How to Read a Statistical Results Table

Overview

A statistical results table is a compressed argument. Its rows, columns, denominators, effect measures and footnotes determine what can be concluded.

The safest approach is not to begin with the p-value. Begin by asking:

1. Who is represented?
2. What outcome is being summarised?
3. Which groups or predictors are compared?
4. What is the effect measure?
5. What is the reference category and direction?
6. How precise is the estimate?
7. Is the estimate crude or adjusted?
8. What observations are missing or excluded?

Statistical reporting guidance recommends presenting effect estimates with appropriate uncertainty rather than relying only on hypothesis tests [1].

Why this matters

A table can look authoritative while concealing important problems:

• different denominators across rows;
• percentages calculated in the wrong direction;
• odds ratios described as risks;
• adjusted estimates without the adjustment set;
• confidence intervals omitted;
• reference categories hidden in footnotes;
• repeated observations treated as independent;
• statistically significant but clinically trivial effects.

Reading the table systematically prevents the eye from being drawn only to bold type or asterisks.

When to use this framework

Use the framework when reading:

• baseline tables;
• outcome tables;
• regression tables;
• diagnostic-accuracy tables;
• survival tables;
• meta-analysis summary tables;
• supplementary analyses.

The framework does not replace knowledge of the study design or statistical method. It helps readers connect the table with those elements.

Anatomy of a statistical table

Simulated primary and secondary outcomes

simulated data

Table 1. Twelve-week outcomes by randomised group

Outcome	Telemonitoring	Usual care	Effect estimate
Systolic blood pressure, mean (SD), mmHg	132.4 (11.9), n=114	138.1 (12.3), n=112	Adjusted mean difference −5.9 mmHg (95% CI −8.5 to −3.3); $p<0.001$
Blood-pressure control, n/N (%)	82/120 (68.3%)	65/120 (54.2%)	Risk difference 14.2 percentage points (95% CI 1.8 to 25.9); risk ratio 1.26 (95% CI 1.03 to 1.55); $p=0.024$

Notes: The numerical-outcome effect is telemonitoring minus usual care, adjusted for baseline systolic blood pressure and clinic. Negative mean differences favour telemonitoring. The binary-outcome analysis follows randomised allocation and uses all 240 participants. The calculations underlying the simulated adjusted estimates, confidence intervals and p-values are not shown here. CI = confidence interval; SD = standard deviation.

Title

A useful title states:

• outcome or table purpose;
• relevant time point;
• analysis population where necessary.

A title such as “Results” is too vague.

Rows

Rows may represent:

• outcomes;
• predictors;
• subgroup levels;
• time points;
• models.

Do not assume all rows use the same denominator or effect measure.

Columns

Columns may present:

• group summaries;
• crude estimates;
• adjusted estimates;
• confidence intervals;
• p-values;
• missing counts.

A table should not force the reader to infer which estimate belongs to which model.

Footnotes

Footnotes should define all non-standard abbreviations and clarify:

• reference groups;
• direction of subtraction;
• adjustment variables;
• statistical method;
• missing-value handling;
• multiplicity correction.

A step-by-step reading method

Step 1: Identify the analytical population

Ask whether the table represents:

• all enrolled participants;
• all randomised participants;
• complete records;
• per-protocol participants;
• participants with a measured outcome;
• events or person-time rather than participants.

In the example, descriptive week-12 means use 114 and 112 observed outcomes, whereas the binary endpoint uses all 120 participants per group. The table must explain why.

Step 2: Identify the variable type

For numerical variables, look for:

• mean and SD;
• median and quartiles;
• units;
• number observed.

For categorical variables, look for:

• count and percentage;
• denominator;
• whether percentages are calculated by row, column or total.

Step 3: Identify the effect measure

Differences have null value 0:

• mean difference;
• risk difference;
• regression coefficient.

Ratios have null value 1:

• risk ratio;
• odds ratio;
• hazard ratio;
• rate ratio.

The numerical value cannot be interpreted without knowing the scale.

Step 4: Determine direction and reference

The mean difference is telemonitoring minus usual care. Therefore, −5.9 mmHg means the telemonitoring mean is lower.

If the subtraction were reversed, the estimate would be +5.9 mmHg with the same substantive result.

For regression tables, the stated reference category determines every categorical coefficient.

Step 5: Read the confidence interval

The confidence interval shows the range of effect sizes compatible with the data and model under the selected procedure [2].

For the adjusted mean difference:

• estimate: −5.9 mmHg;
• 95% CI: −8.5 to −3.3.

The interval excludes the null value 0. It is also narrow enough to exclude very small effects near zero.

Step 6: Assess clinical importance

Suppose a reduction of about 5 mmHg was considered worthwhile during planning.

The interval includes effects larger and smaller than 5 mmHg. Therefore, the data support benefit but leave uncertainty about whether the average effect reaches the planning target.

The planning target should not be treated as an automatic post-study threshold.

Step 7: Read the p-value last

The p-value summarises incompatibility with the specified null model. It does not state:

• the probability the null is true;
• the probability the result occurred by chance;
• effect magnitude;
• clinical importance;
• absence of bias [3].

Step 8: Examine adjustment

An adjusted estimate is conditional on:

• variables included;
• coding;
• functional form;
• interactions;
• missing-data assumptions.

The table should state the adjustment set. “Adjusted model” without details is insufficient.

Worked example

Misleading table patterns

Asterisks without estimates

Asterisks convert continuous evidence into categories and encourage readers to ignore magnitude.

Separate within-group tests

A table showing significant change in one group and non-significant change in another does not test the group difference.

Percentages without counts

68.3% should be accompanied by 82/120.

Excessive decimal places

Reporting blood pressure as 132.437 mmHg implies unsupported precision.

Hidden model changes

Model 1 and Model 2 may contain different participants because of missing covariates. The sample size for each model should be shown.

Baseline significance tests

In randomised trials, baseline p-values are generally not needed. Baseline tables should describe the sample rather than test whether randomisation “worked”.

Interpreting the table

Presenting the results

Suggested reporting

Tables should be understandable without repeated consultation of the main text. ICMJE guidance emphasises clear titles, concise headings, explicit variation measures and explanatory footnotes [4].

Common mistakes

Knowledge check

What should be read before the numerical cells?

Answer and reasoning

The table title, row and column headings, denominator information and footnotes.

What is the null value for a risk ratio?

Answer and reasoning

One. A ratio of one indicates equal risks.

Why is the direction of the mean difference important?

Answer and reasoning

The sign depends on which group is subtracted. Without the direction, −5.9 mmHg cannot be interpreted.

Does p = 0.024 show the effect is clinically important?

Answer and reasoning

No. Clinical importance is evaluated using the risk difference, interval, consequences and decision context.

What must accompany an adjusted estimate?

Answer and reasoning

The variables and functional forms used for adjustment, the analysis population and the effect measure.

Key take-home messages

References

1. Lang TA, Altman DG. Basic statistical reporting for articles published in biomedical journals: the SAMPL guidelines. Int J Nurs Stud. 2015;52(1):5–9. https://doi.org/10.1016/j.ijnurstu.2014.09.006
2. Gardner MJ, Altman DG. Confidence intervals rather than P values: estimation rather than hypothesis testing. Br Med J (Clin Res Ed). 1986;292(6522):746–750. https://doi.org/10.1136/bmj.292.6522.746
3. Greenland S, Senn SJ, Rothman KJ, Carlin JB, Poole C, Goodman SN, et al. Statistical tests, P values, confidence intervals, and power: a guide to misinterpretations. Eur J Epidemiol. 2016;31(4):337–350. https://doi.org/10.1007/s10654-016-0149-3
4. International Committee of Medical Journal Editors. Recommendations for the conduct, reporting, editing, and publication of scholarly work in medical journals: preparing a manuscript for submission. Updated 2026. https://www.icmje.org/recommendations/browse/manuscript-preparation/preparing-for-submission.html