Monday, October 6, 2008

Well this one goes to eleven - NOW NEW AND IMPROVED!

Seeing as I'm still deep in the paper production time, I've been thinking a lot about whether there is an optimal number of figures, and if so, what that number might be. Then Dr. Jekyll + Mrs. Hyde got the ball rolling in wondering why seven figures seems to be the norm. Also, we've probably all heard the anecdote that as the number of figures in a paper increases, the number of people who read it decrease.

I think that the number of figures per se is not terribly meaningful. Unless there are journal dependent requirements (think GlamourMagz), then a paper should have as many figures as are needed to tell the story. If that's 5 figures, good. If that's 15 figures, so be it.

But what's important to me is that each figure contains a coherent point*, conveys that point as succinctly as possible, while still reflecting the reality of the result.

What really gets all up in my grill is when people cram panels into a set number of figures, while blending the points together. That's one way to fit into an external figure limit, while increasing the amount of data presented. I think it fails because it dilutes the take home message from each figure. Also, it destroys any visuo-cognitive impact a figure has. When I look at a figure, I want to be able to see that message almost immediately, and preferably without referring to the legend. When I have to go through each of A-J panels, that ain't gonna happen.

So as a little test, I quickly went through 15 pdfs that were in my "PapersToSummarize" folder, and looked at the total number of figures, as well as the total number of panels in each figure. What struck me was that regardless of the total number of figures, the total number of panels was actually not that different. Now, this is a completely non-random sample, and reflects mostly electrophysiology related papers, with only a smattering of GlamourMagz, but let's look:

Table 1:

Ok, so the median number of figures was 7, perhaps reflecting DrJ+MrsH's impression that 7 is heaven. But also, it's pretty clear that regardless of total number of figures, right about 29-35 panels (overall avg was 32 panels). So those people who don't read a paper with too many figures (if they even exist) are deluding themselves in thinking that really means anything. See, here's a graph to prove it (and they say there's no science on science blogs. There's goddamn regression line in there. You call that no science?):

See, there ya go. Papers with more figures have more panels, but the slope is only ~2 panels/figure. Also, extrapolating to zero figures gives us the predicted absolute minimum panels for a zero figure paper, ~20. Anyway though, the correlation is crap here, with R-squared 0.21.

After doing this, I decided to check back on my papers from my thesis lab, since as those were two author papers, I did the lion's share in planning the figures and their layouts (I'll include the albatross around my neck, one manuscript that will, will see the light of day, even if I have to drown a couple hundred crewmen to publish it; it'll make for a kick ass song though):

11 figures, 27 panels, for 2.5 panel/fig.
11 figs, 27 panels, 2.5
10 figs, 29 panels, 2.9
And the number of panels ranged from 1-5. Pretty consistent.

So readers, tell me: What is your ideal number of figures?

*by point here i mean someting like a quantum of scientific result. Yeah, I just made it up, but maybe other people intuitively understand it.


River Tam said...

Well, I don't know about ideal, but a sample taken from my publications indicates my average number of figs is 7.44 and average number of panels is 10.3

I was actually surprised by this. There's been a big push in my field for shorter articles, so I thought my average would be far less than 7. Perhaps if I focused on taking a more recent subset my number would be smaller....

Nat Blair said...

10.3 panels per paper, with 7.4 figures. How beautiful they must be, with 1 or 2, maybe 3 panels at most!

But people want things shorter? How short do they want them? Is the text relatively longer? I would say that in my very superficial reading of anything at all related to ecology (ok, mostly the 2 papers I was on as an undergrad) the text relative to figures was higher than what I see as typical in a physiology paper.

Is supplementary data a big thing in your journals? I personally don't see that as a truly desirable route to shorter papers, but it's one way.

I will say that in some cases the panel is a very ambiguous thing. In reviewing some of those papers, what the authors chose to call a panel (which I stuck with for the quantification - and boy, did that table come out sucky) is not what I would have called it. In general, they were cramming too much stuff into a panel IMO.

River Tam said...

Oh, yes, I believe in elegant, uncluttered figures! Otherwise, what's the point? (though I have to admit, my distribution of panels per figure is not.... Gaussian)

Yes, the move has been in general for papers overall to get shorter, which means fewer figures, fewer words, fewer citations...really fewer everything. It has resulted in either more focused or less deep papers (depending on your point of view). Supplementary materials are huge (both in popularity and in size). I have mixed feelings about this as well. On the one hand, a good supp is useful for recreating a complicated methodology in an otherwise short paper, but sometimes we end up needing to stick everything but the kitchen sink into the supp to fit into the word requirements and some important things get shoved into a file may people probably won't look at. I also feel that they are reviewed less rigorously than the main text...

Don't know whether ecology papers tend to have a higher text:figure ratio (my reading in the physio lit is even less than your reading of the ecology lit). However, most of our work is probably more focused on a single set of experiments because of the logistic contraints, which may reduce the number of figures that need to be presented? Just a guess.

Dr. Jekyll and Mrs. Hyde said...

Dude, that's hysterical that you quantified. Trust a scientist! I agree that I want each figure to have a straightforward title that sums up the contents--which only works if each figure is a unitary idea, more or less.

And since I have it handy, my most recent paper-to-be: 7 figures (as I posted about), 37 total panels (although 2 or 3 of those are "schematic" non-data panels.)

Love the albatross. We all has them. That'd be a good post topic...

BTW your table is all but unreadable on my platform (Mac, Firefox). The html crap for tables is annoying but workable; here's one site with info.

Nat Blair said...

Thanks for the table info pointer, DrJMrsH. I decided I'd update the post, adding a table as an image, and a figure I made in about 10 minutes this morning.

I think these revisions have like totally answered the reviewers' questions.

Dr. Jekyll and Mrs. Hyde said...

I am especially impressed that the author has made use of extensive analytical techniques, including regression lines and Igor tables, to demonstrate the point. This blog post is likely to spur other groups to further quantification of their own figure/panel ratios. Accept as is!

Nat Blair said...

sweet! *goes to update CV*

I'm gonna bring this comment over to Mike Greenberg and let him know. I think a Hahvahd job is in my future!

Keep your fingers crossed everyon.

Dr. Jekyll and Mrs. Hyde said...

Funny, here I thought Harvard was known for eating its young, not promoting them :)

Nat Blair said...

Hey, Mike, nice new chairman digs ya got here! Check out this blog post I just wrote, it's got a table, a figure, and like 10 people on teh internet just read it so maybe....WAAAAAAA

*disappears into the maw*

whoever said...

So, you really think that David gives a shit about any of this - as long as it gets published in something with a high impact factor?

Nat Blair said...

@whoever - Last I checked this was a blog by Nat Blair, not David Clapham.

So it's not so relevant what David thinks of all this shit. If you want to know, why don't you email him?

Dr. A said...

The impact factor of this blog just went up. I love this post. My average number of figures is 8, not including supplementary figs.

I used to hate supplementary data but now am starting to appreciate it.. another idea for a post.

Nat Blair said...

Thanks Dr. A! I'd definitely be interested on your take regarding suppl. figs. I tend to think they're annoying, but if you've reconsidered that, I'd love to hear why.

Dr. A said...

I always found them annoying as a reader.. especially when you print the article you are going to present in journal club tomorrow so that you can read it on the bus on the way home and then you see that there is a shitload of suppl. figs that you did NOT print and have to go back.

My last paper from my PhD work, however, included 3 supplementary figures. This allowed me to show important data that was fundamental to the conclusions I was making but would have made the paper too long. One of the figs was actually something a reviewer asked for.

The greatest use of suppl. figs is in those places where it would normally say (data not shown) but as a reader you really want to see that data!!

NeuroStudent said...

funny, I always took (data not shown) to mean n=1...