COVAR in Collabora Office: Calculate Population Covariance of Paired Data

=COVAR(Data1; Data2)

Data1 and Data2 are equally shaped ranges or arrays of paired observations. Each row’s first value must correspond to that row’s second value. Sorting one column independently changes the pairs and can change the result.

COVAR uses the population-style denominator n: the sum of paired centered products divided by the number of usable pairs. It is not the sample covariance obtained by dividing by n-1. The current source maps the legacy COVAR operation to the population covariance calculation.

The result has units equal to the product of the two input units. Unlike a correlation coefficient, covariance is not restricted to the interval from -1 to 1.

Use the four pairs from the quick answer. Both column means are 2.5. In C2 enter =A2-AVERAGE($A$2:$A$5) and in D2 enter =B2-AVERAGE($B$2:$B$5); fill both down.

The first deviations are -1.5, -0.5, 0.5, 1.5. The second deviations are -0.5, -1.5, 1.5, 0.5. In E2 enter =C2*D2 and fill down; all four products equal 0.75.

Their sum is three. Dividing by the four observations gives 0.75, agreeing with COVAR. Dividing by three instead gives 1, the corresponding sample covariance for these same pairs.

Replace the second column with 3, 4, 1, 2, preserving the first column. The independent covariance is -0.75. The second column has the same values and mean as before but a different pairing with the first column.

This is why matching records by an identifier or observation time matters before calculating covariance. Equal list lengths alone do not establish that the rows correspond.

Keep both original columns and any matching key when preparing data. A plausible positive or negative covariance does not prove that the join or sorting operation was correct.

Multiply each original B value by ten. COVAR of A and the scaled column becomes 7.5. Adding 100 to every A value leaves the original covariance 0.75 because subtracting the mean removes that common shift.

Multiplying A by two and B by three multiplies covariance by six, giving 4.5. A larger numerical covariance can therefore reflect changed units rather than stronger standardized association.

For the original dataset, each population variance is 1.25, so the standardized ratio 0.75/SQRT(1.25*1.25) is 0.6. This illustrates the relation to correlation without treating covariance itself as a unitless coefficient.

For identical first and second ranges, COVAR(A2:A5;A2:A5) equals the population variance of A, independently 1.25. Its sample variance is 5/3, approximately 1.666666667.

This self-pair control can expose a mistaken n-1 denominator. Use full stored values while comparing, because rounded means or centered products can introduce small numerical differences.

The current source requires equal matrix dimensions and iterates numeric pairs together. It does not independently compact each column and then zip the remaining values. Use complete numeric pairs in reproducible examples, and review missing observations before interpreting a result.

At least one usable pair is required in the population path; with one pair, both centered deviations are zero and the mathematical covariance is zero. That does not provide meaningful evidence about association. Exact error displays and every error-cell or Boolean conversion are not native-tested.

Covariance describes joint variation, not causation. Its sign and magnitude depend on the paired data, units and chosen population/sample treatment. Preserve the underlying observations rather than reporting an isolated number without its context.

The exact function is documented in Collabora Office 26.04 desktop help. The cited help snapshot was retrieved on 2026-10-13 and hash-verified October 15, 2026. Desktop documentation does not establish Online or mobile support. First introduction and last functioning release are unknown in this evidence.

No native Collabora Office execution occurred. All numerical results are independent mathematical expectations, with English names and semicolon argument separators. Native error displays, all coercions and extreme numerical behavior are not certified.

Implementation statements are corroborated by Collabora current engine source, retrieved October 15, 2026. This is moving-main source evidence, not a pinned desktop 26.04 binary or execution.

The legacy operation mapping is corroborated in Collabora dispatch source.

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