Introduction

Variance is a statistical measure that quantifies how much the values of a random variable or dataset spread out from their mean. It captures the average of the squared deviations from the mean, providing a foundational measure of variability that underlies many other statistical concepts, including standard deviation.

Variance is widely used in statistics, finance, and machine learning to understand data spread, assess risk, and evaluate the consistency of a process or model.

Why is Variance Important?

Variance helps to:

• Quantify how spread out data values are from the mean
• Assess risk and volatility in finance and investment analysis
• Support model evaluation through the bias-variance tradeoff in machine learning
• Form the mathematical basis for standard deviation
• Compare variability across different datasets or distributions
• Guide quality control by measuring consistency in processes

Calculating Variance

Whiteboard

Loading diagram...

Variance Formula

Population Variance

σ² = Σ(xᵢ - μ)² / N

Sample Variance

s² = Σ(xᵢ - x̄)² / (n - 1)

where μ = population mean, x̄ = sample mean, N = population size, and n = sample size.

Step-by-Step Example

Data: 2, 4, 4, 4, 5, 5, 7, 9

Step 1: Mean = (2+4+4+4+5+5+7+9) / 8 = 5

Step 2 & 3: Squared deviations from mean
(2-5)²=9  (4-5)²=1  (4-5)²=1  (4-5)²=1
(5-5)²=0  (5-5)²=0  (7-5)²=4  (9-5)²=16

Step 4: Variance = (9+1+1+1+0+0+4+16) / 8 = 32/8 = 4

The variance of this dataset is 4 (in squared units of the original data).

Variance of a Random Variable

Variance can also be expressed in terms of expectation:

Var(X) = E[(X - E(X))²]

This can be simplified to a commonly used computational formula:

Var(X) = E(X²) - [E(X)]²

Properties of Variance

Scaling

Var(aX) = a² × Var(X)

where a is a constant.

Adding a Constant

Var(X + b) = Var(X)

Adding a constant shifts the data but doesn't change its spread.

Sum of Independent Variables

Var(X + Y) = Var(X) + Var(Y),   if X and Y are independent

Variance vs Standard Deviation

Key Properties of Variance

• Variance is always zero or positive; it can never be negative.
• A variance of 0 means all data points are identical (no spread).
• Larger variance indicates greater spread or variability in the data.
• Variance is sensitive to outliers, since deviations are squared.
• Adding a constant to all values doesn't change variance; multiplying does.

Where is Variance Used?

Advantages

• Provides a foundational measure of data spread and variability
• Forms the mathematical basis for standard deviation and other statistics
• Useful for comparing consistency across different processes or models
• Central to key machine learning concepts like the bias-variance tradeoff
• Applicable to both population and sample data

Limitations

• Expressed in squared units, making direct interpretation harder than standard deviation
• Highly sensitive to outliers due to the squaring of deviations
• Assumes the mean is a meaningful measure of center for the data
• Can be misleading for skewed or non-normal distributions
• Requires calculating the mean first, which itself can be affected by outliers

Real-World Examples

Best Practices

• Use sample variance (dividing by n-1) when working with a sample rather than a full population.
• Pair variance with the mean for meaningful interpretation of a dataset.
• Convert to standard deviation when communicating spread in the same unit as the original data.
• Check for outliers before interpreting variance, as they can heavily skew results.
• In machine learning, balance variance against bias to avoid overfitting or underfitting.

Interview Tip

A common interview question is:

"What is variance, and how does it relate to the bias-variance tradeoff in machine learning?"

A strong answer is:

Variance measures how much data values spread out from the mean, calculated as the average of squared deviations from the mean. In machine learning, variance refers to how much a model's predictions change when trained on different subsets of data — high variance often indicates overfitting, where the model captures noise rather than the underlying pattern. This is balanced against bias, which measures how far off a model's predictions are on average, forming the well-known bias-variance tradeoff central to model evaluation.

Mentioning the connection to overfitting and the bias-variance tradeoff makes your answer stronger.

Conclusion

Variance is a foundational statistical measure that quantifies the spread of data around its mean, forming the basis for standard deviation and playing a central role in risk assessment, quality control, and machine learning model evaluation. Understanding variance, alongside its relationship to bias in the bias-variance tradeoff, is essential for building reliable statistical and machine learning models.