understood and remembered. I absolutely detest the obligatory sequence data slide that creeps into many science presentations. Surely a creative scientist will someday discover a better way to effectively communicate sequence data in a presentation. And, how many people are going to stand at your poster for 4 hours to hand-copy all of your sequence data?

Animated and 3D graphics

[ColorTheory]

This page: http://www.jimworthey.com/jimtalk2004nov.html is the graphics that I used for a talk last year. As you read through, you'll see 3D pictures and animated graphics. When you see a 3D graph with a border, that links to a VRML pic that you can zoom and rotate. For free VRML viewer see http://www.parallelgraphics.com/ for example.

Absolutely Shameless Plug

[durandal61]

I'm the author of an easy to use open source C++ library that helps bridge the gap between your science and a final high quality image, and I thought I might point it out, since it's relevant to the topic.

PNGwriter was originally written with scientists in mind. The need to create an image from the result of a scientific computer simulation arises as a natural part of scientific programming. Getting the data out of the program and into a high quality image in an efficient way can sometimes be hard, especially if the user is not a very experienced programmer. The methods used can often be highly inefficient or too complex to be feasible.

PNGwriter is a very easy to use open source graphics library that uses PNG as its output format. The interface has been designed to be as simple and intuitive as possible. It supports plotting and reading in the RGB (red, green, blue), HSV (hue, saturation, value/brightness) and CMYK (cyan, magenta, yellow, black) colour spaces, basic shapes, scaling, bilinear interpolation, full TrueType antialiased and rotated text support, bezier curves, opening existing PNG images and more. Documentation in English and Spanish. Runs under Linux, Unix, Mac OS X and Windows. Requires libpng and optionally FreeType2 for the text support.

It has been packaged for or is a part of Debian (stable), Ubuntu, Arch and FreeBSD.

The website is available in English, Spanish and (in summary form) in Japanese, and contains many examples, an online version of the PDF manual, a FAQ section and more.

Take a look:

http://pngwriter.sourceforge.net/

Hope you find it useful!

diagrams for discrete process modeling

[rp]

As a programmer and sometimes teaching assistant I've been doing a lot of stuff with techniques to represent the structure of information (ER models, UML class diagrams, RDF, etc.) and of discrete processes (state machines, flow diagrams, Petri nets, UML activity diagrams, UML message sequence charts, etc.)

Considering the popularity of such techniques I find it odd how little material I have encountered on their actual useability, compared to other forms of representation. There still appear to be hordes of professionals in the software industry who routinely dismiss diagram techniques as being useless, or worse, a tell-tale sign of a weak mind (as Dijkstra did), without feeling the slightest need to substantiate such sentiment with evidence of any kind. At the same time, none of the proponents of diagram techniques I have seen (speaking or in writing) make any serious useability arguments in favour. Clearly it's easy to draw up small examples on which a particular diagram technique does well, and other examples to discredit the same technique. But that is the full extent to which the matter seems to be dealt with, even among professional software design specialists, such as the designers of the UML.

So what I have been reading, mostly between the lines, is that formulas are "too hard" while diagrams are "too easy". Well, on the whole there may be a grain of truth in this thought, but I'd like to see more details. Are there any serious studies on the useability for diagrams (vs. that of tables, or formulas, or other types of visualizations) for conveying information? Or is this whole subject really as trivial as everybody appears to believe?