Tables for evidence and explanation
Tables organise related words and numbers for comparison or reference. Their structure, headings, captions and notes should make the intended relationships clear without requiring a search through the surrounding text.
Purposes and types
Tables contain words and numbers rather than artwork; drawings, photographs and diagrams belong to figures. Tables offer four advantages.
- Presenting datasets or lists, especially long lists.
- Combining words and numerical data cleanly.
- Replacing long, complex sentences containing many items.
- Breaking up uninterrupted pages of prose.
A short, logically grouped list integrated into prose can serve as a text table, with an introduction but no separate legend or formal callout. A formal table stands separately and needs a reference in the text. Constructing a central table before drafting can clarify the information to communicate, as when tabulating the presenting signs and symptoms of meningococcal infection. Developing tables can be as creative as composing prose.
Four table types serve different reading tasks.
- Reference tables present data in a grid: readers locate a category in the left column, read across for related values and compare rows, as in study reports and manuscripts.
- Process tables show who does each task in a long procedure and when, as in clinical study protocols.
- Decision tables help readers make choices and can replace procedural explanations overloaded with conditions, as in instructions for investigators and trial nurses.
- Action–response tables pair actions with their responses, especially human actions and machine responses; medical writers use them infrequently.
The question the data answer should determine the display's main message and the choice of presentation.
Tables suit tasks requiring actual values, including supplementary data and mixed baseline characteristics such as sex, age, clinical history and employment status. A single chart may not accommodate such varied information.
Components and size
A simple formal table has four components.
- A heading with a number and an appropriate legend or label.
- A top row containing column headings.
- A first column containing row headings.
- A data field.
The left-hand row labels are also called the stub. Tables should complement the narrative rather than duplicate it, remain understandable independently, and use consistent formats, units, symbols and headings throughout the document.
Six characteristics describe a good table.
| Characteristic | Practical meaning |
|---|---|
| Not too long or wide | Ideally fits on one journal or book page; tables extending over several pages can be hard to place and follow. |
| Clearly written title | Makes sense without referring back to the text. |
| Not too much text | Long paragraphs may belong in the narrative instead. |
| Not too many columns | Three to five are suggested as ideal; more may be necessary but can reduce readability. |
| Not too many abbreviations | Excessive abbreviation can compromise readability. |
| Not too many footnotes | Avoid continual movement between the table and its notes. |
Main-text tables should contain enough data to communicate the key message, with detailed supporting tabulations in an appendix or suitable electronic supplement. An overloaded table can be split into simpler ones. A teaching table may include raw data and intermediate calculations to explain how a statistic is derived, whereas a results table may need only the summary and a reference to the raw data. Too many tables can also complicate page layout.
Page and font limits can reveal an overloaded design. The 2012 layout advice was to stay within 1-inch (2.5-cm) margins, use landscape orientation for a wide table and split it if that still did not provide enough room. Readable type is more important than squeezing in data: the same advice suggested avoiding sizes below 9 point for most documents and offered Arial 10 point as a compact option.
Headings, alignment and rules
Seven construction tips support readable medical tables.
- Make the table's intended message clear at first glance.
- Keep column headings concise.
- Put measurement units in headings.
- Make the stub readable; a column of normal values beside it may help readers unfamiliar with the data range.
- Prefer round or reasonably short numbers; tables generally do not need four decimal places.
- Align decimal points vertically.
- Use similar titles and headings consistently throughout the article or chapter.
Every column should have a heading in the top row or rows. One layout convention uses bold headings with initial capitals and rules above and below them. Related rows should be grouped consistently, with enough white space to distinguish columns.
For readers of left-to-right scripts, familiar context can occupy the left columns and new information the columns to the right. For example, a DNA-yield table puts the starting amount of leaf tissue first and the quantity of DNA isolated to its right.
| Starting tissue (mg) | Arabidopsis DNA (µg), mean ± SD | Maize DNA (µg), mean ± SD |
|---|---|---|
| 40 | 8.33 ± 1.15 | 7.67 ± 0.58 |
| 100 | 43.33 ± 2.52 | 40.67 ± 6.66 |
| 150 | 49.83 ± 1.04 | 47.17 ± 2.75 |
| 250 | 59.67 ± 7.57 | 60.00 ± 4.00 |
Alignment connects values with their headings even when grid lines are removed. Rules that do not separate information readers might otherwise confuse add clutter and can be removed. Keeping vertical rules to a minimum allows uninterrupted reading across a row. A process table can legitimately need a grid because its cells form a matrix. The 2012 manuscript advice generally excluded vertical rules, limited horizontal rules to those above and below column headings and at the table's foot, and advised blank rows rather than shading to separate groups where manuscript conventions exclude shading.
Self-contained results displays
Titles and notes should be descriptive without becoming wordy, since some readers use tables independently of the narrative. Information accompanying a results display should explain the following.
- Abbreviations, even when expanded in the narrative.
- Sample summaries, such as mean and standard deviation for symmetrical Gaussian data and median and range or interquartile range for skewed data.
- Analysis populations, including intent-to-treat and per-protocol, and methods for handling attrition and multiple comparisons.
- P values and the statistical tests generating them.
- Analysis of covariance models and prespecified covariates associated with least-squares means.
- The reference group or condition assigned a risk ratio of 1.0 in relative-risk comparisons.
Concise column headings may use abbreviations if notes expand them; for example, “Histopath. Exam.” can be expanded below the table. Notes can explain an empty cell or why rounded percentages total 100.8%, using the destination journal's symbols and sequence.
A note can explain that denominators vary across characteristics because they reflect the number of participants with available data for each.
Eight design checks help expose common problems.
- Choose a design that makes comparisons easy and puts key comparators next to each other.
- Check that column headings accurately describe their fields.
- Check that a total at the bottom of a column is its sum.
- Align text and numbers consistently and top-align column headings.
- Remove clutter; consider a footnote for identical adjacent data and exponents where appropriate.
- Use an unambiguous footnote system, preferably letters if permitted; avoid numbered footnotes in numerical tables.
- Remove footnote ghosts: every symbol and abbreviation explained below the table must appear in it.
- Avoid empty cells that could confuse readers or layout editors.
Row and column comparisons
Placing treatment groups in row headings creates vertical comparisons within a column; placing them in column headings creates horizontal comparisons across a row. The same values can demonstrate both arrangements: an illustrative table reports surgical simulation scores from zero (poor) to 100 (excellent) for 35 colorectal surgeons.
With groups in rows, scores are compared down each task column.
| Test group | Cutting, median (range) | Suturing, median (range) | Ablating, median (range) |
|---|---|---|---|
| Residents (n = 12) | 75 (59–84) | 63 (57–85) | 80 (66–93) |
| Fellows (n = 8) | 85 (80–94) | 87 (78–99) | 88 (77–93) |
| Staff (n = 15) | 91 (75–99) | 92 (69–94) | 93 (88–99) |
| P | 0.03 | 0.04 | 0.05 |
With groups in columns, the same scores are compared across each task row.
| Surgical task | Residents (n = 12), median (range) | Fellows (n = 8), median (range) | Staff (n = 15), median (range) | P |
|---|---|---|---|---|
| Cutting | 75 (59–84) | 85 (80–94) | 91 (75–99) | 0.03 |
| Suturing | 63 (57–85) | 87 (78–99) | 92 (69–94) | 0.04 |
| Ablating | 80 (66–93) | 88 (77–93) | 93 (88–99) | 0.05 |
Proximity, emphasis and page fit
Vertical comparisons put numerical values close together with fewer intervening elements, in an arrangement familiar from arithmetic. Horizontal comparisons can place the values farther apart, sometimes with other numbers between them. Vertical arrangements can also save space when long group labels and sample sizes would otherwise require wide, multiline column headings.
In a schematic survival example, an arrangement with vertical comparisons used seven lines of type and five columns, whereas horizontal comparisons used eight lines and six columns. The example covered medians and interquartile ranges at 6, 12, 24 and 36 months after discharge, with placebo (n = 47), usual care (n = 52), active drug (n = 49), men (n = 71) and women (n = 77).
Column headings are prominent, usually bold and read before row headings by readers following English reading order, which can emphasise a comparison between groups placed across the top. Clinical studies often have more dependent than independent variables, which can also favour putting groups in columns for spatial reasons.
Evidence and conventions
Evidence comparing the arrangements is limited and conflicting. Studies found faster, more accurate currency conversions with vertical tables, faster searches for a single item in vertical lists, and faster same-or-different judgments for pairs of numbers placed one above the other. A different experiment found mean completion times of 73 seconds for vertical arrays and 44 seconds for horizontal arrays when participants marked nonidentical pairs (P = .005), while another found negligible orientation effects for alphanumeric displays. Data-display specialists Wainer, Ehrenberg and Clark have advocated vertical comparisons.
A 2018 review of 109 writing and data-display books found the following distribution.
| Treatment of comparisons | Books |
|---|---|
| Specifically recommended vertical comparisons | 21 |
| Used examples favouring vertical comparisons without specifically recommending them | 11 |
| Used both arrangements or considered each appropriate | 8 |
| Specifically recommended horizontal comparisons | 13 |
| Did not address the issue | 56 |
The books came from the reviewer's personal library; the collection lacked some recent, expensive titles and its representativeness was unknown. Research may also be dispersed across perceptual psychology, linguistics, human factors and graphic design, limiting conclusions about the absence of evidence.
The manual editions compared in 2018 gave different advice.
| Manual | Advice on comparison direction |
|---|---|
| AMA Manual of Style, tenth edition | Generally horizontal, but independent variables in the stub and dependent variables in columns for tables of those variables. |
| Council of Science Editors, Scientific Style and Format | Numbers usually easier to compare side by side, but no simple answer to which data belong in rows and columns. |
| Chicago Manual of Style | Vertical comparisons. |
| US Government Printing Office Style Manual | Examples consistent with vertical comparisons. |
| Publication Manual of the American Psychological Association | Horizontal comparisons. |
| American Chemical Society Style Guide | Did not address the issue. |
| European Association of Science Editors handbook | Did not address the issue. |
Historical recommendations also differed: books from 1922 and 1925 advocated vertical comparisons, whereas a 1936 text recommended horizontal comparisons for adjacent groups, noted their better fit to the page and used both arrangements. Day, Huth and Morgan recommended vertical comparisons in their early textbooks; Huth's 1987 and 1999 books retained vertical examples without explicitly recommending that orientation. O'Connor and Woodford's late-1970s examples supported vertical comparisons, while O'Connor's 1986 book recommended horizontal comparisons. Other horizontal recommendations appeared in Burnett's 1990 text and Lang's books of 1997, 2006 and 2010; Hewitt's and Davidson's 1957 books used both orientations without choosing between them.
Choice for the intended reader
No generally accepted rule assigns the independent variable to one axis: meaning, numbers of variables and entries, heading length and page fit can determine the choice. In many small medical tables, such as those occupying less than half a page, values can be compared quickly in either orientation. Vertical arrangements may suit large tables involving frequent numerical comparisons and audiences familiar with their conventions, such as basic scientists. Horizontal arrangements may suit smaller tables and mixed clinical readerships, including physicians, nurses, patients, lawyers, reporters and the public. Prominent column headings may lead readers to expect comparisons across groups; vertical comparisons may need attention directed to the row headings. The difference does not appear to impose a substantial burden on readers, and fitting the page may be a stronger reason for choosing an orientation.
Anticipated comparisons can be emphasised in either arrangement by highlighting relevant cells, perhaps with a border: these may be threshold values, maxima, minima, larger or smaller values, or groups of similar values. One worked display for 462 preterm neonates with congenital heart defects highlighted the most common risk category, surgery and diagnoses, plus the highest and lowest mortality and median hospital stay.
Captions and tables of tables
A table's title, body and legend should remain close together, with a short informative title and a legend explaining key information, abbreviations and footnotes. When captions generate a table of tables, reviewers use that list to locate the information they need. Putting every explanation in the caption can make the list hard to scan even when each individual table is self-contained.
Two caption rules balance these needs.
- Restrict each caption and corresponding table-of-tables entry to the elements needed to distinguish it from other entries.
- Put other information needed to explain the table in a subtitle or headnote excluded from the table of tables.
In a worked example, eleven captions beginning “Descriptive statistics of quantitative laboratory parameter” and continuing with parameter, units, treatment, time point and analysis set were shortened as follows.
| Table | Shortened caption | Page |
|---|---|---|
| 1B | Laboratory: Leukocytes | 4 |
| 2B | Laboratory: Erythrocytes | 5 |
| 3B | Laboratory: Hemoglobin | 6 |
| 4B | Laboratory: Hematocrit | 7 |
| 5B | Laboratory: MCVolume | 8 |
| 6B | Laboratory: MCHemoglobin | 9 |
| 7B | Laboratory: Platelets | 10 |
| 8B | Laboratory: Neutrophils, total [%] | 11 |
| 9B | Laboratory: Neutrophils, total [10⁹/L] | 12 |
| 10B | Laboratory: Lymphocytes [%] | 13 |
| 11B | Laboratory: Lymphocytes [10⁹/L] | 14 |
For the leukocyte table, the caption became “Laboratory: Leukocytes”, with measurement units and analysis set moved to a headnote. Units may still belong in captions when they distinguish relative from absolute counts; naming those approaches in words is another option. These rules require judgment and apply to other displays, including figures and listings.
Caption placement is a convention rather than a rule of English. Many publications place table titles above tables, but placing them below can also work. Publications should follow the target journal, regulatory documents the applicable house style, and each document should be internally consistent. Different reports within a dossier may use different conventions, while the Common Technical Document text should remain consistent.
Submission and production
Tables should follow the target journal's specific preparation requirements and be numbered in citation order. The 2018 textbook examples described JAMA and Springer instructions to create actual tables using the word processor's table function, rather than images of tables, with JAMA tables placed at the end of the manuscript.
Tables need careful proofreading after production because columns can become jumbled and editors may not know the medical meaning of their contents. Complex, data-heavy tables are especially vulnerable when authors focus on prose and skim the tables. Borrowed tables should identify the source in a note using the journal's style.