Variance Explained — What It Measures and Why It Matters

What Variance Actually Measures

Variance measures the average squared deviation of each value from the mean. "Squared deviation" is key: we square the differences so that positive and negative deviations do not cancel each other out, and to penalise large deviations more heavily than small ones.

The squaring has one downside: the units of variance are the square of the original units. If your data is in metres, the variance is in metres². If your data is in dollars, variance is in dollars². This makes variance hard to interpret directly — which is why we typically take the square root to get standard deviation, which is back in the original units.

Despite the unit problem, variance has important mathematical properties that make it the preferred measure in theoretical statistics. Variance is additive for independent variables: if X and Y are independent, Var(X + Y) = Var(X) + Var(Y). Standard deviation does not have this property, which is why ANOVA (Analysis of Variance) works with sums of squares rather than sums of deviations.

Worked Example — Step by Step

Dataset: Six monthly returns for an investment fund: 3%, 7%, −2%, 5%, 8%, 1%. We will calculate both population and sample variance.

Step 1 — Calculate the mean: x̄ = (3 + 7 + (−2) + 5 + 8 + 1) / 6 = 22 / 6 ≈ 3.67%.

Step 2 — Find each deviation from the mean:

3 − 3.67 = −0.67 | 7 − 3.67 = 3.33 | −2 − 3.67 = −5.67 | 5 − 3.67 = 1.33 | 8 − 3.67 = 4.33 | 1 − 3.67 = −2.67.

Step 3 — Square each deviation:

0.45 | 11.09 | 32.15 | 1.77 | 18.75 | 7.13.

Step 4 — Sum: 0.45 + 11.09 + 32.15 + 1.77 + 18.75 + 7.13 = 71.33.

Population variance (σ²): 71.33 / 6 ≈ 11.89 (%²).

Sample variance (s²): 71.33 / 5 ≈ 14.27 (%²).

Sample SD: s = √14.27 ≈ 3.78%.

Interpretation: Monthly returns typically deviate from the average (3.67%) by about 3.78 percentage points. This is the volatility of the fund.

Why Sample Variance Uses n−1 (Bessel's Correction)

When estimating population variance from a sample, dividing by n produces a biased estimator — it systematically underestimates the true population variance. The reason: when you calculate the sample mean x̄ from the same data, you use up one degree of freedom. The n deviations (xᵢ − x̄) are constrained to sum to zero, so only n−1 of them are free to vary independently.

Dividing by n−1 instead of n corrects this bias. The resulting s² is an unbiased estimator of σ² — meaning that if you drew many samples and calculated s² for each, the average of all those s² values would equal the true σ².

For large samples, the difference between dividing by n and n−1 is negligible. But for small samples (n < 20), the correction matters. At n = 5, dividing by n underestimates variance by 20%; at n = 10, by 10%. Most statistical software defaults to n−1 (sample variance).

Variance vs Standard Deviation — When to Use Each

Standard deviation is the square root of variance. Both measure spread, but they serve different purposes.

Use standard deviation for reporting and communication. SD is in the same units as the data, making it directly interpretable. "The mean salary is $65,000 with an SD of $12,000" is immediately meaningful.

Use variance for mathematical and statistical analysis. Many formulas and statistical tests are built on variance (not SD) because of its additive property and its role in the F-distribution. ANOVA decomposes total variance into between-group and within-group components. The coefficient of determination R² in regression is defined in terms of variance.

Real-World Applications of Variance

Finance and investing: Variance (or SD) of returns is the primary definition of financial risk. Portfolio theory (Markowitz) optimises the variance of a portfolio of assets. The F-test for comparing investment fund volatility uses variance ratios.

Quality control: Manufacturing processes use variance to track consistency. Control charts (SPC) monitor whether process variance stays within acceptable limits. Six Sigma methodology aims to reduce process variance to achieve fewer than 3.4 defects per million opportunities.

ANOVA (Analysis of Variance): This fundamental statistical test compares means across three or more groups by decomposing total variance into components. If between-group variance is much larger than within-group variance (F ratio is large), the groups differ significantly.

Regression analysis: R² (coefficient of determination) measures the proportion of the total variance in the outcome variable that is explained by the predictor variables. An R² of 0.80 means the model explains 80% of the variance in the outcome.