English | 日本語
LAB FORUM — WRITING PAPERS | FIGURES

Building the Claim Through Figures

Figures — Until the figures are decided, do not write the text
Before reading the text, experts look through the figures in order to grasp the whole study. A good paper is one where the figures alone carry "what was asked, how it was tested, and what was found."

What this lecture covers

The Lab Forum lecture "Writing Papers" covered the whole picture of submitting a paper (the mindset for writing, authorship, cover letters, responding to reviews). This page covers the stage before that: designing, testing, and freezing the figures. It corresponds to Figures-A → B → C of the Paper Card, and walks from the Figure 0 you drew in Lab OS Chapter 04 (Framing-B) up to the moment just before Draft.

The principles of visual design (fonts, color, alignment, diagramming techniques) are systematized in the Lab Forum lecture "Design for Research". Here we focus on what is specific to paper figures — how a figure carries a claim — and refer to design principles where needed.

1
Act 1 Figures are claims
From "for myself" to "for the reader"
2
Act 2 Figures-A Design the empty figures
Decide the figures before the data
3
Act 3 Figures-B Test provisionally
Find where it breaks with minimal data
4
Act 4 Figures-C Freeze
Freeze and reduce
5
Act 5 Figure craft
Three levels for not wasting the reader's brain
6
Act 6 Daily habit
The panel collection

B4: Learn that a figure is a "claim," not a "place to put data." M & D: Check the correspondence between the Figure 0 and the main figures on your own Paper Card. Postdocs: Develop the eye to look at a junior's figure and ask "can the claim be read?"

Act 1
Figures are claims
Figures Are Claims — From "for myself" to "for the reader"

A figure "for yourself" does not exist

Records are left for a future other

The first mistake many students make is to create figures "to organize my own data." In science, there is no such thing as information summarized for yourself. Records, including lab notebooks, are left for a future other (including yourself six months from now). Figures are always made for someone else.

Pave the road so that a third party can evaluate the value of the research without stress. That is the job of a figure. A messy figure raises doubts about the integrity of the research itself. Do not make the reader spend brain resources on decoding — this is the essence of technical writing.

A figure is not a memo for yourself; it is road-paving for the reader.

Hack the habit of "reading from the figures"

Aim for a state where the story is complete in the figures alone

Before reading the text closely, experts try to grasp the whole study by looking through the figures in order. An excellent paper is a visual narrative in which "what was asked, how it was tested, and what was found" is complete just by following the figures. Abandon the indulgence of treating figures as supplements to the text, and pursue the completeness where the story stands on the figures alone.

That is why figures are decided before the text. As stated in Lab OS Chapter 04, the moment the Key Contributions are decided, you should be able to say "if this figure comes out, the claim stands." If you cannot say it, the claim has not yet been designed.

Decide the figures before the text. If you cannot say it, the claim has not been designed yet.

Correspondence between figures and claims: one figure, one claim

One box in Figure 0 (the structure diagram of the claim) corresponds roughly to one main figure. If there are three Contributions, there are about three main figures. If you cram claim A and claim B into one figure, the reader no longer knows what to take from it. Claim B goes to the next figure.

When the set of figures works as a sequence of "mini-papers," the story of the whole paper crystallizes.

One figure, one claim. The set of figures becomes the story of the paper itself.
Act 2
Decide the figures before the data
Figures-A — Designing the empty figures

The purpose of Figures-A

Start from empty figures showing only axes, comparisons, and elements

The purpose of Figures-A is to decide, for each figure, "what claim this figure makes." No data is needed yet. You create empty figures that show only the axes, the comparison, and the elements.

Figure list (with roles): First, list the figures that will appear in the paper, each with its role (question / setup / validity / claim) and its location. Next, the Contribution–Figure Mapping: map only the claim figures to Contributions, as in C1 → Fig.4, C2 → Fig.5. If a Contribution has no corresponding figure, it is a claim not supported by a figure. If a claim figure has no corresponding Contribution, it is a figure without a claim. The roles are explained in detail in the next section.

What to write for each figure

A figure whose Claim cannot be written has not yet been designed.

The four roles of figures: one figure, one job

"One figure, one claim" is about the main figures. The figures in a paper are not only those. Some students prepare just three figures to match the Contributions and stop, but that confuses "figures that carry claims" with "figures that appear in the paper." Figures have four roles, and each role has a different design question.

RoleLocationJob of the figureDesign question
Question figureIntroduction or start of Methods (follow the journal's convention)Show the Gap and the Core Question. Overview of the target, conceptual modelWhat "question" can the reader not understand without this figure?
Setup figureMethodsShow the target, apparatus, model domain, and workflow at a reproducible granularityWhat procedure cannot be reproduced without this figure?
Validity figureMethods or start of Results, AppendixShow that the method is trustworthy. Benchmarks, convergence, calibrationWhat is shown to be trustworthy (a claim about the method)?
Claim figureResultsCarry a Contribution. One-to-one with the boxes of Figure 0What is the Claim (one sentence)?

Only the claim figures correspond to the boxes of Figure 0. For an Original paper, a rough guide is 2–3 claim figures and 2–3 question/setup/validity figures, 5–8 figures in total.

Whether a figure may be placed in the Introduction varies by field and journal. In the earth sciences it is common to place a map of the study area or a conceptual model as Fig.1 at the end of the Introduction or the start of Methods, while in engineering and Nature-family journals more papers place a conceptual figure in the Introduction that draws the problem setting and the overview of the method in one panel. There is one common rule: a figure in the Introduction is a conceptual figure and does not show results. When you collect papers from candidate journals in Step 1 of Framing-A, note what Fig.1 draws and in which section it appears, and the placement is settled.

Validity figures actually carry a claim: a claim about the method, "this method is trustworthy." The anticipated reviewer comments "Assumptions are unrealistic" and "Comparison is insufficient" come when the setup figures and validity figures are lacking.

Only claim figures correspond to the boxes of Figure 0. Aim for 5–8 figures in total.

The form of the claim and the form of the figure

The distinctions in Lab OS Chapter 02 and Chapter 04 determine the form of the figure.

Form of the claimFigure needed as evidence
Hypothesis-first: the law holds within the range of …Agreement of prediction vs. observation. The boundary of the range of validity (where it fails)
Data-first: the data show …The distribution of the data and its lack of bias. Reproducibility of the regularity found
Active inquiry: … can be used for designThe effect when used for design (comparison of time, cost, accuracy)
Committed to explanation: … governs …Comparisons that eliminate alternative explanations (conditions with confounders separated)
Committed to prediction: … holds even under unknown conditionsValidation on data not used for training

Look at the form of your claim and choose the form of the figure. If you try to show a regularity found data-first with a hypothesis-first figure (prediction vs. observation), something will always break.

When the form of the claim and the form of the figure diverge, something always breaks.

Write panel titles as "findings"

Toward a state where the result can be foreseen the instant the title is read

Write panel titles not as "Analysis of X" but as the finding obtained, such as "X increases Y." The reader being able to foresee the result the moment they see the title maximizes the speed of understanding. At the empty-figure stage, write provisionally "X increases Y (expected)" and confirm in Figures-B.

This is the same principle as "the key point in 25 characters at the very top" in Design for Research. The key point of a slide becomes the panel title in a paper.

Panel titles may be provisional. Confirm them in Figures-B.

Next Judge

What to confirm at Team Meeting: does each Contribution have a corresponding figure / can the Claim of each figure be said in one sentence / does the form of the figure match the form of the claim. If it passes, on to Figures-B.

Act 3
Find where it breaks with minimal data
Figures-B — Provisional testing

The purpose of Figures-B

The purpose of Figures-B is to confirm with minimal data "whether the figure can support the claim." It is not about collecting all the production data first and then making figures. Fill the empty figures with a small amount of data, rough calculations, or reuse of existing data, and look for where the claim breaks.

This is the stage where the falsifiability of Lab OS Chapter 02 is carried out at the level of figures. Not "going to prove it" but "going to see where it breaks." If it does not break, the claim stands; if it breaks, revise or discard the claim. Either way is progress.

What to confirm

If it does not break, the claim stands; if it breaks, that is progress. Either way, progress.

Stop the multiplication of figures

While testing provisionally, an interesting result appears and you think "I want to include this figure too." This is the multiplication of figures. Added figures often do not correspond to a Contribution and blur the claim of the paper. Leave the interesting result in the Research Log; if there is no box for it in this paper's Figure 0, do not include it. It becomes the Idea-0 of the next paper.

An interesting result is the seed of the next paper. Do not put it in this one.

Next Judge

What to decide at Team Meeting: for each figure, "keep the claim / revise / discard." If revising, go back to Framing-B and rewrite the Contribution. Going back here is normal. If you proceed to Draft with figures that do not support the claim, you will end up patching things together in the Discussion, and it will collapse in review.

Act 4
Freeze and reduce
Figures-C — Freeze & Reduce

Freeze & Reduce

The purpose of Figures-C is to freeze the figures at minimum number, maximum claim power.

Write three things on the Paper Card: Frozen Figures (number, claim, and data source of each finalized figure), Figures to Remove (figures to cut and why), and Minimum Required Additional Work (what remains).

Freeze at minimum number, maximum claim power.

From here on, do not add figures

The meaning of freezing is to concentrate all subsequent work on "turning the figures into text." If, while writing the Draft, you think "actually, this figure too," that is a decision to go back to Figures-B, not a decision to add a figure within the Draft. When a figure moves, the Results and the Discussion move too.

When a figure moves, the Results and the Discussion move too.

Next Judge

Do the frozen main figures correspond one-to-one with the Contributions? If it passes, on to Draft (Lab Forum lecture "Writing Papers — Draft").

Act 5
Do not waste the reader's brain
Figure craft — Three levels

Figure craft: three levels

Once the design of the figures is decided, raise the quality of the figures themselves. Professional figure-making has three stages. Level 1 is the rules that must absolutely be kept, Level 2 is the design of the flow of information, and Level 3 is presentation.

Level 1|Always keep

1
Strict order of appearance (top priority)
Figures and panels are arranged in the order they are explained in the text. A structure that explains 2c after 2a and skips 2b is not allowed. Even in the appendix, there must be no figure that is never referred to in the text
2
Pixel-level alignment
If the edge of a panel, the start position of text, or the height of an axis label is off by one pixel, the reader unconsciously feels something is wrong and loses trust in the author
3
Unified fonts and sizes
Sans-serif such as Helvetica or Arial for Latin text, gothic for Japanese. Unify sizes within a figure and across the whole paper
4
Complete within the figure (self-explanatory)
If you push "black is data A, white is data B" into the caption, the eye goes back and forth between figure and text. Embed the legend inside the figure
5
Statistical transparency and no Excel defaults
A bar chart of means alone hides the distribution of the data. Show individual points with violin plots or jitter plots. Make matplotlib, ggplot2, or Prism the standard
Excel's default settings (shadows, gridlines, color scheme) are a symbol of "dishonesty."

Level 2|Design the flow of information

1
One topic, one figure
As stated in Act 1
2
Guiding the eye
The reader's eye moves in a Z from left to right, top to bottom. Arrange panels a, b, c along this flow
3
Finding-oriented subtitles
As stated in Act 2
4
Color universal design
Avoid red–green, do not distinguish by color alone, combine with line styles. Details in Part 3 of Design for Research

Level 3|Presentation

1
Scientific validity and playfulness together
Lumping missing data and failed computations together as "N.D." is dishonest. Prepare your own icons that distinguish out-of-memory from time-out
2
Imitate and analyze top journals
Learn the patterns of layout, fonts, and typography from Nature, and the abstraction and symbolization of conceptual figures reconstructed by professional artists from Nature Reviews
Excellent visualization begins with imitation.
Act 6
The panel collection
Daily habit — 15–30 minutes

Make a panel every day, 15–30 minutes

The skill of making figures is not only for writing time. When experimental data come out, shape them that same day into a publication-quality panel and record it in the Research Log. Every day, 15–30 minutes.

Repeated daily, figure-making becomes templated, and you can finish it in the little time before going home. When you write the paper, finished panels are already at hand, and writing ends with just thinking about the composition. The "Draft is an unfolding task" stated in Writing Papers presupposes this habit.

The provisional testing of Figures-B can also start from the daily panels. Once the empty figures are designed, you can see which of the panels at hand can be used. Not making figures from zero, but choosing from the collection.

A 15–30 minute habit today, so that you do not panic at writing time.

Final checklist

1
Does the figure list carry roles (question / setup / validity / claim)? Are the claim figures one-to-one with the Contributions, with a one-sentence Claim written for each (Figures-A)?
2
Does the form of the figure match the form of the claim (hypothesis-first / data-first, commitment)?
3
Did you look for where it breaks with minimal data? If it broke, did you go back to Framing-B (Figures-B)?
4
Are the figures frozen, with none added since (Figures-C)?
5
Are the panels arranged in the order they appear in the text?
6
Is the font Helvetica/Arial, with sizes unified across the whole paper?
7
Are panel edges, labels, and legends aligned at pixel level?
8
Have you eliminated Excel defaults and bar charts without distributions?
9
Do the panel titles state findings (results)?
10
Does the eye flow in a Z?
11
Did you turn today's data into a publication-quality panel within 30 minutes?

Takeaways by role

B4
Write the Claim in one sentence
Pick one figure from your thesis and try writing its Claim in one sentence. If you cannot, that figure has no claim
M & D
Make the mapping table
Make a table mapping the Figure 0 on your Paper Card to the main figures. Find the boxes and figures that do not correspond
Postdocs
Look at three points
Look at a junior's figure on just three points: order of appearance, alignment, finding-titles. That alone fixes half of it

Related links