On Writing: Part V… Step 1, .. It’s all about the figures.

A while ago I started a series on writing that I’ve been meaning to come back to. I’d been on sort of an extended hiatus from reviewing papers, just because I had a bunch of grant deadlines in the spring- but the few papers I’ve reviewed recently have made me think about paper writing organization of data and figures, how to decide when you are ready to write, the content of the different sections, what order I write things in, and so on… and so on. An endless topic for blog posts.

Anyway. The process of writing a paper begins long before you actually sit down to type the words into the document itself, at least for me. This process should begin in your head while you are collecting the data, and carefully recording protocols, data, conclusions, and next steps in your notebook. It would be nice if it could start even before that- with planning of the project itself. But I think that in reality it doesn’t really work like that, and projects take unpredictable twists and turns, and sometimes you end up on interesting (and fruitful) tangents that are impossible to plan for.  So- when you are doing experiments and collecting the data- you should be thinking about figures that include all the relevant controls (and if you don’t know whether or not you got them all, try bringing that one up in lab meeting because I’m sure you’ll get some input).  And, you should be thinking about beautiful publication quality figures with all the right statistical tests, properly applied. It is just horrible to have a bunch of data on figures that are messy, ugly, and worst of all – incomplete. And reviewers hate messy looking incomplete data, and they can use one spot of weakness to take you down!!

A note about statistics, because it is always shocking to me the number of papers that I review that come to me without ANY statistics.  I mean error bars alone, do not statistically significant data make.  After you figure the standard error, you have to apply a statistical test (you’ll have to figure out which is the appropriate statistical test for your particular situation), and then determine the probability that the results you got could have occurred by chance alone. I’m always amazed by how many authors forget this critical step. Or they say they do the statistical test, the put asterisks all over the figure, and they forget to tell you what statistical cutoff they used …P< ????, you say. It is ok if you are not a statistical wizard, and most of the time you don’t really need to be-… but if you should find yourself under-educated in this area- get the help of someone that knows what he/she is doing!

While you are doing the experiments- you should also be thinking ahead about what set of figures it will take to make a complete set of data. You are going to have to try to think of all the pieces, and of how the reviewer will think when they look at your data set.  This will probably take some practice- so if your mentor is a genius at this- try to pay attention and learn all that you can. Learning to do this will save you having to do actual experiments to answer reviewer criticisms, although you will probably still have to write long point-by-point rebuttals of reviews.

Once I get a set of data, I fastidiously make all the figures basically just as I would want them to appear in the paper. I lay them all out as I’m going along, and figure out where the holes in the story are- if there are any left at that point. Then we try to do those very last, clean experiments. I don’t know why- but that last experiment that you need, always seems to take nearly forever to get just exactly right … that last western always has more nonspecific background than it should, and just by Murphy’s law these take what feels like an eon to troubleshoot.

While I’m waiting for that last perfect figure, I just start writing…  and I always do the Materials and Methods first..and actually I can even start that earlier while I’m making the figures themselves… because it breaks up the monotony of using all the same font and formatting figures.

So, next up- Materials & Methods.

Advertisements

19 thoughts on “On Writing: Part V… Step 1, .. It’s all about the figures.

  1. I do it exactly the same way… In fact when discussing a project with trainees I ask them to tell me the figures that they envision will be in their manuscript. One difference though, I do materials and methods last, largely because it bores me to death. I’m always happy when the method involves something I’ve done before because then I can just say: yadda-yadda was done as described previously.

  2. JP- I write the materials and methods while doing figures, because it is very easy. I think for students sometimes the thought of writing a whole paper is kind of intimidating- I know it was for me at first. So for me, having an easy piece that I can bite off and accomplish gets the writing ball rolling for the more difficult sections.

  3. my Order:

    1) Figures
    2) Figure Legends
    3) Methods
    4) Results
    5) Intro
    6) Discussion

  4. No one puts together tables? 😛

    Generally, I do my M&M first. I then put together my Figures, Legends and Tables. Then R&D, Intro, and finally the Abstract. I almost never write a paper with separate Results and Discussion sections; I prefer to combine them.

  5. Most Journals actually decide that for you TJ.

    I do tables with Figures…didn’t mention it, as it seems these days I don’t put many tables in.

  6. I have to admit something embarrassing: I have no idea how to do a statistical test. I kind of get how to assess the meaningfulness of one… I mean, you need for other statistics to tell you that your 0.7 or 0.95 or whatever is significant for that dataset… but I don’t know how to do those statistics either.

    Remember, I came from synthetic chemistry training, where people just say “Hey, I made this stuff this time and it turned out this way. The end.”

    I have to try to do some of this for our paper we are working on, but I don’t even know where to begin. I hope some of my students know how to do it…

  7. Arlenna- Judging by the stuff that comes across my desk, you are not alone. But, the first step to learning is knowing what you don’t know. Statistical tests are necessary to determine whether two results (for example, the control and the experimental condition) are different by random chance alone, or not.

  8. Most Journals actually decide that for you TJ.

    Really? I’ve found it’s almost always negotiable. I have submitted articles with combined R&D to journals that recommend separate R & D sections … and had them accepted. No reviewers ever even suggested I split them.

  9. Arlenna: I’d recommend contacting the statistics department where you work (if it’s academia). Someone there surely can sit down with you, figure out what you’ve done, and what tests are applicable. In most cases, you should have your statistical methods in mind when planning your experiments, but if it’s too late for that, they can probably still help you out.

  10. In the experimental biosciences, the vast majority of experimental designs can be analyzed using either parametric group comparisons (t-tests or ANOVA with multiple comparisons) or Chi-squares.

  11. Arlenna, check out the “Graphpad prism statistics guide”. It’s free, and it’s one of the most intuitive intros to the use of stats in biology that I’ve come across.

  12. I’m right with you but the Figure Legends for me are always dead last, at the point when I am utterly sick of the paper. The quality of my figure legends undoubtedly reflects this situation.

    Perhaps I should change my ways.

  13. when you are doing experiments and collecting the data- you should be thinking about figures that include all the relevant controls (and if you don’t know whether or not you got them all, try bringing that one up in lab meeting because I’m sure you’ll get some input).

    In fact when discussing a project with trainees I ask them to tell me the figures that they envision will be in their manuscript.

    AAAIIEEE!!! I know lots of people work well this way, yet I’ve always worried that, depending on how it’s articulated, such a plan comes loaded with so many possibilities of unconscious bias. Of course, insistence on the right controls will minimize problems, but I often wonder what small observations might be unconsciously disregarded because they just don’t fit with the overall flow of the planned manuscript.

    I actually never work this way in the beginning. Neither have most of my mentors (I think I got lucky fitting in with my initial advisors that way). We always started with something like, “Hmm, this seems pretty cool, let’s play around, do some experiments, and see what happens.” Of course we always had some broad question/idea/framework that we were addressing, but the actual reasoning was typically pretty inchoate. Of course, this leaves a really difficult transition from the early pilot experiments to making the figure outline.

    This may be just a difference in degree, as I’m sure that no one is making a figure outline without any first result.

    So, here’s an interesting question: What fraction of experiments in you work were done before a figure outline was made?

    For me I’d guess somewhere in the range of 60-75% were done before an outline.

  14. You guys don’t just fold your minimum acceptable probability into your definition of a decection limit, and hide it in the methods section?

  15. Nat- I do the let’s fool around with this interesting avenue as well- and I don’t start making a figure outline at that point. How could I? But there comes a point in every set of experiments,where you have to think ahead about where you are going with a project, is it going to end up in a manuscript, or not? When you hit that point- you have to be outlining in your head what all you think should be in the paper. For controls- these should be done at every step. Period. It is a huge waste of time to do no controls, then have to come back later on and do them… then figure out your fantastic looking effect is an artifact at that late date!

  16. Nat said,
    “So, here’s an interesting question: What fraction of experiments in you work were done before a figure outline was made?”

    Usually 100% of the experiments I planned to do, and about 80-90% of the experiments I later realise I really ought to do. In the past I’ve tried to get ahead of the curve by preparing figures and figure outlines as I go along, but they end up getting rearranged so much during the actual manuscript prep that the attempt to save time usually ends up having been a waste of time.

    So, basically when I’ve completed the experiments in the initial design plan, I start preparing the figures… at which point either myself or the governor will have an unpleasant thought along the lines of, “A ha! We could make this story even more compelling if we did this, this and this…”

    And so it goes.

  17. Inn general, if you have an experiment that will require statistics, you should plan it with the statistical analysis in mind – this will guide your write up of the results nicely. (gets off soapbox) If you want to know more about stats and experiment design, ask a quantitative psychologist – it’s their stock in trade.

    I can recommend Jim Lindsey’s “revealing statistical principles” as a good intro.

  18. I’m in a slightly different biological field, but . . I go to our local statistician to find out whether the test I am proposing to do is correct, and he says, lets look at the data, and we look at hundreds of different graphs of the data, and I say, okay, so to get back to my question, should I do x? and he says, oh, you shouldn’t be testing that/ there’s no point, there doesn’t seem to be any difference/ these data are really not that interesting (!!!) whatever/ and I eventually leave at the end of my allotted time depressed and frustrated. . . (and as a footnote I should add I have published in several international journals some of these ‘uninteresting’ data.

    It seems to me statisticians are much more cautious about applying tests than scientists.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s