Presenting Results Clearly: Tables, Figures and Text

Most results sections would get better if the authors deleted a third of what's in them. I don't mean the findings. I mean the duplicates: the same means, the same p-values and the same sample sizes, printed once in a table, again in a bar chart and a third time in the prose.

Authors usually think repetition is thoroughness. Reviewers read it as padding, and editors read it as a sign that nobody decided what the paper is actually about. After twenty years of handling manuscripts, I can tell you that a results section where every number has exactly one home almost always belongs to a paper with a clear argument. The two go together.

So this post is about making that decision on purpose: what goes in the text, what goes in a table, what goes in a figure, and how to keep them from overlapping.

The rule I actually use: one home per number

Every number in your results should live in one place. Other places can point to it, interpret it or highlight one piece of it. They shouldn't reprint it.

This isn't just my preference. The ICMJE Recommendations, which most biomedical journals follow, say plainly that authors shouldn't repeat all the data from tables or illustrations in the text, and shouldn't duplicate data in graphs and tables. Plenty of authors have read that sentence. Few take it literally.

Take it literally. When you draft, ask of every number: where does this one live? If the answer is "everywhere," pick one location and cut it from the others.

Text, table or figure: how I decide

Most advice here is vague ("use figures for trends, tables for detail"). That's correct, but it doesn't settle the hard cases. What settles them for me is the question what does the reader need to do with this?

Text is for the handful of numbers that carry the paper

If you have fewer than about four numbers to report for a finding, put them in a sentence. A table with two rows looks like you ran out of ideas, and a figure with two bars wastes half a page to show something a reader could take in from one clause.

Tables are for looking things up

Readers don't read tables from top to bottom. They scan for the row they care about. That makes tables the right place for exact values, for many variables side by side and for anything a meta-analyst will want to pull out later. If someone might need your third decimal place, that number goes in a table.

Figures are for patterns the eye can catch faster than the brain can parse

A figure earns its space when it shows something a table can't: a non-linear relationship, a distribution with a long tail, two groups that diverge at week six. If your figure's message could be stated as "group A was higher than group B," you probably don't need the figure.

One more opinion here, and it's not a universal view: I'd rather see one excellent figure than four adequate ones. Journals cap display items for a reason, and running into the limit is usually a sign of weak selection, not rich data.

How the same number ends up in three places, and drifts

The real danger of repetition isn't wasted space. It's inconsistency.

Years ago, a lab manager I worked with, Denise, kept the master spreadsheet for a multi-site study I was co-authoring. Two days before submission she sent the group an email with the subject line "which n is right?" She had found that our analyzed sample was 412 in the abstract, 409 in Table 2, and 412 in the figure legend, while the text described "over 400 participants." Three people had updated three locations at different points after we excluded three participants with incomplete follow-up. Nobody had been careless, exactly. The numbers had simply lived in too many places.

We caught it. Many authors don't. When a reviewer finds two different values for the same quantity, they stop trusting every other number in the paper, and they're right to. I've seen decent manuscripts get a "major revision" largely because of this.

The fix Denise put in place afterward was simple, and I've recommended it ever since. Every number in the manuscript traces back to one cell or one output file, and the tables are generated from analysis outputs rather than typed by hand. If you work in R or Python, building tables directly from code is worth the setup time. If you don't, keep a "numbers log" listing each reported value and where it comes from.

Writing text that points instead of repeats

The job of the results text is to tell the reader what to notice. A table can't do that. Only a sentence can.

Here's a typical before:

Before: "Mean systolic blood pressure at baseline was 142.3 mmHg in the intervention group and 141.8 mmHg in the control group. At 12 weeks it was 131.6 mmHg and 139.2 mmHg, respectively. Diastolic pressure at baseline was 88.1 and 87.9 mmHg, and at 12 weeks 83.4 and 86.8 mmHg (Table 2)."

That's Table 2 read aloud. The reader learns nothing the table doesn't already show, and they still have to work out the finding themselves.

After: "Groups were similar at baseline (Table 2). By 12 weeks, systolic pressure had fallen more in the intervention group (adjusted difference −7.4 mmHg, 95% CI −10.1 to −4.7). The diastolic difference was smaller and less certain (Table 2)."

The rewrite reports the one number that matters, states it as an effect with uncertainty and sends the reader to the table for everything else. It's shorter and it does more work.

A few sentence patterns I use constantly:

• "Most baseline characteristics were balanced between groups (Table 1); the exception was smoking status, which was more common in the control group."

• "The association was strongest in participants over 65 (Figure 2B)."

• "Results were similar across all sensitivity analyses (Supplementary Table S3)."

Notice that each one makes a claim and gives a location. That's the pattern. Don't write "Table 1 shows the baseline characteristics." That sentence does nothing. The table's title already says so.

Tables reviewers can actually read

The most common table problems I see are mechanical, which means they're fixable in an afternoon:

• Inconsistent precision. Means to three decimals in one column and one decimal in the next. Pick a precision that fits the measurement. Nobody's age needs a decimal place.

• Units buried or missing. Put units in the column or row header, not in every cell.

• "p = 0.000". Report "p < 0.001". Better still, lead with effect sizes and confidence intervals and let p-values be secondary.

• Columns in the wrong order. Put the comparison you care about next to each other. If the point is intervention versus control, those columns should be adjacent, not split by a "total" column.

• Abbreviations defined only in the main text. Every table should stand on its own. Define abbreviations in a footnote, even if you defined them on page three.

• Vertical rules everywhere. Most journals strip them anyway. Use white space and a few horizontal lines.

One test I give students: hand the table, without the paper, to a colleague in a neighboring field. If they can't say in one sentence what it shows, the table isn't finished.

Figures that earn their space

Some years back, I handled a manuscript where the main result was a bar chart: two bars, mean plus standard error, a satisfying gap between them and an asterisk on top. One of the reviewers, someone I've never met and whose name I never learned, wrote a short, polite comment asking the authors to replot it showing each individual data point.

They did. The new figure showed that one group had six observations and the "effect" came almost entirely from two of them. The authors softened their conclusions considerably, and the paper was better for it. I still think about that reviewer whenever I see a mean-and-error-bar plot of a small sample.

That reviewer was echoing a well-documented critique. Weissgerber and colleagues made the case that bar graphs of continuous data hide the distribution and can mislead, and their work is easy to find on PubMed. My practical version:

• For continuous data with small samples, show the individual points (dot plots, strip plots, box plots with points overlaid).

• Reserve bar charts for counts and proportions, where a bar actually represents a total.

• Say in the legend what the error bars are. SD, SE and 95% CI tell very different stories, and reviewers will ask.

• Label axes with the quantity and the unit, in words a reader understands. "Rel. expr. (AU)" is not a label.

• Use color palettes that work for colorblind readers, and make sure the figure still makes sense in grayscale.

• Start the y-axis at zero for bar charts. For other plot types it's a judgment call, but make the judgment honestly.

And if your figures include images such as blots, micrographs or scans, keep the unprocessed originals. Image integrity concerns are among the most common problems editors escalate, and journals increasingly follow COPE guidance when adjustments look like more than brightness and contrast applied to the whole image.

Some display items aren't optional

Reporting guidelines settle part of the table-or-figure question for you, so check them before you design anything. Trials reported under CONSORT need a participant flow diagram and a baseline characteristics table. Systematic reviews need a PRISMA flow diagram showing records identified, screened, excluded and included. Observational studies under STROBE have their own expectations for participant flow and descriptive data.

The EQUATOR Network indexes these guidelines by study type, and it's the first place I send anyone who isn't sure which one applies. Editors notice when a required item is missing, and it's an avoidable reason to be sent back.

Use these required items as anchors. If the flow diagram already shows that 38 participants were lost to follow-up, your text doesn't need to list the reasons again. Point to the figure and move on.

A pre-submission check for your results section

Before I'd send a manuscript out, I'd run through this:

1. List every number in the results. Does each one have exactly one home?

2. Do sample sizes match across abstract, text, tables, figures and legends?

3. Does every paragraph of results text make a claim, rather than just narrating a table?

4. Can each table and figure be understood without reading the main text?

5. Is the primary outcome reported with an effect size and a measure of uncertainty, not only a p-value?

6. Do any figures hide small samples behind bars and error bars?

7. Are all display items required by your reporting guideline present?

8. Could any table or figure move to the supplement without weakening the argument? If so, move it.

9. Are display items numbered in the order they're first cited in the text?

Item 8 is the one people resist. Do it anyway. The supplement isn't a demotion. It's where thoroughness belongs.

Where the drift actually starts

In my experience, inconsistent numbers rarely start at the writing stage. They start earlier, when notes, summaries of other papers and draft paragraphs live in different files and nobody can say which version is current. That's part of why I've been watching tools that keep the whole process in one place. PACR, for example, is an AI-native research workspace that connects searching, reading, summarizing, writing and collaborating, so the context you build while reading carries through into the draft rather than getting lost between apps. That won't choose your figures for you, but it does make "where did this number come from?" an easier question to answer.

If you have a results section open right now, try one thing before anything else: print it, take three highlighters, and mark every number by where it appears, whether text, table or figure. Wherever you see the same number in two colors, you've found an edit. And if you'd like a single place to keep the reading and writing behind those numbers together, you can set up a PACR workspace and try it on your next paper.