Free scoping read. A written assessment of your manuscript in two business days, at no cost.

Start one
Contents

Academy · Research Writing · Tables and figures

Tables and Figures in Research Manuscripts: Selection, Sizing and Captioning

A display item is justified when it performs a function that prose cannot. Tables convey exact values across many variables; figures convey shape, trend and distribution; and a single two-number comparison belongs in a sentence. The caption is usually written last but read first, which reverses the appropriate order of priority.

Pubrica AcademyWritten by the medical writing and biostatistics teams Reviewed byPubrica biostatistics team Published9 October 2026 Last checked9 October 2026 Reading time12 minutes Sources29 peer-reviewed

A results section is constrained by a word limit and typically holds more data than either the text or the word limit can accommodate. The choice of display format is therefore not cosmetic: a table that should have been a figure wastes the reader’s attention, a figure that should have been a sentence wastes a display slot, and a caption that cannot be understood without the body text fails the principal test that journals apply. This article serves both as a decision guide and as an index to related Pubrica Academy material on presenting data in a paper.

Learning objectives

  • The circumstances in which a table is preferable to a figure, a figure to a table, and a single sentence to both
  • The limits that journals impose (rows, columns, page extent and numbering), with one journal’s published caps as a worked example
  • The required content of a self-contained caption, with a reusable skeleton
  • The failure modes that reviewers most often identify: duplication of the text, means without measures of spread, truncated axes, missing denominators and colour-only encoding
  • The reasons a baseline table should not report p-values across randomised arms
  • How display items are checked against the reporting guideline for the study design

Common misconceptions

  • “Figures are more persuasive than tables.” This holds only for shape. When a reader requires the exact value, for comparison with other data or for use in a meta-analysis, a figure withholds it.
  • “The caption describes the figure.” The caption must make the figure interpretable on its own: what is plotted, in whom, how many, what the error bars represent and what the abbreviations mean. “Figure 2. Survival by group” is a label, not a caption.
  • “More rows means more information.” The Radiology instructions cap a main table at 40 rows and 6 to 8 columns and require it to fit on one page. A table that cannot be read across conveys no information.
  • “Table 1 should show whether randomisation worked.” P-values comparing randomised arms at baseline test a hypothesis whose answer is already known; the relevant information is the size of any imbalance, not its p-value.
  • “Display items carry no cost.” Each display item has a cost. A typical manuscript runs to about eighteen paragraphs across five sections, so each display item competes with a paragraph for the reader’s attention and for space.

Choosing between a table, a figure and a sentence

The choice should begin with what the reader must do with the number. If the exact value is required — for comparison with another series, for extraction into a meta-analysis or for verification of the arithmetic — a table is necessary, because a figure conveys only position. If a shape must be seen — a trend over time, a distribution, a relationship between two continuous variables, a survival curve or the spread of individual points around a mean — a figure is necessary, because a table of coordinates does not convey shape to a human reader.

If the entire content is a single comparison of two numbers, it belongs in a sentence. “Mean operating time was 94 minutes in the robotic group and 112 in the open group” occupies one line and no display slot. A two-row, two-column table conveying the same information spends a numbered item on content that prose handles better, and reviewers tend to notice this.

Three further rules of thumb apply. A table with one row is a sentence. A figure with three bars is a sentence. A figure showing a difference that the analysis found inconclusive invites the reader to perceive an effect that the statistics do not support, which makes it a presentation decision with interpretive consequences.

Table, figure or sentence — decide by what the reader has to do with the number

What the reader needsFormWhyWhat goes wrong otherwise
The exact value, to compare or extractTableA figure gives position, not value.Reviewers commonly request the numbers that were plotted.
Many variables across a few groupsTableRows and columns let a reader look things up.A clustered bar chart with eight categories is unreadable.
A shape: trend, distribution, relationshipFigureA human reads shape from a picture, not a coordinate list.A table of 40 time points hides the trend it contains.
Time to eventFigure (plus numbers at risk)The curve is the finding; the table under it gives the precision.A median alone discards the shape of the difference.
Individual values, small sampleFigure (points, not bars)Overlap and outliers are the information.Bars with error whiskers overstate separation.
One comparison of two numbersA sentenceIt costs a line and no display slot.A two-by-two table spends a numbered item on nothing.
Participant or screening flowDiagram, content fixed by the guidelineCONSORT and PRISMA 2020 specify what must appear.Numbers drawn from memory do not reconcile with the text.

Decision framework assembled from the reporting guidelines cited on this page (CONSORT 2010, PRISMA 2020, STROBE) and from editorial practice. The row-by-row judgements in the last two columns are ours, not drawn from a single source.

Journal limits on display items and their rationale

Most journals cap the number of display items and set format requirements that authors discover at proof stage. The Radiology author instructions are a usefully explicit example: each table must fit on a single page, main tables should carry no more than 40 rows and 6 to 8 columns, and tables and figures are numbered in order of first appearance in the text 28. These limits are not arbitrary. A table that runs over a page break cannot be read across; beyond approximately eight columns a reader loses track of the row being followed; and numbering out of order means that the reader encounters Figure 3 before Figure 2 and must search for it.

The limits must also be read against the word budget, because display items and prose compete for space. The same instructions allow an Original Research article 3,000 words in total — 400 for the introduction, 800 for materials and methods, 1,000 for results, 800 for discussion, with a 300-word abstract and a maximum of 35 references 28. Kliewer, writing in the American Journal of Roentgenology, describes the resulting architecture as roughly eighteen paragraphs across five sections, with three introduction paragraphs before the purpose statement and five discussion paragraphs before the summary 19. Against eighteen paragraphs, four or five display items constitute a substantial share of the paper, and each must justify the space of a paragraph.

Where a table genuinely requires 60 rows, the usual solution is to divide it according to what the reader will look up, or to move the detail to a supplement and retain the summary in the paper. The Pubrica manuscript formatting page covers reconciling these requirements across journals when a paper is redirected.

40maximum rows in a main table
6–8columns in a main table
1 pageeach table must fit on a single page

Source: Radiology scientific style guide and author instructions (Radiological Society of North America), which also requires tables and figures to be numbered in order of first appearance in the text. Other journals set different limits; check the instructions for yours.

Required elements of a self-contained caption

The test journals apply is whether the display item can be understood without the body text. That matters more than it used to: figures are extracted into slide decks, indexed separately, pulled into systematic reviews and reproduced in secondary coverage, usually with the caption and nothing else.

A caption that passes this test contains six elements: what is plotted or tabulated, in plain terms; in whom, meaning the population and any restriction; how many, with the denominator for each group; the units and the time point; what any error bars, shading or whiskers represent, which is the most commonly omitted item and the one that most alters how a figure is read; and every abbreviation expanded, including those expanded earlier in the text.

Where a figure shows an estimate, the caption should also say what the model adjusted for, or say that it is unadjusted. GRADE summary-of-findings tables, as set out by Guyatt and colleagues in the Journal of Clinical Epidemiology, are a good model for this discipline: the table footnotes carry the reasons for every certainty rating, so a reader can reconstruct the judgement without going back to the text 13,17.

Dependent and self-contained captions compared

The difference lies almost entirely in information density, not style. The weaker version is not poorly written; it simply assumes that the reader has the methods section to hand. Because the most likely future reader of a figure is someone who has only the figure, that assumption is the error.

The same figure caption, before and after it can stand alone

Unreadable without the methods section

Figure 2. Change in outcome over time in the two groups. *P < 0.05. Error bars show variability.

“Outcome” is not named, so the reader cannot tell what was measured or in what units.

No population and no denominators. “The two groups” could be anyone, at any size.

“Error bars show variability” is the most consequential omission on the page: standard deviation, standard error and a confidence interval look identical and mean quite different things.

An asterisk with no named test, no comparison specified and no mention of multiplicity, so the claim cannot be assessed.

Nothing about adjustment or missing data, so a reader cannot tell what the plotted means represent.

Readable on its own, anywhere

Figure 2. Mean change in six-minute walk distance (metres) from baseline to 12 weeks in adults aged 40–75 with stable COPD, randomised to supervised pulmonary rehabilitation (n = 84) or usual care (n = 81). Points are group means at each assessment; error bars are 95% confidence intervals. Between-group difference at 12 weeks: 38 m (95% CI 14 to 62), from a mixed model for repeated measures adjusted for baseline distance and centre; P = 0.002. Analysis by intention to treat; 12 participants (7%) had no 12-week measurement and were handled under a missing-at-random assumption. COPD = chronic obstructive pulmonary disease.

Outcome, units, population, time points and group sizes are all present, so the figure survives being extracted from the paper.

The error bars are defined, which fixes how the overlap should be read.

The effect, its interval, the model and what it adjusted for are stated, so the figure carries the estimate rather than implying it.

The analysis population and the missing-data handling are named, pre-empting the first two reviewer questions.

The abbreviation is expanded, even though it appeared earlier in the text, because the caption may be read alone.

The second caption is six times longer, and this is intentional: caption text is usually outside the word count, and it is the least costly place in a manuscript to answer questions. It also matters beyond peer review, because this is the version that remains interpretable when the figure is transferred to a slide, indexed by a search engine or extracted for a systematic review, which is the most likely future use of the figure.

Common display-item failures identified by reviewers

The following failures are listed in approximate order of frequency in Pubrica’s experience.

Duplicating the text. A results paragraph that recites every cell of a table, or a table that repeats numbers already given in prose, wastes both. Prose should refer to the display item and state what it shows, rather than restate its contents.

Bar charts of means with no spread. A bar chart of two means conveys nothing about overlap, and bars with standard-error whiskers systematically make differences appear larger than they are. Where the sample is small, the individual points should be plotted.

Truncated or unlabelled axes. An axis that does not start at zero can be legitimate, but this must be evident; a bar chart with a truncated axis is not legitimate, because bar length encodes magnitude. Every axis requires a label and a unit.

Missing denominators. Percentages are reported without the n from which they are derived, or a figure presents groups of different and unstated sizes. The n should be given in the caption or on the axis.

Significance asterisks without the test. An asterisk is uninformative unless the caption names the test, the comparison and the correction for multiplicity, if any.

Colour as the only encoding. Figures are read in greyscale, on poor-quality projectors and by readers with colour vision deficiency. Shape, line style or direct labelling should be used in addition to colour. The Pubrica visual accessibility editing page covers the checks.

Screenshots and resampled images. A figure rebuilt from a screenshot of a statistics package will not survive production. Figures should be exported at the resolution and in the format that the journal specifies.

The baseline table and inappropriate p-values

Table 1 of a trial report describes the randomised groups. It is routinely published with a column of p-values comparing them, and that column should be omitted. Randomisation ensures that any baseline difference arose by chance, so testing for one tests a hypothesis already known to be true; a “significant” baseline difference in a properly randomised trial is a chance finding by construction 14. The CONSORT 2010 statement, set out by Schulz, Altman and Moher in Trials, asks for baseline demographic and clinical characteristics for each group — the numbers, not tests of them 14.

What a reader requires is the size of any imbalance and the denominator for every cell, in order to judge whether an imbalance is large enough to affect the outcome. For observational studies the situation differs: group differences are not chance by construction, so describing and quantifying them is part of the analysis, and the reporting guideline for the relevant study design specifies what to present.

Flow diagrams as display items with fixed content

The participant flow diagram in a trial report and the screening flow diagram in a systematic review are not optional illustrations; they are specified by the reporting guideline and their content is fixed. CONSORT 2010 specifies the flow of participants through each stage with numbers at each step 14. The PRISMA 2020 statement, published by Page and colleagues in BMJ, specifies the flow from records identified through to studies included, with reasons for exclusion at each stage 5.

The common error is to draw the diagram at the end from memory, which produces numbers that do not reconcile with the text. The diagram should be built from the screening log as screening proceeds, and the arithmetic checked in both directions: every number entering a box must leave it, either as an inclusion or as a counted exclusion. A diagram whose numbers do not sum correctly rapidly undermines a reviewer’s confidence, because the error can be verified in thirty seconds.

Pages: systematic review and meta-analysis for the flows themselves, and reporting checklist review for an audit of the completed checklist before submission.

Checking display items against the reporting guideline

Reporting checklists are usually applied to the prose rather than to the tables and figures, which is where several items are located. STROBE, published by von Elm and colleagues in BMJ, contains 22 items, of which 18 apply across all three designs it covers — cohort, case-control and cross-sectional — and 4 are design-specific 12,18,19. Several of those items are satisfied by a display item or not at all: the numbers of participants at each stage, the characteristics of participants with the extent of missing data, unadjusted and adjusted estimates with their precision 18.

The checklist should therefore be read by asking, for each item, “where does a reader find this”. If the answer is a table, the table should contain the item in full, including the missing-data counts, which are the most commonly omitted. If the answer is nowhere, that is itself the finding.

Further resources on presenting data

This page covers the decision and the caption. The surrounding material covers production, alternative formats and the wider structure of the paper.

Related pages

If you need…Go to
The paper’s overall structure and section functionsMedical paper structure
The whole path from finished study to published paperWriting and publishing an original research article
Figures prepared to a journal’s production specificationFigure production
A single figure that summarises the paperGraphical abstract
Colour, contrast and greyscale checksVisual accessibility editing
The checklist for your design, and which items live in a tableReporting guidelines
An audit of the completed checklist before submissionReporting checklist review
Help with the analysis behind the display itemsBiostatistics or statistical review
Formatting to a specific journal’s requirementsManuscript formatting
Why manuscripts get rejected before reviewReasons for manuscript rejection

Navigation table.

Key points

Key points

The form of a display item should be chosen according to what the reader must do with the number: exact values and many variables belong in a table, shape and distribution in a figure, and a single two-number comparison in a sentence. Journal limits should be respected, since a table that runs over a page break or exceeds eight columns cannot be read; one journal caps a main table at 40 rows, 6 to 8 columns and a single page. Effort should then be concentrated on the caption, which is usually outside the word count and is the only part that travels with the figure: it should state what is plotted, in whom, how many, in what units, what the error bars represent and what the model adjusted for.

Application to your study

Application to your study

  1. For every display item, the task that the reader must perform with the number should be recorded. A need for exact values indicates a table, and a need to see shape indicates a figure. If the task is only to see that one number is larger than another, the item should be replaced by a sentence.
  2. The number of display items should be weighed against the number of paragraphs. A manuscript of around eighteen paragraphs cannot support eight tables; authors should decide which items justify the space of a paragraph each.
  3. Each table should be checked against the journal’s limits before formatting (rows, columns and single-page fit), and detail should be split or moved to a supplement rather than the font size being reduced.
  4. Every caption should be rewritten to stand alone: what is plotted, in whom, n per group, units, time point, what the error bars represent, what the model adjusted for, and every abbreviation expanded.
  5. Error bars should be defined explicitly. Standard deviation, standard error and confidence interval appear identical in a figure but are not equivalent. This is the single most common caption omission.
  6. Any p-value column comparing randomised arms at baseline should be removed and replaced with the imbalance itself, together with the denominator for every cell.
  7. Every flow diagram should be reconciled with the text in both directions. Every number entering a box must leave it as an inclusion or a counted exclusion.
  8. Display items should be reviewed in greyscale to confirm that no information depends on colour alone, with shape, line style or direct labels added where necessary.
  9. The reporting checklist should be applied by asking “where does a reader find this” for each item, and by confirming that items expected in a table are present there, particularly the missing-data counts.

Frequently asked questions

The decision should depend on what the reader must do with the number. If the exact value is required — for comparison with other data, for verification of the arithmetic or for extraction into a later meta-analysis — a table is necessary, because a figure conveys position rather than value. If a shape must be seen — a trend, a distribution, a survival curve or the spread of individual points — a figure is necessary, because a human reader does not read shape from a coordinate list. Many variables across a few groups favour a table; few variables across many time points favour a figure. If the content is a single comparison of two numbers, it belongs in a sentence.

Smaller than most authors would prefer. The Radiology instructions cap a main table at no more than 40 rows and 6 to 8 columns and require it to fit on a single page, with tables and figures numbered in order of first appearance. Other journals set different numbers, but the underlying constraints are the same: a table that breaks across a page cannot be read across, and beyond approximately eight columns a reader loses track of the row being followed. If a table genuinely requires 60 rows, it should be divided according to what the reader will look up, or a summary should be retained in the paper and the full version moved to a supplement.

A caption that stands alone requires six elements: what is plotted, in plain terms; in whom, including any restriction on the population; how many, with the denominator for each group; the units and the time point; what any error bars, whiskers or shading represent; and every abbreviation expanded, even those expanded earlier in the text. Where the figure shows an estimate, the caption should also state what the model adjusted for, or that the estimate is unadjusted. The error-bar definition is the item most often missing and the one that most alters interpretation, because a standard deviation, a standard error and a confidence interval appear identical on the page.

The figure will be read without the methods. Figures are transferred to slide decks, indexed separately by search engines, reproduced in secondary coverage and extracted into systematic reviews, usually with only the caption attached. A reader in any of these situations cannot consult the description of the population or the error bars. Caption text is also usually outside the word count, which makes it the least costly place in the manuscript to anticipate a reviewer’s question. The cost is a caption that appears long beside others; the benefit is a figure that remains interpretable for years.

No. Randomisation ensures that any baseline difference between arms arose by chance, so a test for one tests a hypothesis already known to be true, and a “significant” baseline difference in a properly randomised trial is by construction a chance finding. CONSORT 2010, as set out by Schulz, Altman and Moher in Trials, asks for the baseline demographic and clinical characteristics of each group — the numbers, not tests of them. What a reader requires is the magnitude of any imbalance and the denominator for every cell, in order to judge whether it matters for the outcome. Observational studies are a different case, in which quantifying group differences is part of the analysis.

In approximate order of frequency: prose that recites a table cell by cell instead of referring to it; bar charts of means with no indication of spread, or with standard-error whiskers that make differences appear larger than they are; axes without labels or units, and truncated axes on bar charts, where bar length encodes magnitude; percentages without the denominator from which they were derived; significance asterisks with no named test, comparison or correction for multiplicity; colour used as the only encoding, which fails in greyscale and for readers with colour vision deficiency; and figures rebuilt from screenshots that do not survive production. All of these can be checked in a single pass before submission.

Usually fewer than the maximum. The relevant constraint is competition for the reader’s attention rather than a cap: Kliewer, in the American Journal of Roentgenology, describes a typical manuscript as roughly eighteen paragraphs across five sections, with three introduction paragraphs before the purpose statement and five discussion paragraphs before the summary. Against eighteen paragraphs, four or five display items already constitute a large share of the paper, and each should justify a paragraph of prose. The appropriate test for each candidate item is whether its deletion would remove something the reader needs, or only something the authors collected.

Several checklist items are located there and nowhere else, which is why applying a checklist to the prose alone misses items. STROBE, published by von Elm and colleagues in BMJ, contains 22 items, 18 of which apply across cohort, case-control and cross-sectional designs and 4 of which are design-specific; among them are the numbers of participants at each stage, participant characteristics with the extent of missing data, and both unadjusted and adjusted estimates with their precision. CONSORT 2010 and PRISMA 2020 both specify a flow diagram whose content is fixed. Each checklist item should be read by asking where a reader would find it, and the tables checked accordingly.

References

  1. International Committee of Medical Journal Editors (ICMJE). Recommendations for the Conduct, Reporting, Editing, and Publication of Scholarly Work in Medical Journals. 2026. read 9 October 2026. icmje.org
  2. Bloothooft M, Pennings HJM, van der Heyden MAG. The Utrecht University Honours Program review project: example based scientific publishing training aimed at bachelor medical students. BMC Medical Education. 2024;24(1):1-12. doi:10.1186/s12909-024-05098-7 read 9 October 2026. doi.org/10.1186/s12909-024-05098-7
  3. Borowiec BG. Ten simple rules for scientists engaging in science communication. PLOS Computational Biology. 2023;19(7):e1011251. doi:10.1371/journal.pcbi.1011251 read 9 October 2026. doi.org/10.1371/journal.pcbi.1011251
  4. Churruca K, Pomare C, Ellis LA, Long JC, Henderson SB, Murphy LED, Leahy CJ, Braithwaite J. Patient-reported outcome measures (PROMs): a review of generic and condition-specific measures and a discussion of trends and issues. Health Expectations. 2021;24(4):1015-1024. doi:10.1111/hex.13254 read 9 October 2026. doi.org/10.1111/hex.13254
  5. Page MJ, McKenzie JE, Bossuyt PM, et al. The PRISMA 2020 statement: an updated guideline for reporting systematic reviews. BMJ. 2021:n71. doi:10.1136/bmj.n71 read 9 October 2026. doi.org/10.1136/bmj.n71
  6. Jawaid SA, Jawaid M. Common reasons for not accepting manuscripts for further processing after editor's triage and initial screening. Pakistan Journal of Medical Sciences. 2019;35(1):1-3. doi:10.12669/pjms.35.1.28 read 9 October 2026. doi.org/10.12669/pjms.35.1.28
  7. Pasarica M, Bailey M, Cendan JC. Increasing students' publication productivity: could launching a university scientific journal be a catalyst? Cureus. 2019;11(1):e3953. doi:10.7759/cureus.3953 read 9 October 2026. doi.org/10.7759/cureus.3953
  8. Sterne JAC, et al. RoB 2: a revised tool for assessing risk of bias in randomised trials. BMJ. 2019;366:l4898. read 9 October 2026. frontiersin.org
  9. Fujita MSL, Agustin-Lacruz MdC, Terra AL. Journals' guidelines about title, abstract and keywords: an overview of Information Science and Communication Science areas. European Science Editing. 2018;44(4):76-79. read 9 October 2026. europeanscienceediting.org.uk
  10. Bannas P, Reeder SB. How to write an original radiological research manuscript. European Radiology. 2017;27(11):4455-4460. doi:10.1007/s00330-017-4879-8 read 9 October 2026. doi.org/10.1007/s00330-017-4879-8
  11. Wager E, Woolley K, Adshead V, Cairns A, Fullam J, Gonzalez J, Grant T, Tortell S. Awareness and enforcement of guidelines for publishing industry-sponsored medical research among publication professionals: the Global Publication Survey. BMJ Open. 2014;4(4):e004780. doi:10.1136/bmjopen-2013-004780 read 9 October 2026. doi.org/10.1136/bmjopen-2013-004780
  12. von Elm E, Altman D, Egger M, Pocock S, Gotzsche P, Vandenbroucke J. The STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) statement: guidelines for reporting observational studies. International Journal of Surgery. 2014. doi:10.1016/j.ijsu.2014.07.013 read 9 October 2026. doi.org/10.1016/j.ijsu.2014.07.013
  13. Guyatt G, Oxman AD, Akl EA, et al. GRADE guidelines: 1. Introduction - GRADE evidence profiles and summary of findings tables. Journal of Clinical Epidemiology. 2011;64(4):383-394. read 9 October 2026. link.springer.com
  14. Schulz KF, Altman DG, Moher D. CONSORT 2010 Statement: updated guidelines for reporting parallel group randomised trials. Trials. 2010;11(1). doi:10.1186/1745-6215-11-32 read 9 October 2026. doi.org/10.1186/1745-6215-11-32
  15. Mokkink LB, Terwee CB, Patrick DL, et al. The COSMIN study reached international consensus on taxonomy, terminology, and definitions of measurement properties for health-related patient-reported outcomes. Journal of Clinical Epidemiology. 2010;63(7):737-745. doi:10.1016/j.jclinepi.2010.02.006 read 9 October 2026. doi.org/10.1016/j.jclinepi.2010.02.006
  16. Kuper A, Reeves S, Levinson W. An introduction to reading and appraising qualitative research. BMJ. 2008;337:a288. read 9 October 2026. link.springer.com
  17. Guyatt GH, Oxman AD, Vist GE, et al. GRADE: an emerging consensus on rating quality of evidence and strength of recommendations. BMJ. 2008;336(7650):924-926. read 9 October 2026. link.springer.com
  18. von Elm E, Altman DG, Egger M, Pocock SJ, Gotzsche PC, Vandenbroucke JP, et al. Strengthening the reporting of observational studies in epidemiology (STROBE) statement: guidelines for reporting observational studies. BMJ. 2007;335(7624):806-808. doi:10.1136/bmj.39335.541782.AD read 9 October 2026. doi.org/10.1136/bmj.39335.541782.AD
  19. Kliewer MA. Writing it up: a step-by-step guide to publication for beginning investigators. American Journal of Roentgenology. 2005;185:591-596. read 9 October 2026. tech.snmjournals.org
  20. Sollaci LB, Pereira MG. The Introduction, Methods, Results, and Discussion (IMRAD) structure: a fifty-year survey. Journal of the Medical Library Association. 2004;92(3):364-371. read 9 October 2026. genreacrossborders.uwaterloo.ca
  21. International Committee of Medical Journal Editors. Uniform requirements for manuscripts submitted to biomedical journals. Annals of Internal Medicine. 1997;126(1):36-47. read 9 October 2026. stat.cmu.edu
  22. Vickery B. The Royal Society Scientific Conference of 1948. Journal of Documentation. 1992;54(3):281-283. read 9 October 2026. stat.cmu.edu
  23. Atkinson D. The evolution of medical research writing from 1735 to 1985: the case of the Edinburgh Medical Journal. Applied Linguistics. 1992;13(4):337-374. read 9 October 2026. stat.cmu.edu
  24. Huth EJ. Structured abstracts for papers reporting clinical trials. Annals of Internal Medicine. 1987;106(4):626-627. read 9 October 2026. stat.cmu.edu
  25. Meadows AJ. The scientific paper as an archaeological artifact. Journal of Information Science. 1985;11(1):27-30. read 9 October 2026. stat.cmu.edu
  26. Bazerman C. Modern evolution of the experimental report in physics: spectroscopic articles in Physical Review, 1893-1980. Social Studies of Science. 1984;14:163-196. read 9 October 2026. stat.cmu.edu
  27. GBD 2021 Nervous System Disorders Collaborators. Global, regional, and national burden of disorders affecting the nervous system, 1990-2021: a systematic analysis for the Global Burden of Disease Study 2021. The Lancet Neurology. 23(4):344-381. read 9 October 2026. ro.uow.edu.au
  28. Radiology (Radiological Society of North America). Radiology scientific style guide and author instructions: scientific editing. Radiology. read 9 October 2026. pubs.rsna.org
  29. Fernandez E, Garcia AM, Seres E, Bosch F. Students' satisfaction and perceived impact on knowledge, attitudes and skills after a 2-day course in scientific writing: a prospective longitudinal study in Spain. BMJ Open. 8(1):e018657. read 9 October 2026. bmjopen.bmj.com

Every source above is a peer-reviewed article, a publisher’s record of one, or a standard issued by the body that owns it. Each carries the date it was last read. Where a standard moves, this article is revised rather than left up with a note.

How to cite this article

Pick your referencing style and copy it. Your university or target journal may vary the punctuation — check its guide if it has one.

Pubrica Academy. (2026, October 9). Tables and Figures in Research Manuscripts: Selection, Sizing and Captioning. Pubrica. https://pubrica.com/academy/tables-figures/

This article is written to be learned from. Cite it if you are referring to what it says; if you are using the evidence behind it, cite the sources listed above instead.

Caption skeleton, for a figure and for a table

The bracketed slots should be completed and inapplicable elements deleted. The order is deliberate: what, in whom, how many, in what units, what the bars represent and what the model did. Caption text is usually outside the word count, so brevity is unnecessary here.

FIGURE CAPTION SKELETON

Figure [n]. [What is plotted, named in full, with units] in [population,
including age range and any restriction], [grouped / randomised] to
[group A] (n = [__]) and [group B] (n = [__]), from [start time point]
to [end time point].
[Markers] are [group means / medians / individual participants];
error bars are [95% confidence intervals / standard deviations /
interquartile ranges -- SAY WHICH].
[Between-group difference / effect estimate] at [time point]:
[estimate] [units] ([95]% CI [lower] to [upper]), from
[model], adjusted for [covariates / state "unadjusted"]; P = [value],
[test named].
Analysis by [intention to treat / per protocol]; [n] participants
([__]%) had no [time point] measurement and were handled by
[method].
[ABBREVIATION] = [expansion]. [Repeat for every abbreviation, even if
expanded earlier in the text.]


TABLE CAPTION SKELETON

Table [n]. [What is tabulated] in [population], by [grouping variable]
([group A], n = [__]; [group B], n = [__]).
Values are [n (%) / mean (SD) / median (IQR)] unless stated otherwise.
Denominators differ where data were missing; the number with available
data is given in each cell / in footnote [x].
[Footnote a] [Definition of any derived variable.]
[Footnote b] [Source of any figure not generated in this study.]
[Footnote c] [Why any cell is blank -- not applicable vs not measured.]
[ABBREVIATION] = [expansion].


PRE-SUBMISSION DISPLAY-ITEM CHECK
  For EACH table and figure:
  [ ] Form matches the reader's need (exact value -> table; shape -> figure;
      one comparison -> delete and write a sentence)
  [ ] Caption stands alone: what, in whom, n, units, time point
  [ ] Error bars / whiskers / shading DEFINED
  [ ] Adjustment stated, or "unadjusted" stated
  [ ] Every abbreviation expanded in the caption itself
  [ ] Denominator present for every percentage
  [ ] Axes labelled with units; no truncated axis on a bar chart
  [ ] No significance marker without a named test and comparison
  [ ] Legible in greyscale; nothing encoded by colour alone
  [ ] Within the journal's row / column / page limits
  [ ] Numbered in order of first mention in the text
  [ ] Exported in the journal's named format and resolution
  [ ] No number here contradicts the text or the flow diagram
  [ ] No baseline p-values comparing randomised arms
  WHOLE PAPER:
  [ ] Display items counted against paragraphs -- each one worth its space
  [ ] Reporting checklist read item by item asking "where does a reader
      find this?", and the table items verified present
  [ ] Flow diagram reconciles with the text in both directions

Pubrica AcademyWritten and maintained by Pubrica’s subject matter experts. Articles are revised when the underlying standards change, and each article states the date on which it was last reviewed.

How Pubrica can help

Pubrica’s editors and statisticians review display items as a set rather than individually: whether each should be a table, a figure or a sentence, whether the captions stand alone, and whether the numbers reconcile with the text and with the flow diagram. Pubrica also prepares figures to a specific journal’s production specification, including greyscale and colour-blind checks, so that the published version matches the version the authors approved.

Find something else

Search the Pubrica Academy

Access is free and requires no account. Enter a topic, or open the full library to filter by category, sub-topic or resource type.

On receipt of a manuscript with its tables and figures, Pubrica assesses which display items are justified.

A scoping assessment is free and takes two business days. It provides a written assessment of what the work requires and which specialist would undertake it.

Get a free quote

Regulations

Check your project against this rule

  1. 1Your work
  2. 2Your details
  3. 3Send
What do you need?

Material you send is held under signed confidentiality, with access limited to the assigned team and encrypted transfer. It is not used to train anything and is not passed to third-party tools. Ask for the agreement before you send anything you would rather cover first.