FTEST Function (OpenOffice Calc)

• Introduced: First release not established. The OpenOffice.org 3.3 function appendix documents this name; that establishes historical presence, not its introduction.
• Supported in: Apache OpenOffice 4.1.16; no final supported release established.
• Scope: Desktop Calc. Examples use English function names and semicolon argument separators; the numeric contract is release-specific.
• Checked: October 14, 2026. The two equal-variance order cases were recalculated in Apache OpenOffice 4.1.16 on Linux x86-64. Other examples are independent mathematical references.

FTEST compares the sample variances of two independent normal populations. It computes variances from the observations, rather than accepting two precomputed variance cells. The classical F reference is sensitive to departures from normality; it is not a general-purpose guarantee about arbitrary distributions. Pinned implementation.

A p-value summarizes compatibility with equal population variances under that sampling model. It is not the probability that the populations are identical. Different means can coexist with equal variances, and a non-small p-value is not proof of equality.

=FTEST(Data1; Data2)

Exactly two arguments are required. Each sample needs at least two numeric observations and a nonzero sample variance. The ranges may have different sizes. Use complete numeric ranges so sample size and variance are unambiguous; these examples do not rely on blank or text coercion. Zero is a valid observation and is not a missing-data marker.

The evaluator places the larger sample variance in the numerator and carries that sample’s n−1 degrees of freedom with it. It then doubles the upper-tail result. That last step is the source of the unequal-size caveat below.

Enter the following data. D has only three observations; keep D5 empty and exclude it explicitly in the formulas.

These are independently computed values from the cited release expression, rounded to twelve decimal places. The value above 1 is not a meaningful probability. Both equal-variance order cases were also reproduced in the native check described below; other rows remain mathematical references.

For A and B, VAR gives 1.666666666667 and 6.666666666667, respectively. The selected ratio is 4 with df1=df2=3. Keep the original observations: entering only those two variance cells would leave each FTEST sample with a single observation.

C contains 1, 1, 1, 3 and D contains 1, 2, 3. Both sample variances are exactly 1. An XLSX-import and hard-recalculation check in Apache OpenOffice 4.1.16 on Linux x86-64 (4116m3, Build 9816) saved approximately 1.070483996910 for the three-observation sample first, and 0.929516003090 with the samples exchanged. The saved ODS retained both FTEST formulas and all seven input observations. Stale and uncached SUM controls confirmed recalculation. The equality branch assigns different numerator degrees of freedom when the ranges are exchanged.

To calculate a conventional doubled-smaller-tail result explicitly, put the fixed-order variance ratio in F2 with =VAR(D2:D4)/VAR(C2:C5), then put its upper tail in F3 with =FDIST(F2;COUNT(D2:D4)-1;COUNT(C2:C5)-1). =2*MIN(F3;1-F3) gives approximately 0.929516003090. Both sample variances must be positive.

Do not repair a result above 1 merely with MIN(1;FTEST(…)). Clipping hides the tail-selection issue and does not reproduce the doubled-smaller-tail calculation. Do not choose between a one-sided and two-sided scientific alternative after inspecting the data.

Before using the result, inspect outliers and data-entry problems without automatically deleting observations. Keep measurement units consistent: multiplying just one sample by a conversion factor changes its variance by the square of that factor. Errors from insufficient data or constant samples should remain visible rather than be replaced with zero.

• Function-specific Apache OpenOffice 4.1.16 implementation

The numerical references were recomputed independently against the retained release contract. Only the two opposite-order equal-variance cases have the stated 4.1.16 native observation. Other releases, locales and import formats were not tested.