KURT Function (Calligra Sheets)

• Introduced: First release not established; the name is present in Calligra 2.4.0 source.
• Supported in: Calligra Sheets 26.08.2, with no retirement established.
• Scope: Desktop Sheets with the Statistical module. Examples use English semicolon-separated formulas.
• Checked: October 14, 2026. Selected formulas ran through the official 25.04.2 engine library. Source and independent mathematics were checked. This is engine evidence, with no GUI-cell or import verification; 26.08.2 was source-checked only.

KURT calculates the finite-sample corrected excess-kurtosis estimator for a numeric dataset. It uses a fourth central moment and subtracts the normal-reference offset. Excess kurtosis is a tail-shape statistic; it is not a test that the data are normal and should not be interpreted merely as peak height. Function help.

=KURT(value1[;value2;...])

Supply at least four numeric observations through one or more arguments. The function accepts arrays and ranges by registration; these engine examples use explicit scalar lists, not a native-cell range fixture. A constant dataset has zero spread and cannot produce a defined statistic. Brackets indicate optional arguments and are not entered in the formula. Registration and implementation.

The engine and mathematical columns are separate. Results are rounded to twelve decimal places, or thirteen significant digits in scientific notation for very small or large values. This is a display choice, not an accuracy guarantee. NaN and infinity are nonfinite outputs; an engine error flag is checked separately from those values.

For values 1,2,3,4, the sample mean is 2.5 and the sample variance is 5/3. The corrected sample excess kurtosis is −1.2, matching the ordinary engine example. For 1,1,1,2, the result is 4. A small dataset can change this statistic dramatically when a single observation changes.

Do not automatically delete the largest or smallest observation to force the statistic toward zero. Investigate measurement errors using an established rule and retain genuine observations when the model requires them.

Writing s²=Σ(x−mean)²/(n−1), the estimator is n(n+1)Σ((x−mean)/s)^4/((n−1)(n−2)(n−3))−3(n−1)²/((n−2)(n−3)). This is distinct from the uncorrected population moment ratio.

Translation and nonzero rescaling preserve kurtosis. The table compares 1,2,3,4 with 2,4,6,8 and 11,12,13,14, and the executed difference below confirms the translation check for those explicit lists.

The executed advanced check was:

=KURT(1;2;3;4)-KURT(11;12;13;14)

Its engine result was 0.

Keep this calculation numeric-only. The callback combines full-value count/mean defaults with a numeric-only standard deviation, so mixed types are handled inconsistently. Adding text “abc” or text “5” to 1,2,3,4 gives 7.3 in the tested engine, rather than ignoring the text or consistently including a fifth numeric observation. TRUE and FALSE also alter the result. The boundary table preserves these observations; they are not recommended data-cleaning semantics. Fewer than four observations gives #VALUE!, while a constant four-value list gives #DIV/0!.

• Function-specific implementation
• Versioned function help
• Shape Defaults helper
• Shape Aggregation helper
• Shape Walk helper
• Stddev helper

The statistical module and value-calculation helpers are byte-identical between the inspected 25.04.2 and 26.08.2 sources. That comparison supports the source discussion but does not establish later-version native output, every locale or arbitrary extreme-parameter accuracy.

AVERAGE, STDEV.