Standard Deviation Explained — What It Is and How to Calculate It

What Standard Deviation Actually Measures

Standard deviation quantifies the typical distance between a data point and the mean. If a dataset of exam scores has a mean of 70 and an SD of 10, most scores fall within 70 ± 10 — i.e., between 60 and 80. An SD of 25 would mean scores are much more spread out, with many students far above or below the mean.

Mathematically, SD is the square root of variance. Variance is the average squared deviation from the mean. We square the deviations to: (1) eliminate negative signs that would cancel out, and (2) penalise large deviations more heavily than small ones. Taking the square root brings the result back to the original units of measurement — so if data is in kilograms, the SD is also in kilograms, making it directly interpretable.

SD is expressed in the same units as the original data. This is its key advantage over variance (which is in squared units). When someone says "the average delivery time is 4 days with an SD of 0.5 days," you can immediately understand what that means.

Step-by-Step Calculation — Worked Example

Dataset: Five students scored 72, 85, 90, 68, 95 on a test. We will calculate the sample standard deviation.

Step 1 — Calculate the mean: x̄ = (72 + 85 + 90 + 68 + 95) / 5 = 410 / 5 = 82.

Step 2 — Find each deviation from the mean: 72−82=−10, 85−82=+3, 90−82=+8, 68−82=−14, 95−82=+13.

Step 3 — Square each deviation: (−10)²=100, (3)²=9, (8)²=64, (−14)²=196, (13)²=169.

Step 4 — Sum the squared deviations: 100 + 9 + 64 + 196 + 169 = 538.

Step 5 — Divide by n−1 (sample variance): 538 / (5−1) = 538 / 4 = 134.5.

Step 6 — Take the square root: s = √134.5 ≈ 11.6.

Interpretation: The typical student score differs from the mean (82) by about 11.6 points. Scores of 70–95 are all within one SD of the mean.

Population SD vs Sample SD — Which to Use

The key difference is the denominator: population SD divides by N (total population size); sample SD divides by n−1. The n−1 adjustment is called Bessel's correction.

Why n−1? When you calculate the mean from the same sample data you use to calculate deviations, you lose one degree of freedom — the deviations are constrained to sum to zero. Dividing by n−1 corrects for this, producing an unbiased estimate of the population variance.

Rule: Use population SD (σ, divide by N) only when your data represents the entire population — every member. Use sample SD (s, divide by n−1) when your data is a subset drawn from a larger group — which is nearly always the case in practice. Most calculators and software default to sample SD.

The Empirical Rule (68–95–99.7 Rule)

For data that follows a normal (bell-curve) distribution, standard deviation has a predictable relationship with the proportion of data within each interval:

• ±1 SD from the mean: approximately 68% of data falls here.

• ±2 SD from the mean: approximately 95% of data falls here.

• ±3 SD from the mean: approximately 99.7% of data falls here.

Example: IQ scores are designed with a mean of 100 and SD of 15. By the empirical rule: 68% of people have IQ between 85–115 (±1 SD), 95% have IQ between 70–130 (±2 SD), and 99.7% have IQ between 55–145 (±3 SD). A score above 145 is extremely rare (< 0.15% of the population). This rule only applies to approximately normal distributions — verify normality before applying it.

Standard Deviation vs Other Spread Measures

SD is not always the right measure of spread. The choice depends on your data type and distribution.

Range (max − min): Simple but severely distorted by a single outlier. Useful only as a rough first look.

Interquartile Range (IQR = Q3 − Q1): Covers the middle 50% of data. Robust to outliers. Better than SD for skewed distributions or when outliers are present.

Coefficient of Variation (CV = SD / mean × 100%): Expresses SD as a percentage of the mean. Use when comparing spread across datasets with different units or means — e.g., comparing volatility of stocks with different price levels.

Standard Error (SE = SD / √n): Measures the precision of the sample mean, not the spread of the data. Use SE for confidence intervals and hypothesis tests; use SD to describe the data itself.

Common Mistakes When Interpreting Standard Deviation

1. Confusing SD with Standard Error. SD describes how spread out data values are. SE describes how precise your estimate of the mean is. They answer different questions. SE shrinks as sample size grows; SD does not.

2. Using SD with heavily skewed data. SD is most meaningful for symmetric, approximately normal distributions. For skewed data (e.g., incomes, house prices), use median and IQR instead. A few extreme values can make the SD misleadingly large.

3. Dividing by n instead of n−1 for sample data. If you use n in the denominator for sample data, you get a biased (too small) estimate of the population SD. Always use n−1 for sample SD unless you truly have the entire population.

4. Treating a "small SD" as proof of accuracy. A dataset of all wrong answers that are consistently wrong (e.g., all measurements are 5 units too high) can have a small SD but be very inaccurate. SD measures precision (consistency), not accuracy (closeness to the true value).