MEDIAN in Collabora Office: Find the Middle of Numeric Observations

• Introduced: original Collabora release and launch date are not established by the sources reviewed.
• Availability: documented in Collabora Office 26.04 help, scoped to the desktop-help release.
• Platform scope: online and mobile deployments are not inferred from desktop documentation.
• Checked: October 14, 2026. Function-specific documentation and independent calculations; no native Collabora test was performed.

=MEDIAN(argument1;[argument2];...)

Collabora documents up to 255 arguments, which may be numbers, text, cell references or ranges. Its function section explicitly says text and empty cells within a data range are ignored. The examples use numeric measurements and clearly marked text controls, without extending that range rule to every direct-argument coercion case.

The median is a positional summary, not the arithmetic mean of all observations. The source order need not already be sorted; sorting below is a transparent manual check. Formulas use English function names and semicolon separators.

Put these six task durations, in minutes, in A2:A7:

=MEDIAN(A2:A7)

Expected result: 9.5. The sorted values are 5, 6, 8, 11, 11, 17. Positions three and four contain 8 and 11, so their mean is (8+11)/2=9.5. A median need not equal a value appearing in the original data.

Add numeric 21 in A8 and extend the formula:

=MEDIAN(A2:A8)

Expected result: 11. Seven values now have one central position, the fourth. Adding another observation can change which positions define the median even when that new value is not itself central.

For the original six values, AVERAGE gives approximately 9.666666667, from total 58 divided by six. Median and mean are different statistics; neither result indicates an arithmetic error in the other.

Create C2:C6 containing numeric 0, numeric 6, numeric 14, text Pending, and a genuinely empty cell.

=MEDIAN(C2:C6)

Expected result: 6. Only the three numeric values participate, including the genuine zero. Replacing Pending with numeric 100 changes the expected median to 10, the mean of 6 and 14. Merely displaying text that resembles a number is not the same controlled edit as supplying a numeric measurement.

Preserve units as numbers, with Minutes in a heading rather than embedded in each cell as text such as “6 minutes”. The ignored-value rule can otherwise remove observations without producing an obvious formula error.

Put 1, 5, 90 in E2:E4 and 20, 30 in G2:G3.

=MEDIAN(E2:E4)
=MEDIAN(G2:G3)
=MEDIAN(E2:E4;G2:G3)

Expected results are 5, 25 and 20. Averaging the two group medians gives 15; weighting them by counts gives (5*3+25*2)/5=13. Neither reconstructs the combined median of 20.

The group medians discard the distribution of the original values. Combine the underlying observations when the intended question concerns the middle of all records. A count-weighted average is appropriate for reconstructing certain means, not a general reconstruction rule for medians.

The controls =MEDIAN(2;4;6;8;100) and =MEDIAN(2;4;6;8;1000) both have expected median 6. Their means are 24 and 204. The unchanged center does not prove that the extreme record is unimportant or erroneous.

Duplicates also remain observations: =MEDIAN(1;3;3;3;12) has expected 3. Do not deduplicate a measurement register unless each unique value, rather than each observation, is the intended unit of analysis.

LARGE and SMALL return explicit ordered positions. MEDIAN chooses its central position or pair based on the participating count. Its direct section does not establish every entirely empty-range, Boolean, error-containing or overflow case, so exact native outcomes for those boundaries remain unverified.

Retain precision until the presentation stage. Displaying 9.5 as 10 after rounding does not change the underlying median, and a display preference should not silently redefine the statistic.

• Collabora Office 26.04: MEDIAN.
• Editorial testing standards.

The function-specific Collabora documentation was reviewed. Worked results are independent arithmetic or documented type-rule controls, not captured output from the application. Client-specific entry procedures and the boundaries identified above remain subject to native verification.