Effect Sizes in SPSS: Choose, Read and Report the Right Measure
A p-value addresses evidence against a null hypothesis. An effect size describes the magnitude of an estimated relationship or difference. Here is a practical route from the research question to a defensible report.

Quick answer: Match the effect-size measure to the analysis and design before reading the SPSS table. For a two-group mean comparison, consider a standardized mean difference such as Cohen’s d. For an ANOVA, report the effect-size measure supplied for the relevant factor and name it exactly; eta squared and partial eta squared are not interchangeable. State the estimate, an interval when available, units or definition, and the study context. Do not describe an effect as practically important from a universal “small/medium/large” cutoff alone.
Choose the measure before opening the output
The measure depends on what the study compares. IBM’s paired-samples t-test documentation describes an effect-size option, while its one-way ANOVA documentation describes an ANOVA Effect Sizes table. Check the procedure and software version actually used in your project; output labels differ by analysis.
| Research comparison | Common measure to examine | Read with care |
|---|---|---|
| Two independent groups on a continuous outcome | Mean difference in the original units and a standardized difference such as Cohen’s d | Check group order, spread and confidence interval. |
| Before/after measurements on the same participants | Paired mean difference and the paired-design effect-size estimate | Do not substitute an independent-groups formula. |
| One-way ANOVA across groups | The ANOVA effect-size estimate reported for the overall comparison | Name eta squared or another statistic exactly as shown; a post-hoc pairwise comparison is a different question. |
| Factor in a multi-factor ANOVA | Often partial eta squared for that factor | It is conditional on the model and should not be compared casually with a one-way eta squared. |
A statistically significant difference can be small in practical terms, while an imprecise interval can leave several plausible magnitudes. A peer-reviewed primer on effect sizes for t-tests and ANOVAs explains why the study design and estimator matter. Your programme or target journal may require a particular statistic and format; check that rule before finalizing the table.
Worked fictional example: training scores
Every figure in this example is invented for instruction, not a Labeeb client result. A researcher compares assessment scores for two independent groups of 40 participants. Suppose the analysis reports a mean difference of 5 points, Cohen’s d = 0.42 and a 95% interval for d of 0.10 to 0.74. A complete interpretation would say which group scored higher, define the assessment scale, give the 5-point difference and report the standardized estimate with its interval.
The 5-point difference is easier for a reader to judge if the assessment is scored out of 20 than if it is scored out of 1,000. The interval suggests uncertainty about the standardized magnitude. It does not prove that training caused the difference if the groups were not randomly assigned or important baseline differences remain.
| Weak sentence | Better sentence |
|---|---|
| “The effect was medium and therefore important.” | “The groups differed by five assessment points; the estimated standardized difference was d = 0.42 (illustrative 95% interval 0.10–0.74). Practical importance depends on the assessment scale and study context.” |
| “ANOVA found d = 0.42.” | “Report the effect-size statistic associated with the actual ANOVA table and identify the factor or comparison it describes.” |
Never copy an example value into a real dissertation. Export or preserve the actual output, and check the selected row, group order, missing observations and calculation method.
Check whether your output answers a different magnitude question
This guide focuses on group differences and ANOVA. If your research question asks about association between two variables, a correlation coefficient such as r describes direction and strength on its own scale. If it asks how a regression model explains variation, R2 addresses a model-level question. Neither should be copied into the same sentence as Cohen’s d or partial eta squared as though all were interchangeable measures.
Before reporting, write the estimand in plain language: difference between which groups, relationship between which variables, or variation explained by which model? Then check the actual SPSS procedure, selected row and confidence interval. For a correlation choice, use the linked Pearson-versus-Spearman guide; for a model, follow your institutional reporting rule and the model’s assumptions. A visually neat table cannot rescue a statistic answering the wrong question.
Before submitting the results chapter
- Trace the question: write the exact outcome, groups or factors and whether observations are independent or paired.
- Inspect assumptions and data: compare the analysis with the measurement scale, missing-data decisions and planned design.
- Read the correct table: record the estimate, interval if shown, sample sizes and direction. Do not mix statistics from different models.
- Report both scales: give the original-unit difference where meaningful and a standardized estimate when appropriate.
- Interpret cautiously: explain what the magnitude means in this field and population, and distinguish association from causation.
- Reconcile text and tables: the result, abstract, discussion and figure should describe the same analysis.
For the preceding decision about which analysis fits your variables, use Labeeb’s correlation method guide when it becomes live, or the existing SPSS output and APA guide for table-by-table reporting. This article focuses on magnitude, which those broader guides cover only briefly.
Need to interpret your real SPSS results?
Bring your research question, analysis plan and output tables. Labeeb can help check that the statistic, interpretation and written results agree with your actual data and programme requirements. We do not invent findings or guarantee an academic outcome.
Explore statistical analysis support and dissertation guidance for the project scope.