A distribution-free probabilistic bound that states for any random variable with finite mean μ and finite nonzero variance σ², the probability that the variable differs from its mean by at least k standard deviations is at most 1/k², for k>0.

Principle

A mean-variance tradeoff: variance controls the worst-case tail probability without any assumptions on distribution shape.

Demonstration

For a random variable X with mean μ and variance σ², applying Markov's inequality to the nonnegative variable (X−μ)² yields P(|X−μ| ≥ kσ) ≤ E[(X−μ)²]/(k²σ²) = 1/k². This holds whether X is continuous, discrete, or heavy-tailed so long as σ² is finite.

Misapplication

Using Chebyshev's inequality to claim tight tail behavior for small k or for distributions known to have light tails; or applying it when the variance is infinite or undefined, which makes the bound meaningless.

Consequence

Provides a simple, model-free upper bound on tail probabilities and a baseline guarantee for variance-based concentration; it is often used to justify conservative confidence intervals when little is known about the distribution.

Reversal

Inverting the claim gives that rapid tail decay (for example exponential tails) implies much stronger bounds than 1/k²; thus Chebyshev is the weak, distribution-free end of a spectrum of concentration results.

Boundary

Requires finite variance and a meaningful mean; it does not provide sharp or informative bounds for small k, for heavy-tailed variables with infinite variance, or when moment information of higher order is available.

Semantic Tension

Competes with sharper concentration inequalities (Hoeffding, Chernoff, sub-Gaussian bounds) that assume more structure; Chebyshev trades strength of conclusion for minimal assumptions.

Synthesis

Chebyshev's inequality ties a single second-moment statistic to a guaranteed upper bound on tail probability: it is the universal, weakest concentration bound that follows from knowing only mean and variance.