TTEST in Calligra Sheets: Compare Two Samples Carefully

=TTEST(sample1; sample2; tails; type)

All four arguments are required, with no defaults.

Select the test from the study design and defensible assumptions before looking for a small p-value. Do not choose paired versus independent solely because one gives a smaller answer, or select equal variance just because two sample variances look superficially similar.

Suppose the same five test jobs are measured with two configurations. The example times are in seconds:

1. Keep each job’s two measurements on the same row. Do not sort the two columns independently.
2. In E2 enter =TTEST(B2:B6;C2:C6;2;1).
3. Independent expected p-value: 0.00283784592673.
4. The within-job differences are [2, 1, 2, 1, 2], with mean 1.6 seconds. Report this effect size alongside the p-value rather than treating p as the magnitude of improvement.
5. The paired test concerns the mean difference under its model. It does not establish that these five jobs represent every production workload.

For an ordinary paired t-test, the pairs should be independent across units and the difference distribution should be suitable for the t procedure, particularly with such a small sample. A process change is not established as causal merely because a spreadsheet returns a small probability; study design and confounding matter.

Using these same numbers as independent samples produces very different mathematical answers: approximately 0.297313963931 for type 2 and 0.297340459089 for type 3. These are not interchangeable estimates of one identical question. The paired calculation uses the strong within-job association; the independent calculation does not.

For genuinely independent observations, type 3 estimates uncertainty using separate sample variances and an effective degrees of freedom. It can accommodate unequal sample sizes. Type 1 requires matching counts and meaningful one-to-one pairs. Do not invent pairing to gain significance.

The inspected TTEST implementation passes its t statistic to TDIST, which uses t². As a result, tails = 1 produces the same magnitude-tail probability when the two sample ranges are reversed. For the paired example, the independent source-aligned expectation is 0.00141892296337 in either order, half the two-sided p-value.

That is not a complete implementation of a preselected directional hypothesis. If the observed effect points opposite to the hypothesis, the appropriate directional p-value is not automatically this small half-two-sided number. Check the sign of the mean difference and the predefined alternative separately; use a statistical tool that explicitly accepts the direction when needed. Do not reinterpret the direction after viewing the result.

• Each sample must contain at least two counted numeric observations. A one-value range fails the source check.
• Paired type 1 requires equal counted sample sizes; the provided unequal-length case returns an internal value error.
• Tails outside 1–2 or type outside 1–3, after integer conversion, return an internal value error.
• Constant samples or zero-variance differences can make denominators zero. Such degenerate inputs are not covered by an ordinary significance interpretation, even if an error or number is returned.
• Missing/text observations can disrupt pair alignment and counting. Use clean numeric, explicitly matched samples rather than assuming mixed ranges are handled as intended.

A p-value is a probability of results at least as incompatible with the null model under that model’s assumptions, not a probability that the observed difference is “due to” a particular cause. Report the design, sample sizes, effect estimate and uncertainty, and account for multiple testing where applicable.

KDE TDIST supplies the t-tail calculation; SciPy paired t-test documentation distinguishes two-sided and directional alternatives and describes the independent reference calculation. These SciPy references are not claims of Calligra’s runtime behavior. Recheck type, tail convention and sample handling when moving formulas between applications.

Checked October 18, 2026. Original introduction date is unknown. TTEST is registered in both inspected source versions, Calligra 25.04.2 and 26.08.2. Their relevant statistical-function, calculation-engine and converter source files are byte-identical. Version inspection establishes availability in those releases, not a first-introduction date or a runtime guarantee for every platform.

Primary sources: KDE TTEST handbook, 26.08.2 implementation, 25.04.2 statistical functions, and numeric converter. Mathematical values above are independent calculations. Native observations below were obtained in Calligra Sheets 25.04.2 on Linux, October 18, 2026; no native 26.08.2 execution is claimed. The handbook is a living reference and can differ from the pinned implementation at the explicitly described boundaries.

The following formula results were obtained in the actual application and checked in its completed native-saved workbook. The screenshot is an unmodified application capture. The values in this table are native results; the earlier reference values and model calculations remain separately identified. Small final-digit differences do not make the independent reference a native observation.

Native H2:H6 contains 12, 14, 15, 16, 18 and I2:I6 contains 10, 13, 13, 15, 16. The equal one-tail results after reversing these ranges confirm the implementation’s direction limitation; they do not establish two identical directional hypotheses.

The completed saved values, including error text, determine this table; display rounding can show fewer digits. These tests do not establish all locales, input types, extreme numerical ranges or cross-application import behavior.