Levene Test Calculator

Provide each group as an array of numeric observations, and it returns Levene's statistic with its upper-tail p-value. This implementation uses the classical mean-centered version of the test: it converts every observation to its absolute distance from that group's mean, then compares those distances with a one-way analysis of variance. The calculation is deterministic, requires no network access, and is useful when a normality-sensitive two-sample F test is not a good fit.

What Levene's test measures

Levene's test begins with the null hypothesis that all represented populations have the same variance. For every group, the calculator finds the arithmetic mean and replaces each original observation with its absolute distance from that mean. It then performs a one-way analysis of variance on those distances. If the typical absolute distances differ substantially among groups compared with the variation remaining inside the groups, the statistic becomes large. The returned p-value is the probability, under the equal-variance model and the F reference distribution, of obtaining a statistic at least as large as the observed one. A small p-value therefore provides evidence against equal variances. It is not the probability that the null hypothesis is true, and it does not quantify whether a variance difference is important in the application. A large p-value likewise does not prove that variances are identical; limited sample sizes may simply provide little power to detect a difference. Set a significance threshold before inspecting the result, and interpret the test alongside plots, sample sizes, measurement knowledge, and the practical consequences of unequal spread.

How the calculation is performed

Enter `groups` as an array containing at least two arrays of finite numbers. Every group needs at least two observations, although group sizes do not need to match and the order of observations has no effect. The calculator first computes each group's mean with an incremental update that avoids an unnecessarily large raw sum. It forms absolute deviations from that mean, calculates the mean deviation within every group, and calculates the overall deviation mean weighted by group size. The reported statistic is the between-group mean square divided by the within-group mean square for those deviations. Its numerator degrees of freedom equal the number of groups minus one; its denominator degrees of freedom equal the total observation count minus the number of groups. The p-value is the upper tail of the corresponding F distribution, evaluated deterministically through the regularized incomplete beta function. The response intentionally contains only `statistic` and `p_value`, keeping it straightforward to store, compare, or feed into a documented analysis pipeline. When reporting a result, also preserve the group definitions, sample sizes, chosen threshold, and mean-centered method.

Assumptions, interpretation, and limitations

Observations should be independent within and between groups, and the samples should reasonably represent the populations about which you want to conclude. Levene's procedure is generally less sensitive to departures from normality than the classical F test for two variances, but it is not assumption-free. Strong dependence, biased sampling, measurement rounding, mixtures of populations, or influential outliers can still produce a misleading result. The mean-centered form used here is the original Levene construction. A median-centered Brown–Forsythe variant can be more resistant to heavy tails and extreme observations, but it is a different calculation and is not silently substituted by this capability. Inspect distributions and consider the data-generating process before turning a p-value into a decision. The calculator rejects fewer than two groups, groups shorter than two observations, nonnumeric or non-finite values, and inputs beyond its declared bounds. It also reports an error when all within-group absolute deviations lack variation, because the denominator of the statistic is then zero and an ordinary finite F statistic cannot be calculated. Statistical significance alone does not establish causation, data quality, or meaningful effect size.