For any data set or probability distribution, Chebyshev's inequality guarantees that at least (1 − 1/k²) of observations lie within k standard deviations of the mean, where k > 1. Compute the exact minimum proportion, visualize the interval [μ − kσ, μ + kσ], and explore real‑world applications.
Chebyshev's theorem, also known as Chebyshev's inequality, is a fundamental result in probability theory and statistics. It provides a universal lower bound on the proportion of observations that lie within a specified number of standard deviations from the mean, irrespective of the underlying distribution shape — as long as the distribution has finite mean and variance.
For any random variable X with finite mean μ and finite variance σ², and for any real number k > 1:
\[ P(|X - \mu| \ge k\sigma) \le \frac{1}{k^2} \]
Equivalently: \[ P(|X - \mu| < k\sigma) \ge 1 - \frac{1}{k^2} \]
At least (1 − 1/k²) × 100% of the data lies within k standard deviations of the mean.
The inequality is named after the Russian mathematician Pafnuty Chebyshev (1821–1894), who first proved the result in its general form (though earlier versions appeared in the work of Bienaymé). Chebyshev's inequality is distribution‑free: it holds for any dataset or probability distribution with finite variance, making it one of the most robust tools in statistics. It does not assume normality, symmetry, or any specific shape, which is why it is often applied in quality control, finance, and non‑parametric inference. The bound is sharp — for certain discrete distributions (e.g., a two‑point distribution), equality can be approached arbitrarily.
Later mathematicians such as Markov and Cantelli extended the inequality, leading to one‑sided versions (Chebyshev–Cantelli inequality). Today Chebyshev's theorem is a cornerstone of probability theory and appears in the proof of the Weak Law of Large Numbers.
The proof relies on Markov's inequality applied to the non‑negative random variable (X−μ)². Let Y = (X−μ)², then E[Y] = σ². For any a > 0, Markov's inequality gives P(Y ≥ a) ≤ E[Y]/a. Set a = k²σ², then P((X−μ)² ≥ k²σ²) ≤ σ²/(k²σ²) = 1/k². Thus P(|X−μ| ≥ kσ) ≤ 1/k², and complement yields the desired bound. No additional assumptions about symmetry are required, which explains the theorem's remarkable generality. The inequality can be expressed in terms of proportions: at least (1 − 1/k²) of any dataset's values lie within k standard deviations of the mean.
Note: For k ≤ 1 the bound is trivial (since 1 − 1/k² ≤ 0). Therefore the theorem is meaningful only when k > 1.
| k (number of σ) | Minimum proportion inside [μ − kσ, μ + kσ] | Interpretation |
|---|---|---|
| 1.5 | ≥ 55.56% | At least 55.6% of data lies within 1.5σ of μ. |
| 2.0 | ≥ 75.00% | At least three‑quarters of observations. |
| 2.5 | ≥ 84.00% | Conservative bound for many real datasets. |
| 3.0 | ≥ 88.89% | Often used in outlier detection. |
| 4.0 | ≥ 93.75% | Very high guaranteed coverage. |
| 5.0 | ≥ 96.00% | Extreme bound, rarely needed. |
An e‑commerce platform analyzes daily transaction amounts. The mean purchase is $85 with standard deviation $22. Using Chebyshev's theorem with k = 2.5, we guarantee that at least 84% of transactions fall between $85 − 2.5×22 = $30 and $85 + 2.5×22 = $140. This interval is used to set fraud detection thresholds: any transaction below $30 or above $140 is flagged for review — even without knowing the distribution shape. The same logic helps in inventory management and budget forecasting.
The empirical rule applies only to normal distributions and gives sharper bounds: about 68% within 1σ, 95% within 2σ, 99.7% within 3σ. Chebyshev's theorem is much weaker but universally valid. For normally distributed data, Chebyshev's bound is conservative (for k=2, 75% vs actual 95%). For heavily skewed or leptokurtic distributions, Chebyshev may be the only usable bound. Use the empirical rule only after verifying normality; otherwise, rely on Chebyshev.