To interpret a correlation table, identify the variable pair, read its coefficient, check the usable sample size and any inference, then explain the relationship within your study's limits. For Pearson correlation, the coefficient describes linear association. A p-value alone does not tell you the relationship's size, importance or cause.
You have already done the work of analysing your data; the next task is making the report match it. UniDraft helps organise your confirmed methods and verified findings into an editable final-year project document. Start your project write-up, keep the original analysis beside the draft, and check every value and interpretation. UniDraft does not run statistical tests or supply results.
Start with the table heading and method
Before reading a number, record what the output actually contains: Pearson's r, Spearman's rank correlation or another measure. Do not call a Spearman coefficient Pearson's r merely because both appear in similar-looking tables.
Check the variable labels against your codebook. A column called score_total is not enough: what instrument, scoring direction and units does it represent? A reverse-coded score can change the meaning of a positive or negative association. Use the data-cleaning guide where coding or missing-value handling is unresolved.
The Kent State University Pearson tutorial explains the coefficient, direction, linear association and output. Its software walkthrough is a reference, not evidence that your own analysis is appropriate. Use the statistical-test decision guide and methods advice before interpreting an unsuitable test.
Read r, p and N for the same pair
| Output item | What to check | What it does not tell you |
|---|---|---|
| Coefficient, such as Pearson r | Direction and magnitude of the sample linear association | Whether one variable causes the other |
| Significance, often labelled Sig. | Test, null hypothesis, one- or two-sided setting and reported p-value | The probability that your hypothesis is true |
| N | Usable paired observations for this relationship | Automatically the number invited or the sample for every other pair |
| Confidence interval, if supplied | Method, confidence level and bounds for the association | A guarantee that the estimate is practically important |
A coefficient lies between −1 and +1. For Pearson r, a positive sign indicates that higher values tend to occur together; a negative sign indicates an inverse linear direction. A value near zero can conceal a curved relationship, so inspect the scatterplot rather than write “no relationship of any kind.”
Do not use a universal weak/moderate/strong label without a justified convention and context. The same numerical association can have different practical meaning across measurements and questions. Report the coefficient itself, even when your department also requires a verbal label.
A fictional matrix: count each relationship once
Original teaching illustration only. These selected coefficients describe how to read a matrix; they are not collected student data, executed software output or a significance analysis. No p-values or confidence intervals have been calculated for this illustration.
| Pearson r | Practice score A | Task score B | Completion time C |
|---|---|---|---|
| Practice score A | 1.00 | 0.40 | −0.30 |
| Task score B | 0.40 | 1.00 | −0.20 |
| Completion time C | −0.30 | −0.20 | 1.00 |
Read it in this order:
- Find A's row and B's column: r = 0.40.
- Find B's row and A's column: it is the same relationship, repeated for readability.
- Ignore the diagonal as a substantive comparison: A correlated with itself is not evidence for your research question.
- Read A with C: r = −0.30 indicates an inverse linear association in the illustration. The minus sign does not mean the association failed.
- Return to the research question. Three variables produce three distinct pairs, but your report should explain why each reported pair matters.
Do not turn the six off-diagonal cells into six independent discoveries. If you examined many relationships, discuss the analysis plan and any multiple-testing issue with your adviser instead of selecting only attractive results.
Why can N change between pairs?
Suppose, in a separate fictional missing-data planning example, 60 records contain A and B, but only 54 contain A and C. Pairwise handling may use different sets of cases for these two correlations. Listwise handling instead restricts the analysis to cases complete for the selected variables. The actual procedure and software settings determine the output.
Record your real missing-data method and the N for each reported pair. Do not copy the questionnaire return total into every correlation row, replace blanks with zero without justification, or compare coefficients as though they necessarily came from identical participants. Explain relevant exclusions in the methods and results. Different missing-data mechanisms can affect interpretation; changing the deletion setting does not automatically resolve bias.
Check the inference before writing “significant”
Read the test settings and planned significance level. If software rounds a p-value to .000, do not report a literal probability of zero; use the reporting precision supported by the output and your departmental convention, commonly p < .001 where justified. Preserve an exact value when appropriately available.
The American Statistical Association's statement on p-values explains why a threshold alone is insufficient for a scientific conclusion. A small p-value does not measure the importance of an effect, and a larger one does not establish that the population association is exactly zero.
Check independence, sampling and the relevant assumptions with a methods adviser. Repeated observations from the same person or students grouped within classes can require an approach beyond a simple independent-case correlation test. A visually compelling number does not repair the design.
Turn your verified output into a result paragraph
Use this blank reporting worksheet before drafting. Fill it from your records, not from the teaching matrix above.
| Record | Your verified entry |
|---|---|
| Question or planned hypothesis | The relationship being investigated |
| Variables and scoring direction | What each measure means; units or scale |
| Method | Pearson, Spearman or the actual procedure used |
| Usable N and missing-data handling | Pair-specific cases and documented exclusions |
| Coefficient | Sign and numerical value from the correct cell |
| Inference where appropriate | Actual p-value, test setting and/or confidence interval with method |
| Diagnostic or design issue | Outliers, relationship shape, dependence or sampling limits |
| Bounded answer | What the evidence supports, without a causal leap |
An adaptable sentence frame, not a completed result, is:
Using [method], [measure A] and [measure B] had a [direction] association in [usable N] observations ([coefficient and verified inference]). [State a relevant design or interpretation limit].
Do not fill the gaps with plausible numbers. If uncertainty information is required but missing, return to the analysis with appropriate advice. Write the result under its research question, and use the Chapter 4 guide for table captions and the boundary between findings and wider discussion.
Replace unsupported interpretations
| Draft wording | Better action |
|---|---|
| “Practice caused better performance because r is positive” | Report the association; assess causal claims against the design |
| “The relationship is 40% because r = .40” | Report r = .40; it is not a percentage of participants or a causal percentage |
| “p is small, so the relationship is strong” | Interpret magnitude from the coefficient and context; report inference separately |
| “No relationship exists because the result is not significant” | Describe the estimate and uncertainty; do not turn lack of evidence into proof of absence |
| “All respondents were included” | Check the actual N and missing-data settings for this pair |
In simple settings r squared has a specific mathematical interpretation, but do not introduce it as a causal explanation or silently substitute it for your planned analysis. If regression or adjustment for other variables is required, confirm that method separately.
Before you submit
- The method, variable names and scoring direction match Chapter 3 and the codebook.
- Every coefficient, N and inference belongs to the same intended variable pair.
- Mirrored cells and diagonal entries have not been counted as extra findings.
- Scatterplots and relevant diagnostics have been examined, with unresolved issues referred for advice.
- Missing-data handling and departures from the plan are documented honestly.
- The abstract, results and conclusion quote consistent verified values.
- The report explains association without inventing causation or unsupported practical importance.
Once those checks are complete, use UniDraft to prepare the written document around your confirmed evidence, then compare the editable output with your original analysis before submission. Start your project document.
Sources and scope
- Kent State University Libraries: Pearson Correlation, checked 25 October 2026: coefficient, matrix output and limits of linear association. No source dataset or screenshots reproduced here.
- American Statistical Association: Statement on Statistical Significance and P-Values: interpretation principles, not a substitute for advice on your study design.
This guide addresses reading and reporting checked correlation output. It does not prescribe a statistical method for every project or override your supervisor's confirmed instructions.