TTEST Function (Gnumeric)

=TTEST(array1,array2,tails,type)

All four arguments are required; none has an omitted-argument default.

Use the exact integer choices shown. The retained help says fractional selectors are truncated, but the retained 1.12.62 callback tests exact equality before dispatching. Do not rely on fractional inputs being accepted. These formulas use comma separators; a different locale may require different argument separators.

For type 1, corresponding positions must represent the same subject or matched unit. Ranges must have the same number of positions. Gnumeric omits nonnumeric pairs together rather than independently compacting each sample. For types 2 and 3, samples are collected separately, so unequal sample sizes are allowed. Strings, blanks and logical values do not count as observations; input errors should be corrected, not concealed.

Enter A1:A5 as 1, 2, 3, 4, 5 and B1:B5 as 2, 4, 6, 8, 10. Enter each formula in its own cell:

These expected values are independently calculated from the t distribution, not interchangeable answers to a single study. Use type 1 only when row pairs genuinely belong together. The paired differences are -1, -2, -3, -4, -5, with mean -3 and sample standard deviation √2.5. The absolute t statistic is 3√2, with 4 degrees of freedom. For independent tests, the means are 3 and 6, the sample variances are 2.5 and 10, and the absolute t statistic is 3/√2.5. The pooled test has 8 degrees of freedom; Welch’s test has approximately 5.882352941.

Changing tails to 1 returns half the corresponding two-sided result for these examples. Gnumeric uses the absolute t statistic, so TTEST(A1:A5,B1:B5,1,1) and the same formula with the ranges reversed return the same small-tail probability. This argument does not encode a signed alternative such as “the first mean is larger.”

If your predeclared directional hypothesis is that the first mean is larger but the observed first mean is smaller, the relevant directional probability is the large tail, not the small value returned by blindly choosing 1. Check the sample means and the direction before interpreting a one-sided test. Do not choose the direction after seeing the data. For exploratory comparisons, a two-sided test usually avoids that particular mistake.

• Paired ranges of different total sizes produce #N/A; align the observations before changing the formula.
• Invalid tails or type choices produce #NUM! in the examined callback. Use literal 1, 2 or 3 as applicable.
• Too few usable observations produce #DIV/0! in the variance calculations. For a paired test, you need at least two usable pairs.
• Constant paired differences have zero estimated standard error and produce #DIV/0!. A constant shift does not justify reporting an invented p-value of zero.
• Both independent samples being constant also leaves the ordinary test undefined. Examine sample variances before interpreting results.
• Normality assumptions concern paired differences for type 1 and the relevant populations for independent tests. A spreadsheet result does not establish independence, representative sampling or a valid experimental design.

The contract and selectors were inspected in retained Gnumeric 1.12.62 source. Official NEWS lists TTEST among the functions added in Gnumeric 0.27. No retirement or last functioning release is inferred. Independent SciPy calculations provide the expected numeric examples; native results, where available, have the separately stated version and fixture scope below.

AVERAGE, STDEV and COUNT help audit the inputs and the standard-error calculation.

• Gnumeric function reference
• Official function catalog
• Pinned Gnumeric implementation

On 17 October 2026, Gnumeric 1.12.57 recalculated a synthetic workbook through ssconvert in the C.UTF-8 locale. The 6 recorded TTEST fixture cases matched the independently computed numerical results or the source-predicted error cases within floating-point precision. The cases covered all three two-sided tests, rejection of tails 1.9 and type 2.9 with #NUM!, and unequal-sized paired ranges returning #N/A. These records are engine tests; the separately captured worked-example screenshot is scoped below. It does not certify other versions, locales, workbook formats or untested edge cases.

• Official NEWS introduction record