You have just run an independent-samples t-test or a one-way ANOVA, and the p-value looks promising. Before you celebrate, there is one assumption you need to check: do your groups have roughly equal variances? If they do not, your p-value may be misleading. That is exactly what Levene’s test checks, and in SPSS it takes less than a minute to run. In this tutorial, you will learn what Levene’s test does, how to run it three different ways in SPSS, how to read the output, and what to do when the test comes back significant.
What Is Levene’s Test?
Levene’s test checks the assumption of homogeneity of variance — the idea that the spread (variance) of scores is similar across the groups you are comparing. It is an assumption of the independent-samples t-test and of one-way ANOVA, because both procedures pool the group variances into a single estimate. If one group’s scores are far more spread out than another’s, that pooled estimate becomes unreliable, and so does the resulting p-value.
The logic of the test is simple. For every score, SPSS calculates how far it lies from its own group centre. Levene’s test then asks whether those distances differ systematically between groups. If the average distance is about the same in every group, the variances are plausibly equal. If one group consistently has larger distances, the variances differ.
The hypotheses are straightforward:
- Null hypothesis: the population variances are equal across groups.
- Alternative hypothesis: at least one group’s variance differs from the others.
Note the direction of interpretation, which trips up many students: a NON-significant result (p > .05) is the outcome you want here. It means you have no evidence against equal variances, so you can proceed with the standard test. A significant result (p ≤ .05) means the equal-variance assumption is violated and you need a different approach, covered later in this guide.
There are several versions of the test. The classic version centres scores on the group mean. A popular alternative, often called the Brown-Forsythe version, centres scores on the group median instead, which makes it more robust when your data are skewed. SPSS lets you choose between them in the Explore procedure, and both are reported in the same output table.
When Should You Run Levene’s Test?
Run Levene’s test whenever you are about to run a test that assumes equal variances, which in practice means:
- Before an independent-samples t-test, when comparing the means of two unrelated groups.
- Before a one-way ANOVA, when comparing the means of three or more unrelated groups.
- Before a factorial ANOVA, where the assumption applies to the variance within each cell of the design.
You do not need it for paired (dependent-samples) t-tests or repeated-measures ANOVA, because those procedures work with difference scores rather than comparing separate groups, and they carry a different set of assumptions.
One caution: like all significance tests, Levene’s test is sensitive to sample size. With very large samples, it can flag tiny, practically unimportant differences in variance as significant. With very small samples, it may lack the power to detect real differences. Always look at the actual variances and standard deviations in your descriptive statistics alongside the test result, and let the size of the difference — not just the p-value — guide your decision.
How to Run Levene’s Test in SPSS
Method 1: Through Explore (the Most Detailed Option)
The Explore procedure gives you the fullest version of Levene’s test, including all four centring options. It is the best choice when Levene’s test is the main thing you want.
- Click Analyze > Descriptive Statistics > Explore.
- Move your dependent variable (the scores you measured) into the Dependent List box.
- Move your grouping variable into the Factor List box.
- Click the Plots button.
- Under Spread vs. Level with Levene Test, select Untransformed.
- Click Continue, then OK.
SPSS produces a table titled “Test of Homogeneity of Variance” with four rows: Based on Mean, Based on Median, Based on Median and with adjusted df, and Based on trimmed mean. For most purposes, report the “Based on Mean” row, which is the classic Levene’s test. If your data are noticeably skewed, the “Based on Median” row (the Brown-Forsythe version) is the safer choice.
Method 2: Through One-Way ANOVA
If you are already running an ANOVA, you can get Levene’s test as part of the same analysis, with no extra procedure needed.
- Click Analyze > Compare Means > One-Way ANOVA.
- Move your dependent variable into the Dependent List and your grouping variable into the Factor box.
- Click the Options button.
- Tick Homogeneity of variance test.
- Click Continue, then OK.
The output will include a “Test of Homogeneity of Variances” table showing the Levene statistic, its two degrees of freedom, and the significance value. This is the same classic mean-based version of the test.
Method 3: It Comes Free with the Independent-Samples T-Test
If you are running an independent-samples t-test, you do not need to run Levene’s test separately at all. SPSS includes it automatically. Click Analyze > Compare Means > Independent-Samples T Test, set up your test as usual, and the first section of the output table is labelled “Levene’s Test for Equality of Variances”, with an F value and a significance value. SPSS then gives you two rows for the t-test itself: “Equal variances assumed” and “Equal variances not assumed”. The Levene’s result tells you which row to read, as explained below.
SPSS Syntax for Levene’s Test
If you prefer syntax, or you want a record of exactly what you ran, paste the following into a syntax window and run it. Replace Score with your dependent variable and Group with your grouping variable.
EXAMINE VARIABLES=Score BY Group /PLOT SPREADLEVEL /STATISTICS DESCRIPTIVES /MISSING LISTWISE /NOTOTAL.
For the One-Way ANOVA version, the syntax is:
ONEWAY Score BY Group /STATISTICS HOMOGENEITY /MISSING ANALYSIS.
Running your analysis from syntax has a real advantage for theses and dissertations: your method is fully reproducible, and you can re-run the identical analysis later without remembering which boxes you ticked.
How to Interpret Levene’s Test Output
The output table has four columns you need: the Levene statistic (an F value), df1, df2, and Sig. (the p-value). Here is an example of what the table might look like for three groups of 30 students each:
Levene Statistic = 4.82, df1 = 2, df2 = 87, Sig. = .010.
Because the p-value (.010) is below .05, you reject the null hypothesis of equal variances. The assumption is violated, and you should not report the standard ANOVA result. Instead, use one of the alternatives in the next section.
If instead the output read Sig. = .324, you would fail to reject the null hypothesis. There is no evidence that the variances differ, so you proceed with the standard t-test or ANOVA and read the “Equal variances assumed” row.
Two practical notes. First, always check the descriptive statistics too: if one group’s standard deviation is more than about twice another’s, treat the variances as unequal even if Levene’s test is non-significant, especially with small samples. Second, remember that failing to reject the null hypothesis is not proof that the variances are exactly equal — it is simply an absence of evidence against equality, which is all the assumption requires.
What to Do When Levene’s Test Is Significant
A significant Levene’s test is not a dead end. You have several good options:
- Use Welch’s test. For ANOVA, Welch’s ANOVA does not assume equal variances and is available in SPSS under Analyze > Compare Means > One-Way ANOVA > Options (tick Welch). For two groups, read the “Equal variances not assumed” row of the t-test output — SPSS has already done the Welch correction for you.
- Use the Games-Howell post hoc test. If you follow up a significant Welch’s ANOVA with pairwise comparisons, Games-Howell is the post hoc procedure designed for unequal variances, and it does not require equal group sizes either.
- Consider a data transformation. If your data are skewed, a transformation such as a log or square-root transformation can sometimes stabilise the variances as well as improve normality. Re-run Levene’s test after transforming to check.
- Consider a nonparametric alternative. If the variances differ wildly and transformations do not help, the Mann-Whitney U test (two groups) or the Kruskal-Wallis test (three or more groups) avoid the equal-variance assumption altogether — though they test a slightly different hypothesis, so report them as such.
For most thesis work, option 1 is the right default: Welch’s correction is robust, easy to report, and widely accepted.
Reporting in APA Style (7th ed.)
APA style does not require a long discussion of Levene’s test. A single sentence in your results section, usually before the main test result, is enough. Report the test statistic with its degrees of freedom and the exact p-value.
Example when the assumption holds:
“Levene’s test indicated equal variances across groups, F(2, 87) = 1.24, p = .294, so the standard one-way ANOVA was used.”
Example when the assumption is violated:
“Levene’s test indicated unequal variances across groups, F(2, 87) = 4.82, p = .010, so Welch’s ANOVA was used instead.”
If you used the median-based version, name it so a reader can reproduce your analysis:
“Levene’s test based on the median (Brown-Forsythe) indicated equal variances across groups, F(2, 87) = 0.96, p = .387.”
Note that the numbers in these examples are illustrative. Always report the values from your own output, with the exact p-value SPSS gives you rather than a threshold like p < .05, unless your department or journal specifies otherwise.
Key Takeaways
- Levene’s test checks whether group variances are equal, an assumption of the independent-samples t-test and one-way ANOVA.
- A non-significant result (p > .05) means you can proceed with the standard test; a significant result means you need an alternative such as Welch’s test.
- The quickest route is through Analyze > Descriptive Statistics > Explore, but ANOVA and the t-test procedures include the test automatically.
- Always look at the group standard deviations alongside the p-value, and remember that very large samples can make trivial variance differences significant.
- Report Levene’s test in one sentence with the F value, degrees of freedom, and exact p-value before your main result.
FAQs
Is Levene’s test the same as Bartlett’s test?
No. Both test homogeneity of variance, but Bartlett’s test is very sensitive to departures from normality, while Levene’s test is more robust. For most real-world data, Levene’s test is the safer default, and it is the one SPSS provides.
What if Levene’s test is significant but my group sizes are equal?
Equal group sizes make t-tests and ANOVA fairly robust to unequal variances, so a significant Levene’s test is less alarming when your groups are the same size. Still, reporting Welch’s test costs nothing and removes the doubt, so it remains the recommended choice.
Do I need Levene’s test for a paired-samples t-test?
No. The paired-samples t-test works with difference scores within the same people, so there are no separate group variances to compare. The relevant assumption there concerns the distribution of the difference scores.
Which row should I report from the Explore output?
For most analyses, report the “Based on Mean” row — that is the classic Levene’s test. If your data are skewed, report the “Based on Median” row and name it as the Brown-Forsythe version.
Can I skip Levene’s test and just always use Welch’s test?
Many statisticians would say yes. Welch’s test performs well whether or not variances are equal, so it is a defensible default. Even so, examiners and reviewers commonly expect to see the assumption checked and reported, so running Levene’s test remains good practice for thesis work.
Need this done on your data? Get a free quote →