Bad and Good Presentation Graphs

Posted by Chad Orzel on June 29, 2010
(10)
More »

I gave a short introduction to how to give a presentation today to the students who will be presenting their research in our twice-weekly Summer Student Seminar Series. This included examples of a data slide that is bad in the ways that students’ first attempts at data slides tend to be bad, and the same graph re-done in a more appropriate manner. As long as I’m doing format conversions of this anyway, I figure it might be amusing to post them here. So, here’s the bad graph, with the bullet points highlighting the mistakes:

i-ce1358dbaf38baa4a3badfa15d0695df-BadData.png

And here’s the good version:

i-f45f7ca0def863213c9be4121ba20c32-GoodData.png

(The seminar series features three talks per session, at twenty minutes per talk (15 minutes plus 5 minutes for questions), usually with a short break between speakers when students can re-load on food and drink.)

Edward Tufte, I’m not, but I think these help to make the important points. Of course, I’m probably forgetting lots of things that really ought to be mentioned regarding presentation slides. Feel free to remind me of them in comments.

(10)
More »

Comments

  1. #1 Rich
    June 29, 2010

    A pet peeve that I have with Excel-generated graphs (not that it’s the only way, but certainly a common tool) is that the default seems to make the content area as tiny as possible while making huge margins and legends. I wish that there were some sort of checkbox for “maximize the actual graph” so that I didn’t have to resize it manually every time!

  2. #2 Chad Orzel
    June 29, 2010

    Yeah, it was really easy to make that “bad” graph– the only Excel default I needed to change was making the color uglier.

  3. #3 Tim Eisele
    June 29, 2010

    Maybe you could add one more bullet point to the first slide:

    “The Excel defaults are all awful. Do not use them.”

    Also, I personally like to use Arial or Helvetica fonts (20 point or larger), because the varying line width in, say, Times New Roman makes parts of the letters indistinct at the resolution of the average screen.

    A rule of thumb I like to use is that if you make the slide the size of an 8.5″x11″ sheet of paper on your computer screen, and then sit back *At Least Ten Feet* from the screen, then that is about what it is going to look like when it is projected in front of the audience. People tend to dramatically overestimate how big a screen is going to look to people in the middle or back of the room.

  4. #4 Eric Lund
    June 29, 2010

    One important do for any kind of presentation (and this applies to posters and manuscripts as well as talks) is to have some kind of drawing/artboard program for cleaning up any disasters that you do get out of your graphing program* (or handed to you by your artistically clueless collaborators). That’s the main purpose for which I use Adobe Illustrator, although there are other options as well.

    And yes, you need to make the label fonts bigger. You might get away with 18 point for numbers on the axes, but everything else should be 24 point or larger (subscripts can be downscaled a bit, but don’t go smaller than 14 and 18, respectively). Use a sans-serif font if you can;** Helvetica works best with numbers and Greek text (if your machine and software support Unicode), while Gill Sans does a good job with Western European text.

    *Excel’s defaults are particularly egregious, but I run into trouble with the allegedly publication-grade IDL PostScript output as well. Things like lines too thin, labels colliding, using vector fonts instead of hardware fonts, etc.

    **For the love of your chosen deity avoid Comic Sans (that font has legitimate uses, but a presentation slide is not one of them). Also, I find Arial to be much uglier than the otherwise similar Helvetica, but YMMV.

  5. #5 R Simmon
    June 29, 2010

    You could also use gray axes instead of black (to de-emphasize them relative to the data), remove the left-hand axis entirely, and use ticks instead of lines. Stay away from pure RGB colors (i.e. red = 255,0,0; blue= 0,0,255 (the default Excel colors in the current version aren’t bad)), and make sure your labels are anti-aliasing correctly. You can also adjust the font weight & size to establish a clear hierarchy, and unless it’s required, I don’t think “Union College” needs to be on any slide except the first one. If you do choose to go with sans-serif fonts, stick with one unless you really know what you’re doing. I’d use either points or a line (probably points), connecting the dots just adds clutter, and just use a plain circle rather than the diamonds. Finally, tone down the arrows (but not the “food breaks” label), they’re one of the least important pieces of information.

  6. #6 Cosma Shalizi
    June 29, 2010

    Remove the horizontal lines going across the plot area, or, if you need them for some actual part of the presentation, grey they out so they’re not as visually prominent as the data. Similarly, if the curve is just intended as a guide to the eye, make it less visually prominent than the data. If the curve is a theoretical prediction (from a lattice seminar phenomenology model, no doubt), keep it as prominent as the data, but perhaps distinguish it by color.

  7. #7 Chad Orzel
    June 29, 2010

    As several of these suggestions are things that I quite deliberately did not do, let me explain my reasoning:

    - Tufte makes a lovely sounding argument in favor of removing plot axes, but I find that it’s a great technique for rendering graphs completely unintelligible. On the rare occasions when I encounter such a graphic and really need to understand it, I usually end up drawing my own axes.

    - I’m usually against connecting the dots unless it’s a fit, but for this sort of data, I think the line helps emphasize the dramatic up-tick at 20 and 40 minutes. This fake dataset is clean enough and the points numerous enough that it would probably be ok without the line, but with noisier datasets, the connect-the-dots line can be a big help. If this were for a journal, I would probably leave the line out, but for a talk, I think it helps.

    If it were a fit, rather than a guide-the-eye line, I would’ve made it a different color.

    - I find the horizontal lines helpful for making quantitative estimates on the fly, particularly in seminar talks. If I were putting this graph in a journal, I would leave them out, on the grounds that anyone who needs to make quantitative estimates can draw their own lines, or hold a ruler up to the monitor, or something. For seminar talks, I tend to put them in.

    I agree that the lines could be lighter; I think they are in the original figure, but the process of format conversion and scaling made them appear more prominent (I don’t have the original here to check).

    - The fonts are just my default choices, out of pure laziness. Provided the text is big enough, I don’t find that fonts make a great deal of difference, and I tend to prefer serif fonts. You probably need slightly bigger text with serif fonts, but it’s not like I’m charged by the pixel. The aliasing you see here is another artifact of format conversion and scaling– at full size on the screen, it looks fine, even from the back of the room.

  8. #8 Danil
    June 29, 2010

    In addition to the lines, I’ll normally gray the labels for the tic marks as well, and even the labels for the axes; data to the foreground, chart junk to the back.

    Also, that underline of the header turns out to be unfortunate given the spacing of the horizontal lines in the graph.

  9. #9 CCPhysicist
    June 30, 2010

    Great presentation. What the comments overlook is that you stuck to your own rules to KISS while conveying the all important 20 minute rule about any presentation (or any class you are teaching). More is not always more.

    Although gnuplot is better in many ways than Excel, it also suffers from flaws in its default settings. The rule should be “Make the program tell your story, not its own story”.

  10. #10 Zeynep
    June 30, 2010

    R.Simmon: Why stay away from pure RGB? Ignorant & curious person asking.