The atomic probability measure that places equal mass 1/n at each of n observed sample points, typically written μ_n = (1/n)∑_{i=1}^n δ_{x_i}, and serving as a nonparametric estimator of the underlying distribution.

Principle

Replace an unknown population distribution by the discrete distribution supported on the sample to estimate expectations and law-dependent quantities by simple averages of point evaluations.

Demonstration

Given i.i.d. samples x_1,…,x_n from a distribution P on a measurable space, the empirical measure μ_n assigns mass 1/n to each x_i; for any bounded continuous test function f, ∫ f dμ_n = (1/n)∑ f(x_i).

Misapplication

Treating the empirical measure as if it were a smooth density without smoothing or kernelization; using μ_n directly for inference problems that require absolute continuity leads to invalid likelihood evaluations.

Consequence

As sample size grows the empirical measure concentrates around the true law so that sample averages approximate population expectations; it forms the basis of bootstrap resampling and many nonparametric procedures.

Reversal

A smoothed density estimator (e.g., kernel convolution of the empirical measure) replaces point masses by localized continuous kernels to produce a density with respect to a reference measure.

Boundary

Defined for finite samples on measurable spaces and as an estimator it does not encode parametric model structure; it excludes smoothed or regularized versions unless explicitly convolved with kernels.

Semantic Tension

Contrasts with parametric estimators: empirical measures make minimal modelling assumptions and capture the sample exactly, whereas parametric fits introduce structure that can reduce variance at the cost of potential bias.

Synthesis

The empirical measure is the simplest nonparametric representation of sampled data, an atomic probability measure whose integrals reproduce sample averages and which converges to the generating distribution with growing sample size.