METHOD GUIDE
One-Way ANOVA in SPSS
Unequal Variances, Welch and Post Hoc Choice
SPSS offers the standard F test, two robust alternatives and a choice of post hoc methods, and the right combination depends on your variances and group sizes. This guide covers the checks, the output, the effect sizes and the APA 7 write-up.
IN SHORT
What the Test Answers and What Follows It
A one-way ANOVA asks whether the means of a continuous outcome differ across three or more independent groups, and the omnibus F answers it with a single p-value. The question researchers usually care about comes afterwards: which groups differ, by how much, and with what uncertainty. To run a one-way ANOVA in SPSS, open Analyze, Compare Means (Compare Means and Proportions in recent versions), One-Way ANOVA, request the Welch test under Options, and choose a post hoc method that matches the group variances. OnlineSPSS, a specialist statistical consultancy for academic, clinical and professional research, carries out and reviews this analysis through an in-house PhD-level statistician.
ANALYSIS SUPPORT
Analysis Support for This Method
Once variances and post hoc methods are settled, two pages take it further: the full multi-group analysis on your data, and a rank-based route for outcomes that means do not describe well.
One-Way ANOVA
The comparison of several group means carried out within your study design, with the variance handling, the post hoc strategy, effect sizes with intervals and the reporting agreed for your data.
Kruskal-Wallis Test
A rank-based comparison of three or more independent groups, for an ordinal outcome or when a comparison of means does not answer the research question.
BEFORE YOU RUN IT
Choices That Shape the ANOVA Output
Four choices fix what the SPSS output can legitimately say, and they are easier to defend when they are made from the design before any table has been seen.
Independent Groups Each participant sits in one of three or more groups and is measured once. Repeated measurements or participants nested in classes or sites call for a model that respects that structure.
The Comparisons You Need Decide whether you need all pairwise comparisons, comparisons with a control, or a few planned contrasts. A focused question answered by a planned contrast keeps more power than all pairs.
Variances and Group Sizes Unequal spread matters most when group sizes also differ, and worst when the smallest group has the largest variance. Welch's F costs little when variances are equal, so it is a defensible default.
Outcome and Residual Shape Examine the outcome within each group in a plot. A clearly ordinal outcome may suit a rank-based test, which answers a question about distributions and not about means.
IN SPSS
How to Run a One-Way ANOVA in SPSS
The outline is for IBM SPSS and covers only the choices that change the output, since dialog labels differ a little between versions.
- Confirm the design: one continuous outcome, one factor with independent groups.
- Plot the outcome by group, and compare group sizes and standard deviations.
- Open Analyze, Compare Means (Compare Means and Proportions in recent versions), One-Way ANOVA, place the outcome under Dependent List and the factor under Factor, and in recent versions select the effect size estimate in the same dialog.
- Under Options, request descriptive statistics, the homogeneity test and the Welch and Brown-Forsythe tests.
- Under Post Hoc, choose a method that fits the variance situation and the group sizes.
- Read the descriptives, then the robust tests, then the multiple comparisons.
READING THE OUTPUT
What Each Part of the SPSS One-Way Output Tells You
Two tables test the overall hypothesis, ANOVA and Robust Tests of Equality of Means, and the second is the one most often missing from the write-up.
| Output element | What it tells you | What to check |
|---|---|---|
| Descriptives | The N, mean, standard deviation and 95% interval for each group | That group sizes are as designed and that the largest and smallest standard deviations are compared |
| Test of Homogeneity of Variances | Levene’s statistic, in several variants in recent versions | That it informs the choice but does not decide it, since spread and group sizes matter more |
| ANOVA | The standard F test, which assumes equal variances | That it is reported only when equal variances are plausible, with both degrees of freedom |
| Robust Tests of Equality of Means | Welch and Brown-Forsythe statistics with adjusted denominator degrees of freedom | That the Welch row is reported with its non-integer degrees of freedom when variances differ |
| Multiple Comparisons | Pairwise mean differences with adjusted p-values and intervals for the method chosen | That the method matches the variances: Tukey’s HSD assumes equal ones, Games-Howell does not |
| Homogeneous Subsets | Groups that do not differ, for certain equal-variance post hoc methods | That it is not read as proof of no difference, and that it is absent for Games-Howell |
| ANOVA Effect Sizes (recent versions) | Eta-squared, epsilon-squared and omega-squared with intervals | That the one reported is named, since eta-squared overstates the effect in small samples |
REPORTING
The APA 7 Template for a One-Way Result
Use the wording below as a frame, completing each blank from your own output. It reports no result.
Write the omnibus result as F(df1, df2) = ___, p = ___, using the Welch row and its fractional degrees of freedom when variances differ, followed by an effect size such as omega-squared = ___ with its 95% CI [___, ___]. Then report the post hoc comparisons that answer the research question as mean differences with their adjusted intervals, and name the method and the family of comparisons it adjusts over. Give the mean and standard deviation of each group with its N. When Welch ANOVA in SPSS was used, say so in the text and in the table note.
COMMON MISTAKES
Mistakes Reviewers Find in Reported One-Way Analyses
Six habits draw the most reviewer comments on a multi-group comparison, and most concern what is reported after the omnibus F.
Tukey After Unequal Variances Tukey's HSD assumes similar variances. After clear differences in spread and group size, it can give misleading error rates, and a method that does not assume equal variances is the better match.
Only the Omnibus Result A significant F says that the means are not all equal. Readers need the pairwise differences with adjusted intervals to know which groups differ and by how much.
Subsets as Evidence Homogeneous subsets group means that were not distinguished. A group left out of a difference is not shown to be equal, so the intervals for the differences carry the interpretation.
Post Hoc Method Not Stated Adjusted p-values depend on the method and the number of comparisons. A reader cannot judge them unless the text names the method and the family of comparisons covered.
Eta-Squared Without a Name Eta-squared, epsilon-squared and omega-squared differ in size, particularly in small samples. The effect size is named, and its interval is given when the output provides one.
Tiny Groups Overlooked A group with very few cases makes the variance estimate unstable and the robust and post hoc methods less reliable. Group sizes belong in the report, and thin cells are discussed.
Run the comparison already? Independent Statistical Review checks the variance handling, the post hoc choice and the write-up before you submit.
FREQUENTLY ASKED QUESTIONS
Questions About One-Way ANOVA Output
Short answers to what researchers ask when the SPSS tables disagree, from Welch and post hoc methods to effect sizes, confidentiality and the assessment.
Where is the Welch test in the SPSS one-way output?
It appears in a separate Robust Tests of Equality of Means table, and only when Welch is ticked under Options before the analysis is run. The standard ANOVA table is always printed, which is why it often gets reported by default. Report the Welch row with its non-integer denominator degrees of freedom when group variances differ, and consider it the default when group sizes are unequal.
Which post hoc test follows a Welch result?
Games-Howell in SPSS pairs naturally with Welch’s F, because it does not assume equal variances or equal group sizes. It can be too liberal when groups are very small, so group sizes are reported. Tukey’s HSD suits designs with equal spread. When only a few comparisons matter, planned contrasts are more powerful than all pairs.
Does the overall F have to be significant before post hoc tests?
Not always. Procedures such as Tukey’s HSD and Games-Howell control the familywise error rate on their own. The practical point is to prespecify which comparisons answer the research question, so that nothing is chosen after seeing which differences are large.
Can OnlineSPSS settle the variance handling and post hoc choice for my data?
Yes. An in-house PhD-level statistician can look at your group sizes and spread, confirm whether the Welch and Games-Howell route or the standard one fits, and carry out the analysis or review one you have run. If only the presentation of the comparisons is outstanding, Results Reporting & Interpretation covers that stage. The research question and conclusions remain with you.
How do you protect my research data and confidentiality?
Your data and project details are handled confidentially, and project information is not published. Researcher names and institutions are not used publicly without permission. An NDA is available where required and can be arranged before sensitive materials are shared. Read more about confidentiality and data security.
What happens when I request an assessment, and how long does it take?
You share your research question, design and relevant materials, such as the dataset or the output you have, and a statistically trained Research Coordinator reviews the request and responds the same day. The analysis, deliverables, timeline and one fixed project fee are then confirmed before any work begins, so the timeline follows from the agreed scope.
Still have a question? Find more answers in our full FAQ.
NEXT STEP
Unequal Groups and
Unsure Which Tests to Report?
Share the group sizes, the outcome and the tables you have run. The variance handling, the comparisons to report and the scope are confirmed with you before any work begins.
No Commitment Required
Strictly Confidential
Fixed Project Fee
In-House PhD-Level Statisticians
Same-Day Response
Serving Researchers Since 2014