VAR.S in Microsoft Excel: Sample Variance and Squared Units

Use VAR.S in Microsoft Excel to estimate variance from a sample, dividing the sum of squared deviations by n minus one. The working example below describes daily order totals; its result is measured in squared orders. The official reference supports the function’s identity and purpose.

This guide separates documentation review from application testing. The expected numbers are independently calculated, not screenshots or results captured from Microsoft Excel.

VAR.S(number1, [number2], ...)

Numeric inputs: individual numbers or cell ranges holding the observations. One range can contain many observations; a header is not a measurement.

Additional inputs: optional extra numeric values or ranges. Omit the square brackets shown in the syntax notation when entering a formula. The numeric-only examples avoid relying on text or logical-value conversion.

Use sample variance when the available observations represent a larger population and your analysis needs squared deviations. The n minus one denominator distinguishes it from population variance. This distinction is about the question being asked, not the number of rows or how many decimal places you display.

Variance is in squared input units. For measurements recorded in minutes, the result is in minutes squared, not minutes. Standard deviation is often easier to explain to a non-specialist because its units match the original data. Neither statistic alone establishes causation, a prediction interval or whether a particular observation should be discarded.

An operations workbook with one row per trading day provides the scenario. Enter the following numeric observations into A2:A9, one per row, with a header in A1. Use a spare result cell outside the input range.

Download the Microsoft Excel VAR.S practice data (CSV). Import it into a new, blank sheet starting at A1, with a comma as the file delimiter and the header row retained. The CSV contains input values only, not formulas or native test results. Blank spacer columns place the requested position in D2 and any additional test data in column F. Keep observations numeric rather than importing them as text.

Example 1: Measure spread in the working register

=VAR.S(A2:A9)

Expected result: 18514.285714 (displayed to six decimal places).

The numeric count is 8, the mean is 300, and the sum of squared deviations is 129600. Divide by 7 (n minus one); the result is a variance in squared measurement units.

Example 2: Check the denominator with three known numbers

=VAR.S(2,4,6)

Expected result: 4.

The mean is 4 and the squared deviations are 4, 0 and 4, for a total of 8. The sample denominator is 2. Keep the fractional result during later calculations rather than rounding the input values.

Example 3: Check squared units when values are rescaled

For this check, use F2:F5 containing 0, 0, 80, 80, in that order. These values are included in the practice file.

=VAR.S(F2:F5)

Expected result: 2133.333333 (displayed to six decimal places).

These four observations have mean 40 and squared deviations totalling 6400. They are ten times the numbers 0, 0, 8, 8, so their variance is one hundred times as large. Both zeros still count as observations.

Subtract the mean from each observation and square each difference. Add the squared differences and divide by n minus one. Do not take a square root when reporting variance; its units remain squared. This mathematical definition is used to check the numbers in the examples, independently of the spreadsheet software. Keep stored results at full available precision and round only the final presentation when a report requires it. The displayed expectations above are rounded where necessary; a different number of displayed decimal places is not by itself a calculation error.

The output looks much larger than the original measurements. Variance is expressed in squared units. Compare it with other variances using the same unit scale, not directly with an average expressed in the original units.

The sample has fewer than two numeric observations. The n minus one denominator is then unusable. Show that the data is insufficient rather than quietly substituting zero.

A very different value appears after an import. Check units, included records and numeric storage before changing the formula. A mixture of individual observations, totals and text-coded numbers is not a consistent statistical input. Do not average subgroup variances as though that automatically gives the combined variance.

The Microsoft function reference is the source for this guide’s Excel name and purpose. The reviewed Excel help distinguishes values supplied as arguments from values inside referenced arrays or ranges. Keep imported observations numeric rather than depending on automatic conversion of text or logical values.

For the daily order workbook, keep one comparable trading day per row. Exclude a grand-total row from the statistical input; it is not another trading day. When combining weeks, check that repeated dates are legitimate records rather than duplicated imports. The formula cannot make that distinction for you.

These examples use English function names and comma-separated arguments. A workbook configured for another regional separator may require semicolons. The data addresses stay the same. Test a copied formula on a small known input after an import or file conversion; this guide is not a certification of every Excel edition or exchange format.

For the same statistical question in other programs, see VAR.S in Google Sheets, VAR.S in ONLYOFFICE, VAR.S in SoftMaker PlanMaker. These links identify corresponding measures, not a guarantee that names, mixed-data rules, error codes or file conversions are identical.

MEDIAN, LARGE, SMALL, RANK.EQ, STDEV.S, STDEV.P, VAR.P. Browse the Microsoft Excel function library for the other published guides.

Editorial verification: The vendor documentation was reviewed on 2026-10-12. Scope: Microsoft Excel worksheet functions; the current Microsoft help entry is the reference, not a claim about every historical edition. Three numeric examples were checked with an independent arithmetic model. No native application run was performed. Single-value population edge cases, mixed-data coercion, exact error-code parity and round-trip workbook compatibility remain untested; they are not labelled application-verified.