F.TEST in Collabora Office: Compare Two Sample Variances

=F.TEST(Data1; Data2)
=FTEST(Data1; Data2)

Both arguments are required sample arrays or ranges; there are no tail, significance-level, or paired-test options. Samples may have different lengths. In the pinned Collabora source F.TEST and FTEST have identical two-argument metadata and dispatch to the same ScFTest evaluator without a name-dependent branch. This guide covers both spellings; the dotted name is not a different test.

For predictable data handling, use clean numeric ranges without headers or error cells. The implementation counts the numeric observations collected from each matrix and computes sample variances with denominator n−1. Text and empty entries are not additional numeric observations; zero is an observation. Do not replace missing data with zero unless zero is genuinely measured. Exact coercion of unusual array elements and logical cells is not native-tested.

Enter:

In D2 use =F.TEST(A2:A4;B2:B4), and in D3 use =FTEST(A2:A4;B2:B4). Both have independent target 0.4.

The sample means are 2 and 4. Their squared deviations sum to 2 and 8; dividing by n−1=2 gives sample variances 1 and 4. The implementation puts the larger variance in the numerator, so F=4, with numerator and denominator degrees of freedom both 2. For this special case the F distribution CDF is F/(1+F)=0.8; twice the smaller tail is 2×0.2=0.4.

This hand calculation is an independent reference, not an observed spreadsheet run. Swapping the two sample ranges gives the same expected result because the evaluator also swaps the degrees of freedom when it orders the variances.

The standard interpretation compares equal population variances against a two-sided difference. It assumes independent random samples from normally distributed populations. The spreadsheet can calculate a number without establishing those assumptions. Repeated measurements on the same individuals, dependence, severe outliers, or nonnormal distributions can make the classical interpretation unsuitable.

A p-value of 0.4 does not establish equal variances or a 40% chance that the populations are identical. It says the chosen variance statistic is not unusually extreme at common small significance levels under the equal-variance model. It says nothing directly about equality of means.

The Help’s population-probability wording should not be read as a posterior probability. The pinned evaluator explicitly calculates twice the smaller F-distribution tail.

If D2 contains the p-value and D5 contains a preselected alpha such as 5%, =D2<D5 is FALSE in this example. This is a decision rule, not an argument to F.TEST. Choose the threshold and analysis before inspecting results, and account for multiple comparisons if testing many pairs.

A useful invariance check is to add the same constant to every observation in either sample; variances and the p-value are unchanged mathematically. Scaling both samples by the same nonzero factor also leaves the variance ratio unchanged. Scaling only one sample changes the comparison. Avoid using very large offsets as a numerical stress test in a production analysis.

Each sample must contain at least two numeric observations and have nonzero sample variance. The pinned implementation returns a no-value error for insufficient count or either zero variance; it does not return p=1 for two constant samples. The implementation also rejects inputs that cannot be obtained as matrices.

Do not quote a particular localized error display from these source checks: error rendering and propagation have not been tested in Collabora here. An apparently valid result does not validate input selection, independence, normality, or an analysis plan. Extreme tails can be affected by floating-point precision.

Z.TEST addresses a sample mean using a normal statistic, while T.TEST is a separate mean-comparison function. Neither is a spelling of F.TEST. Collabora Help identifies the dotted F.TEST namespace as COM.MICROSOFT.F.TEST and says it is outside OpenFormula 1.3; verify file exchange in the destination application.

Reviewed on 19 October 2026 against saved Collabora Office 26.04 Help retrieved on 17 October 2026 and the Collabora co-26.04 source snapshot at commit c1013fad0d38683630b48c01b6d5208b8d81181e. The repository is hosted under LibreOffice/core, but the evidence used here is the pinned Collabora branch snapshot, not a claim about another spreadsheet product. The Help URL marked latest can change. Fresh Help retrieval was unavailable during this review, so Help statements were checked against the saved first-party pages; the pinned implementation files were freshly retrieved and verified.

Examples are independently calculated expectations, not recorded Collabora output. Native Collabora and its GUI were unavailable for this review; no screenshots or application runs are claimed. The first release introducing this function and the last working binary version have not been established. Online, mobile, older releases, localized function names, and round-trip compatibility need separate verification. Formulas use English names and semicolon separators; use the separators configured in your installation.

• Collabora Help: F.TEST and FTEST — reviewed 19 October 2026.
• Pinned two-sided F-test evaluator — reviewed 19 October 2026.
• Pinned same-dispatch spellings — reviewed 19 October 2026.
• Pinned argument metadata — reviewed 19 October 2026.