COVAR in Zoho Sheet: Covariance with Paired Data and Examples

Use COVAR in Zoho Sheet to ask whether two paired numeric variables tend to rise and fall together. For numeric measurements in A2:A7 and B2:B7, the official Zoho syntax is:

=COVAR(A2:A7;B2:B7)

For the six-row dataset below, the independently calculated conventional population-covariance expectation is 12.833333333. A positive covariance suggests the two variables tend to vary in the same direction; a negative one suggests opposite directions. Covariance magnitude depends on the variables’ units and scales, so 12.83 is not automatically a “stronger” relationship than 3.2 in another dataset. Zoho lists the function and two required ranges in its official COVAR help (checked 2026-10-12).

Important calculation caveat: Zoho’s individual COVAR page publishes numerical examples but does not explicitly label the denominator as n or n−1. The guide illustrates the conventional legacy population method (sum of paired deviation products divided by n) and separately shows the sample alternative; the predicted COVAR outputs are not actual Zoho Sheet observations. Native testing is required to verify the exact engine result and edge cases.

COVAR(x; y)

No optional arguments or defaults are documented in the Zoho COVAR function page. Zoho’s English examples use a semicolon argument separator. Never assume the same separator applies to every localized spreadsheet editor.

The official COVAR remarks state that text, logical expressions and empty cells are ignored in the given ranges. The documentation does not fully define whether and how unmatched nonnumeric positions are excluded as complete pairs. For dependable results, make two equal-length clean numeric columns, not two independently filtered columns with misaligned rows. Unlike a single-column variance function, covariance requires meaningful row-to-row pairing.

Availability, source and release scope

Zoho’s official statistical reference documents COVAR as of October 12, 2026. The checked individual page does not establish a first-supported version, minimum subscription tier or identical calculation behavior on web, iOS, Android, offline or desktop clients. The Zoho Sheet release history was consulted as context but does not justify inventing a COVAR debut date. No Zoho account or native application was used to execute the examples.

Create a worksheet with row 1 headers and values shown here. Confirm that columns A, B, C and D are numeric.

1. Click a blank cell, for example F2, and enter =COVAR(A2:A7;B2:B7).
2. Under the conventional population covariance model, the expected result is 12.833333333. A positive sign means the two variables tended to move together within this dataset.
3. In F3, use =COVAR(A2:A7;C2:C7). Because column C is 35 − B, the model predicts -12.833333333.
4. In F4, use the first three paired values: =COVAR(A2:A4;B2:B4). The population-model prediction is 1.333333333.
5. In F5, calculate a variable with itself: =COVAR(A2:A7;A2:A7). The population-model prediction is 11.666666667, the population variance of column A.

All four outputs are mathematical predictions independently verified in this package. None is presented as a screenshot, a logged Zoho execution or proof of the app’s denominator choice.

Why the model gives 12.833333333

For A, the mean is (2+4+6+8+10+12)/6 = 7. For B, the mean is (12+15+14+18+21+23)/6 = 103/6 ≈ 17.166666667.

Compute (A_i−7) × (B_i−103/6) for every matching row. The sum of these six deviation products is 77. For the conventional population covariance, divide by the six observations: 77/6 = 12.833333333. If instead you intend sample covariance, the conventional divisor is 6−1, giving 77/5 = 15.4. This distinction matters in statistical inference. Zoho’s individual COVAR help does not explicitly specify the divisor; that unverified aspect is prominent here rather than silently assumed.

When B is the response (dependent variable) and A is the explanatory (independent variable), the least-squares simple-regression slope is mathematically the covariance divided by the variance of A, provided the same valid paired observations and compatible denominators are used. On this clean numeric sample:

=COVAR(A2:A7;B2:B7)/VARP(A2:A7)

The independently calculated expected result, using population covariance and population variance, is 1.1. Here the model predicts that a one-unit increase in activity index corresponds to an estimated 1.1 additional qualified leads in the fitted linear relationship. For a direct Zoho regression calculation, its separately documented syntax puts the dependent y_values first:

=SLOPE(B2:B7;A2:A7)

The model also predicts 1.1 for this formula; Zoho’s official SLOPE reference explicitly documents the y_values; x_values order. The /VARP formula depends on COVAR’s denominator convention, which still requires native verification. The VARP guide explains population variance.

Covariance alone does not tell you association strength independently of the measurement units. If you multiply all entries in B by 100 (for example, switching from one scale of leads to another), the covariance also multiplies by 100, even though the underlying linear relationship is the same. Pearson correlation corrects for this scale effect by dividing covariance by the population standard deviations:

=COVAR(A2:A7;B2:B7)/(STDEVP(A2:A7)*STDEVP(B2:B7))

Under the conventional numeric-only population model, the result is approximately 0.965648520, the same coefficient that the independently evaluated =CORREL(A2:A7;B2:B7) gives. This is a mathematical identity under compatible conventions, not a claim that the native Zoho functions have been checked with mixed data. Compare with the prepared STDEVP guide for population standard deviation and the official CORREL reference for a unitless association score.

To avoid a frequent order error: COVAR(x; y) is symmetric mathematically on complete numeric pairs, but SLOPE(y_values; x_values) is not symmetric. Swapping SLOPE’s arguments changes the model being fitted.

The COVAR help page provides no native screenshot or step-by-step result for this exact review dataset. Avoid claiming test evidence where none exists. Preserve the original row pairing when cleaning text, blanks or Boolean entries; deleting cells independently from only one column can silently change your analysis.

Microsoft’s legacy COVAR documentation explicitly defines covariance as the average of paired deviation products, with the number of observations in the denominator. Zoho’s COVAR page provides official example results but no comparable explicit denominator formula. A matched set of native application tests is still necessary before declaring numerical parity. Google describes COVAR as covariance and directs readers to CORREL or PEARSON for normalized association; it does not establish Zoho’s behavior. The checked sources are linked above rather than relying on assumptions about a function name.

Related guides: VARP and STDEVP. Other directly checked primary sources: Zoho COVAR, Zoho CORREL, Zoho SLOPE and the Zoho functions catalogue.

Sources checked 2026-10-12. Documentation research: completed for the listed references. Independent mathematical calculations: completed and documented in the batch evidence. Native Zoho Sheet execution, platform-parity testing: not completed.