Turn a dense results table into a careful evidence statement without losing denominators, units or uncertainty
A research results table compresses many decisions into rows, columns, symbols and footnotes. English matters because the reader must rebuild a sentence: who or what was measured, compared with what, on which outcome, in which unit and time frame, with how much uncertainty and under which analysis.
Choose the route closest to your task, or read straight through for the complete system.
Searches for how to read a research table, p value table, confidence interval and interpreting results often treat tables as containers of answers. A table is better understood as an interface between methods and claims. Its title, population, row labels, column headings, units, reference categories and notes determine what each number means.
Readers should move in a fixed order. First identify the table’s question. Then find the analysed sample and denominators. Next read one row horizontally and one column vertically. Only after the comparison is reconstructed should the reader interpret effect size, uncertainty and model adjustments. This order prevents decorative significance markers from controlling the conclusion.
Numbers also need language. A relative reduction can sound dramatic when the absolute event difference is small. A mean can hide a skewed distribution. An adjusted estimate can differ from a raw comparison. A subgroup result can be unstable when many comparisons were explored. Clear English names the measure and boundary before giving the conclusion.
This article is educational, not statistical, clinical or policy advice. Interpretation depends on study design, analysis plan and domain knowledge. Read the methods, table notes, supplement and full paper before using a result in a consequential decision.
Did You Know? Reporting guidelines such as CONSORT ask authors to present results with estimates and precision rather than relying only on significance labels. A table can display a small p-value while the effect is too small to matter, or show an important effect with wide uncertainty. Reading therefore begins with the estimate, denominator and interval—not with a hunt for stars.
Find the section you need
Orient the table · 2 chapters
Decode structure · 2 chapters
Interpret estimates · 2 chapters
Check complications · 2 chapters
The title should identify the outcome, population, comparison or analysis represented.
Rewrite it as a question and confirm the table actually contains the variables needed to answer it.
“Outcomes by treatment group at eight weeks” does not report long-term follow-up or baseline change unless columns say so.
What to check
- Name outcome
- Name groups
- Find time point
- Identify analysis type
Create a four-column evidence record for chapter 1, “Read the title as a compressed question”. Copy the exact source wording or observation, explain what it means in context, name the decision it affects, and state the boundary that prevents a larger conclusion.
Now explain the same point to a reader who has not seen the source. Preserve the governing noun, verb, population, comparison, time frame and uncertainty. If the shorter explanation changes the research decision, restore the missing detail.
Boundary: A short title may require the caption and methods for full meaning.
Learning transfer: Titles guide disciplined reading in every data document.
Write one sentence that states the evidence, one that states its limit, and one that states the next responsible action. Then test each noun, number, condition and source against the controlled document. Replace vague confidence with a traceable reason.
Percentages and rates are uninterpretable without the number and population behind them.
Locate total N, group n, analysed n, missing observations and any changing denominator across rows.
Thirty per cent could mean 3 of 10 or 300 of 1,000; the uncertainty and practical meaning differ.
What to check
- Find total N
- Check row denominators
- Track exclusions
- Look for weighted counts
Create a four-column evidence record for chapter 2, “Find the analysed sample and denominators”. Copy the exact source wording or observation, explain what it means in context, name the decision it affects, and state the boundary that prevents a larger conclusion.
Now explain the same point to a reader who has not seen the source. Preserve the governing noun, verb, population, comparison, time frame and uncertainty. If the shorter explanation changes the research decision, restore the missing detail.
Boundary: The enrolled sample and analysed sample may not be identical.
Learning transfer: Denominator literacy protects everyday numerical claims.
Write one sentence that states the evidence, one that states its limit, and one that states the next responsible action. Then test each noun, number, condition and source against the controlled document. Replace vague confidence with a traceable reason.
Tables often nest time, group and statistic under several header levels.
Trace a single cell upward to every header and leftward to every row label before interpreting it.
A value may belong to the intervention group, follow-up period and adjusted model, not to the entire sample.
What to check
- Follow header hierarchy
- Check spanning labels
- Read stub column
- Mark model number
Create a four-column evidence record for chapter 3, “Map rows, columns and multi-level headers”. Copy the exact source wording or observation, explain what it means in context, name the decision it affects, and state the boundary that prevents a larger conclusion.
Now explain the same point to a reader who has not seen the source. Preserve the governing noun, verb, population, comparison, time frame and uncertainty. If the shorter explanation changes the research decision, restore the missing detail.
Boundary: Do not assume repeated columns use the same unit or population.
Learning transfer: Spatial reading complements sentence comprehension.
Write one sentence that states the evidence, one that states its limit, and one that states the next responsible action. Then test each noun, number, condition and source against the controlled document. Replace vague confidence with a traceable reason.
The same number changes meaning with percentage, proportion, score, rate, odds or transformed unit.
Write the unit beside each key estimate and open the methods for scale direction and range.
A lower score can indicate improvement on one scale and worsening on another.
What to check
- Copy units
- Check scale direction
- Find range
- Identify transformations
Create a four-column evidence record for chapter 4, “Check units, scales and coding”. Copy the exact source wording or observation, explain what it means in context, name the decision it affects, and state the boundary that prevents a larger conclusion.
Now explain the same point to a reader who has not seen the source. Preserve the governing noun, verb, population, comparison, time frame and uncertainty. If the shorter explanation changes the research decision, restore the missing detail.
Boundary: Standardised scores can aid comparison but are not automatically intuitive in practice.
Learning transfer: Unit discipline links mathematics and English.
Write one sentence that states the evidence, one that states its limit, and one that states the next responsible action. Then test each noun, number, condition and source against the controlled document. Replace vague confidence with a traceable reason.
5. Distinguish descriptive and inferential statistics
Back to contentsCounts, means and medians describe the observed sample; effect estimates and intervals support model-based inference.
Label each column as descriptive, unadjusted comparison, adjusted comparison or model diagnostic.
A baseline mean is not evidence of an intervention effect; the relevant estimate may be a change difference or model coefficient.
What to check
- Classify statistic
- Find comparison
- Identify reference
- Check model label
Create a four-column evidence record for chapter 5, “Distinguish descriptive and inferential statistics”. Copy the exact source wording or observation, explain what it means in context, name the decision it affects, and state the boundary that prevents a larger conclusion.
Now explain the same point to a reader who has not seen the source. Preserve the governing noun, verb, population, comparison, time frame and uncertainty. If the shorter explanation changes the research decision, restore the missing detail.
Boundary: Summary statistics can hide distributions and outliers; inspect figures or supplements.
Learning transfer: Readers learn not to make one number do every job.
Write one sentence that states the evidence, one that states its limit, and one that states the next responsible action. Then test each noun, number, condition and source against the controlled document. Replace vague confidence with a traceable reason.
Relative measures and absolute differences answer different practical questions.
Recover group risks or rates where possible, then state both absolute and relative comparisons with time frame.
A 50% relative reduction from 2 in 1,000 to 1 in 1,000 is an absolute reduction of 1 in 1,000.
What to check
- Find baseline risk
- State time
- Calculate carefully
- Avoid percentage-point confusion
Create a four-column evidence record for chapter 6, “Translate relative and absolute effects”. Copy the exact source wording or observation, explain what it means in context, name the decision it affects, and state the boundary that prevents a larger conclusion.
Now explain the same point to a reader who has not seen the source. Preserve the governing noun, verb, population, comparison, time frame and uncertainty. If the shorter explanation changes the research decision, restore the missing detail.
Boundary: Derived values should be labelled and based on compatible denominators.
Learning transfer: This skill improves public-information reading.
Write one sentence that states the evidence, one that states its limit, and one that states the next responsible action. Then test each noun, number, condition and source against the controlled document. Replace vague confidence with a traceable reason.
7. Read confidence intervals as ranges of compatible estimates
Back to contentsAn interval expresses estimate precision under model assumptions; it is not a guarantee that the true value lies there.
Identify the effect scale, null value, interval width and clinically or practically meaningful range.
A risk ratio interval from 0.60 to 1.10 includes substantial benefit, little effect and slight harm; a single point estimate hides that uncertainty.
What to check
- Know null value
- Check width
- Compare meaningful threshold
- Keep assumptions
Create a four-column evidence record for chapter 7, “Read confidence intervals as ranges of compatible estimates”. Copy the exact source wording or observation, explain what it means in context, name the decision it affects, and state the boundary that prevents a larger conclusion.
Now explain the same point to a reader who has not seen the source. Preserve the governing noun, verb, population, comparison, time frame and uncertainty. If the shorter explanation changes the research decision, restore the missing detail.
Boundary: Different interval methods and repeated analyses affect interpretation.
Learning transfer: Calibrated uncertainty improves decisions.
Write one sentence that states the evidence, one that states its limit, and one that states the next responsible action. Then test each noun, number, condition and source against the controlled document. Replace vague confidence with a traceable reason.
A p-value addresses compatibility with a statistical model and null hypothesis, not importance or truth.
Read the estimate, interval, sample size, design and multiplicity before using the p-value.
A huge sample can detect a tiny difference that has little practical consequence.
What to check
- State effect size
- Read interval
- Check hypothesis
- Avoid significant equals important
Create a four-column evidence record for chapter 8, “Do not let p-values replace the effect”. Copy the exact source wording or observation, explain what it means in context, name the decision it affects, and state the boundary that prevents a larger conclusion.
Now explain the same point to a reader who has not seen the source. Preserve the governing noun, verb, population, comparison, time frame and uncertainty. If the shorter explanation changes the research decision, restore the missing detail.
Boundary: A non-significant result is not proof of no effect, especially with imprecision.
Learning transfer: Precise statistical language prevents binary thinking.
Write one sentence that states the evidence, one that states its limit, and one that states the next responsible action. Then test each noun, number, condition and source against the controlled document. Replace vague confidence with a traceable reason.
9. Understand adjusted models and reference categories
Back to contentsAdjusted estimates depend on chosen covariates, coding and model form.
Read the footnote or methods for covariates, reference level, interactions and analysis population.
An odds ratio for “high exposure” is meaningless until the reader knows the reference exposure and adjustment set.
What to check
- Find reference category
- List adjustments
- Check coding
- Compare model versions
Create a four-column evidence record for chapter 9, “Understand adjusted models and reference categories”. Copy the exact source wording or observation, explain what it means in context, name the decision it affects, and state the boundary that prevents a larger conclusion.
Now explain the same point to a reader who has not seen the source. Preserve the governing noun, verb, population, comparison, time frame and uncertainty. If the shorter explanation changes the research decision, restore the missing detail.
Boundary: Adjustment may reduce some confounding but can introduce bias when variables are chosen poorly.
Learning transfer: Conditional statements clarify complex comparisons.
Write one sentence that states the evidence, one that states its limit, and one that states the next responsible action. Then test each noun, number, condition and source against the controlled document. Replace vague confidence with a traceable reason.
10. Inspect missing data, subgroups and multiple comparisons
Back to contentsSparse cells and repeated testing can make striking results unstable.
Look for missing counts, imputation, subgroup sample sizes, prespecification, interaction tests and multiplicity control.
One subgroup may show benefit by chance when many subgroups are examined; separate within-group significance is not proof that groups differ.
What to check
- Count subgroup n
- Find interaction test
- Check prespecification
- Read missing-data note
Create a four-column evidence record for chapter 10, “Inspect missing data, subgroups and multiple comparisons”. Copy the exact source wording or observation, explain what it means in context, name the decision it affects, and state the boundary that prevents a larger conclusion.
Now explain the same point to a reader who has not seen the source. Preserve the governing noun, verb, population, comparison, time frame and uncertainty. If the shorter explanation changes the research decision, restore the missing detail.
Boundary: Exploratory findings can be valuable when labelled and followed up, not presented as settled.
Learning transfer: Exception handling strengthens analytical writing.
Write one sentence that states the evidence, one that states its limit, and one that states the next responsible action. Then test each noun, number, condition and source against the controlled document. Replace vague confidence with a traceable reason.
11. Reconcile table, text, figure and conclusion
Back to contentsA trustworthy reading checks whether different parts of the paper describe the same result.
Compare sample size, estimate, time point, model and wording across methods, table, figure, abstract and discussion.
The discussion may emphasise a secondary outcome while the primary-outcome table shows uncertainty; both belong in the summary.
What to check
- Match numbers
- Check primary outcome
- Compare verbs
- Look for correction
Create a four-column evidence record for chapter 11, “Reconcile table, text, figure and conclusion”. Copy the exact source wording or observation, explain what it means in context, name the decision it affects, and state the boundary that prevents a larger conclusion.
Now explain the same point to a reader who has not seen the source. Preserve the governing noun, verb, population, comparison, time frame and uncertainty. If the shorter explanation changes the research decision, restore the missing detail.
Boundary: Small rounding differences can be harmless; material inconsistencies need resolution.
Learning transfer: Cross-document reconciliation supports audits and study notes.
Write one sentence that states the evidence, one that states its limit, and one that states the next responsible action. Then test each noun, number, condition and source against the controlled document. Replace vague confidence with a traceable reason.
How to build this kind of English before the job title arrives
Back to contentsStudents do not need professional authority to practise the underlying language. They can learn to define scope, annotate conditions, compare versions, rebuild a timeline, match a number to its unit and explain uncertainty without embarrassment. Parents can ask calm questions: What decision is this document supporting? Which sentence carries the strongest claim? What would make that claim false? Which source would you open next?
Use a three-pass routine. First, map the document: title, owner, audience, date, sections and decision. Second, trace one claim from source to conclusion. Third, explain the limit and next action. This routine strengthens comprehension, science, mathematics, humanities and workplace readiness because it turns English into an evidence tool.
A student should never imitate professional authority. The useful goal is to recognise when a text needs a qualified reviewer, current rule or official decision. Knowing when to pause is part of literacy.
Cell trail
Outcome: assignment score at eight weeks. Intervention: mean 74.2, n=96. Comparison: mean 71.8, n=92. Adjusted mean difference 2.1 points, 95% confidence interval −0.4 to 4.6.
Responsible sentence
In the analysed sample at eight weeks, the intervention group’s adjusted mean score was estimated 2.1 points higher, with an interval compatible with a small disadvantage through a 4.6-point advantage.
What remains
The reader still needs scale range, baseline comparability, missingness, prespecified importance threshold and model details.
Overclaim to avoid
“The intervention definitely improved performance.” The interval and design do not support that certainty.
Population
Who or what was analysed?
Comparison
Which groups, exposures or times are contrasted?
Outcome
What was measured and on which scale?
Time
When was it measured?
Estimate
What magnitude and direction were observed?
Uncertainty
How precise is the estimate?
Boundary
Which design and analysis conditions govern use?
For each case, identify the tempting shortcut, the controlling evidence and the sentence that needs repair.
The percentage without n
“Thirty per cent” sounds stable until the reader discovers it means 3 of 10.
Write a short evidence note naming the source, applicable date, decision owner and boundary. Then state the smallest next action that remains justified if one assumption proves wrong.
The starred result
A tiny p-value distracts from a negligible effect. Magnitude and interval lead the summary.
Write a short evidence note naming the source, applicable date, decision owner and boundary. Then state the smallest next action that remains justified if one assumption proves wrong.
The reference trap
An odds ratio is copied without its reference category. The comparison is reconstructed.
Write a short evidence note naming the source, applicable date, decision owner and boundary. Then state the smallest next action that remains justified if one assumption proves wrong.
The subgroup headline
One subgroup looks positive, but many were tested and the interaction is unclear. The finding remains exploratory.
Write a short evidence note naming the source, applicable date, decision owner and boundary. Then state the smallest next action that remains justified if one assumption proves wrong.
A parent or tutor can ask for the source and reasoning, but the student should make the final record in their own words.
Use a live, current source where possible. Keep source wording separate from interpretation and preserve dates.
Trace five cells
Read every header, row label, unit and note.
Write a short evidence note naming the source, applicable date, decision owner and boundary. Then state the smallest next action that remains justified if one assumption proves wrong.
Recover denominators
Rewrite percentages as counts where data permit.
Write a short evidence note naming the source, applicable date, decision owner and boundary. Then state the smallest next action that remains justified if one assumption proves wrong.
Translate estimates
State direction, magnitude, time and interval in plain English.
Write a short evidence note naming the source, applicable date, decision owner and boundary. Then state the smallest next action that remains justified if one assumption proves wrong.
Compare models
Record how adjustment changes the estimate and population.
Write a short evidence note naming the source, applicable date, decision owner and boundary. Then state the smallest next action that remains justified if one assumption proves wrong.
Reconcile the paper
Match the table with methods, figure, abstract and discussion.
Write a short evidence note naming the source, applicable date, decision owner and boundary. Then state the smallest next action that remains justified if one assumption proves wrong.
Finish with three lines: what is supported, what remains uncertain and what the next research action is.
The evidence should remain stable even when tone, detail and next action change with the reader.
Student lens
Structure must be decoded before calculation.
Write a short evidence note naming the source, applicable date, decision owner and boundary. Then state the smallest next action that remains justified if one assumption proves wrong.
Researcher lens
Every cell should connect to the method.
Write a short evidence note naming the source, applicable date, decision owner and boundary. Then state the smallest next action that remains justified if one assumption proves wrong.
Statistician lens
Estimates need uncertainty and assumptions.
Write a short evidence note naming the source, applicable date, decision owner and boundary. Then state the smallest next action that remains justified if one assumption proves wrong.
Public lens
Plain language must preserve denominators and units.
Write a short evidence note naming the source, applicable date, decision owner and boundary. Then state the smallest next action that remains justified if one assumption proves wrong.
Decision-maker lens
Practical importance differs from statistical detection.
Write a short evidence note naming the source, applicable date, decision owner and boundary. Then state the smallest next action that remains justified if one assumption proves wrong.
Adaptation is useful only when the shortened version preserves design, population, comparison, uncertainty and boundary.
These checks slow the reader where a plausible summary can hide a consequential gap.
The model cascade
When several models are shown, track how the estimate, sample and adjustment set change. A stable number can reassure, but it is not proof that every bias is removed.
Write a short evidence note naming the source, applicable date, decision owner and boundary. Then state the smallest next action that remains justified if one assumption proves wrong.
The sparse-cell warning
Small subgroup counts can produce unstable estimates, wide intervals or disclosure concerns. Treat precise-looking decimals cautiously.
Write a short evidence note naming the source, applicable date, decision owner and boundary. Then state the smallest next action that remains justified if one assumption proves wrong.
The decision threshold
Ask whether the interval crosses a threshold that matters in practice, not only the statistical null. Name who defined the threshold.
Write a short evidence note naming the source, applicable date, decision owner and boundary. Then state the smallest next action that remains justified if one assumption proves wrong.
A strong final record shows not only the answer reached, but also the evidence boundary and the condition that would require revision.
Answer each question in a complete sentence tied to a source, date, observation or authorised clarification. Write “unknown” when the record does not support an answer.
- What question does the table answer? Record the evidence and the conclusion it permits.
- Which population is analysed? Record the evidence and the conclusion it permits.
- What is each denominator? Record the evidence and the conclusion it permits.
- Which headers govern the cell? Record the evidence and the conclusion it permits.
- What unit and scale apply? Record the evidence and the conclusion it permits.
- What is the reference category? Record the evidence and the conclusion it permits.
- Is the statistic descriptive or inferential? Record the evidence and the conclusion it permits.
- What is the absolute effect? Record the evidence and the conclusion it permits.
- What does the interval include? Record the evidence and the conclusion it permits.
- Which adjustments were made? Record the evidence and the conclusion it permits.
- How are missing data handled? Record the evidence and the conclusion it permits.
- Does the text match the table? Record the evidence and the conclusion it permits.
Reader reconstruction
Give the record to a careful reader who was not present. Ask them to identify the question, method, evidence, uncertainty and next action without extra explanation. If they cannot, repair the missing link rather than adding decorative formality.
Change test
Imagine that the database, sample, outcome definition, analysis or source page changes tomorrow. Mark every sentence that would need updating. This makes cycle-sensitive claims visible and prevents an old note from controlling a new decision.
Should I start with the p-value?
No. Start with the table question, sample, denominator, estimate, unit and uncertainty.
What does the reference category mean?
It is the category against which other estimates are expressed; its choice affects interpretation, not the underlying observations.
Why do denominators change across rows?
Missing responses, eligibility, branching questions or different analysis sets may change them. The notes should explain.
Is a narrow confidence interval always good?
It indicates greater precision under the model, but a precise estimate can still be biased or unimportant.
Can I compare two subgroup p-values?
Not directly. Evidence of a difference between subgroups normally requires an interaction or equivalent comparison.
What if the table and text disagree?
Check rounding, model, time point and analysis population, then consult corrections or contact authors if the discrepancy matters.
- NCBI Bookshelf — How to Read a Scientific Manuscript
- EQUATOR Network — CONSORT Reporting Guideline
- Why English? | Reading a Research Abstract
- Why English? | Explaining Scientific Evidence
Cycle-sensitive facts were checked against current official pages on 11 October 2026. Always reopen the authority’s live page before a real decision.