Steve Hsu has a post comparing his hand-drawn diagrams to computer-generated ones that a journal asked for instead:

i-9e21cbe4b23ff54421a15e4ccf2b6ad3-hsu_figure_comparison.jpg

He’s got a pretty decent case that the hand-drawn versions are better. Though a bit more work with the graphics software could make the computer ones better.

This reminded me, though, of something I’ve always found interesting about scientific publishing, namely the evolution in the use of figures through the years. Whenever I need to do literature searching, I always suspect you could guess the approximate date of a paper’s publication by looking at the figures.

If you go back far enough, reproducing figures was a very difficult process, so there tend to be relatively few of them. What figures you do get, though, are exquisite:

i-38e7e6f370802ffb9ffe4b2fc79d8f78-072mich.jpg

This is from the original Michelson-Morley paper, showing their apparatus. It doesn’t just give you a sense of the layout of their experiment, it’s also a pretty nice drawing. The various optics mounts are drawn with a fair bit of detail, and you even get nice little touches like the individual bricks of the support for their rotating granite optical table.

This was almost certainly the work of a professional draftsman (though I suppose it’s conceivable Michelson or Morley originally drew it, and just had it copied over). Getting a drawing into print was a non-trivial matter back in 1887, and if you were going to do it at all, you would have it done right.

This was the general state of affairs up until around the 1980’s: figures showing bits of apparatus, when they appeared, were very well-done, because they were generally handed off to professional draftsmen to make, so you get realistic perspective drawings of key components, and so on. The style changes a little to reflect the general aesthetic of the time– the Michelson-Morley figure looks like it’s from the late 1800’s– but the majority of figures in papers, particularly the ones significant enough to still be cited today, are professionally done.

Sometime in the early 1990’s, though, computers advanced to the point where any reasonably competent scientist could make his or her own figures on a desktop computer. At which point, you see a rapid shift to electronic submission of the figures (within my graduate career (1993-1999), we moved from sending in full-page printouts of our figures, to be scanned and reproduced at the journal, to emailing them .eps files containing the final figures) by the authors, without assistance from anyone else.

And, in that era, you have a dramatic decrease in the artistic quality of the figures. In fact, you get a lot of diagrams that look like Steve’s computer figure above: everything is represented by a featureless rectangle with a label on it. Or possibly a labeled oval or rounded rectangle, but always basic shape from a vector drawing program, because that’s what nearly everyone was using to make the figures.

This is pretty much the standard aesthetic I internalized as I became a professional scientist, and it’s what I default to even today. Here’s my rendering of the Michelson-Morley apparatus, from How to Teach Relativity to Your Dog:

i-2652aad7c08437aec814e8cfeccd3266-relativity_michelson_bw.jpg

That’s done in PowerPoint, because I’m familiar with its basic vector drawing tools, and that’s all I really need. I probably ought to learn to use something else, but if all I’m drawing is labeled basic shapes, why bother?

Of course, computers have continued to get more powerful, so we’ve entered yet another era. The figures are still done by individual scientists on their own machines, but it’s now relatively easy to do fancy graphics, so we’re back to having 3-d perspective renderings of everything, such as this figure from a recent paper about quantum optics:

i-4a7de17bd13714b10b539b174d2fda63-HBT_apparatus_fig.jpg

The final version of this is in Nature Physics, and if you expect to get in a glamour journal like that, you can’t really get away with labeled boxes any more. Now everything’s in some desktop CAD program, with spiffy three-dimensional effects and color shading and all the rest.

There’s a bit of an uncanny valley effect to a lot of these, but over time, I bet that will go away, too. Twenty years from now, when scientific articles are beamed directly into our neural implants as open-access holograms, everything will probably look pristine and photorealistic. And the whole notion of schematic diagrams pieced together with basic vector-graphic shapes will seem as quaint and primitive as hand-drawn overhead transparencies do today.

Comments

  1. #1 Thad
    March 30, 2012

    I pretend that I have an artistic side, but in reality it only flares up now and again, and is not at all reliable.

    Nevertheless, I have on occasion in the past, when someone else took the lead in the paper authorship, been horrified at the powerpoint-drawn figures submitted. I didn’t have the time to do them right, and couldn’t bring myself to ask them to do them over, or ask them to get someone else to do it right.

    My advice: gnuplot and inkscape work very well together. Learn them both. Never submit a 2D plot as a jpeg image. There is no excuse for not submitting postscript generated from and svg formatted drawing!

  2. #2 Eric Lund
    March 30, 2012

    Color is another respect in which figures have changed. With the standard offset printing techniques that were in use for most of the 20th century, color figures were horrifically expensive. You may have seen occasional references to “plates” instead of figures; the distinction was because of the different handling color figures required. So color was used only when absolutely necessary. By the mid-1990s, computers had advanced enough that it became possible to include color more freely in figures, and the plate/figure distinction became arbitrary. Today it would be unthinkable to publish any but the simplest of figures in black and white.

    Don’t get me started about the many papers I have seen, most if not all from the 1970s and 1980s, in which author-produced camera ready copy was obviously created on a typewriter. Two programs put an end to that: LaTeX and Microsoft Word. Today almost all authors use one of those two packages to create manuscripts. Thus there was much wailing and gnashing of teeth a few years back when Microsoft rolled out the .docx format, which many of the major publishers couldn’t deal with for at least a year or two. (Being on the LaTeX side of that holy war, I don’t know whether journals can accept .docx files today.)

    Yes, you probably should learn a better drawing program than PowerPoint. Do you never have to annotate your data plots after the fact? I don’t know anybody in my field who draws figures for publication with PowerPoint (presentations are a different matter). I personally use Adobe Illustrator, which does a fine job cleaning up the messes that IDL’s Postscript generator creates (my field is heavily locked into IDL for plotting data). There are probably other products that work at least as well.

  3. #3 Chad Orzel
    March 30, 2012

    I use PowerPoint to generate things like schematic diagrams of the apparatus used for some experiment. For displaying actual data, I use SigmaPlot to generate the graphs, and it has all the annotation capability I need, plus it generates very nice .eps output for inclusion in LaTeX documents. I’ve also used MatLab for image-processing purposes, and if I needed to do more quantitative analysis of pictures, that’s what I’d go with.

    Illustrator is (or at least was, the one time I looked) too expensive to justify buying for making silly little block diagrams, which would be the only thing I’d use it for.

  4. #4 Michael Richmond
    March 30, 2012

    GNUPLot will draw simple figures much more cleanly than Powerpoint, and export them in many formats.

  5. #5 Rosie Redfield
    March 30, 2012

    I’ve become able to do very sophisticated drawings in PowerPoint; that’s how I make the final versions of all the figures for my talks and manuscripts. I like it because it gives me much more control over such visual features as fonts and placement. I often start with a graph or other figure from Excel, or that the postdoc has created in R, but I change the labels and legends to make everything as easy as possible to understand.

  6. #6 Michael Richmond
    March 30, 2012

    Xfig will draw simple figures much more cleanly than Powerpoint, and export them in many formats.

  7. #7 Jacques Distler
    March 30, 2012

    arXiv:1106.5410 has 268 colour figures (and, yes, the colours are significant, as are the embedded TeX equations).

    I used to use Illustrator. Now I use only open-source solutions.

    Works much better …

  8. #8 onymous
    March 30, 2012

    Eric Lund wrote:

    Today it would be unthinkable to publish any but the simplest of figures in black and white.

    Unfortunately, there are still journals that charge authors for printing color figures. So I have papers with color figures in the arxiv and online-journal version, but they’re black & white in the printed journal. But who uses paper journals now anyway?

  9. #9 Joemac53
    March 31, 2012

    I would give my eye teeth (if I could find them) for software that would make it easy to make a nice unit circle with labels. You can be sure I’m watching all these comments for ideas.

  10. #10 Zack
    March 31, 2012

    I try very hard to make all my figures black and white, because I’ve seen far too many xeroxed course readers in which the pretty color figures have been reduced to a uniform gray rectangle.

    Sometimes, though, it’s just unavoidable.

  11. #11 Anna
    April 1, 2012

    If you go back a century or two, there’s more interesting stuff to be found in the images. In the early 1800s images weren’t “reporting” but “instructing” so people drew what they thought they ought to see, rather than what they saw. Later in that century the fashion flipped around and the pictures were more or less “raw”, with individual lens blemishes and all.

    It’s a really enjoyable area of science history.

  12. #12 CCPhysicist
    April 1, 2012

    Your comment also applies to textbooks. Over the past two centuries there was a gradual transition from beautiful engravings of actual devices to photos of actual products to a graphic artist’s impression of something they have never used. I am always pleased when I see a textbook that has actual drawings by the author, drawings that look like what a student should draw.

    That figure of the Michelson-Morley apparatus was almost certainly produced by an artist unless one of the authors was a good draftsman with india ink (and that assumes the journal had shifted from engravings to photoengraving). Even photoengraving by journals needed a quality original (original india ink drawing, 8×10 glossy made from same, or a negative) with rules on line thicknesses and contrast well into the 1980s.

    Is there any info from the AIP on when they switched from engravings to photoengraving (which was still new in 1887), or when they moved away from actual typesetting?

The site is currently under maintenance and will be back shortly. New comments have been disabled during this time, please check back soon.