Overview

In statistics an estimator is a procedure or rule for using observed data to compute an approximate value of an unknown quantity of interest (the estimand). The numerical result produced by applying the rule to a particular data set is called an estimate. These three terms — estimator, estimand and estimate — refer respectively to the method, the target parameter and the realized value.

Key properties

Evaluating estimators relies on properties that describe their behavior across repeated samples. Important properties include:

  • Bias: the difference between the estimator's average value and the true parameter. An estimator with zero bias is called unbiased.
  • Variance: how much the estimator varies from sample to sample.
  • Mean squared error (MSE): combines bias and variance and is often used as an overall measure of quality.
  • Consistency: whether the estimator converges to the true value as sample size grows.
  • Efficiency: relative variance among unbiased estimators; an efficient estimator has low variance.
  • Robustness: sensitivity to departures from model assumptions.

Common methods and examples

Several general methods produce widely used estimators. The sample mean and sample variance are canonical point estimators for population mean and variance under independent sampling. The maximum likelihood estimator (MLE) chooses parameter values that make the observed data most probable under a specified model. The method of moments matches sample moments to theoretical moments. Bayesian estimation uses the posterior distribution to produce point summaries like the posterior mean or median. Shrinkage estimators, such as those inspired by James and Stein, trade bias for lower variance in multivariate settings.

Uses and distinctions

Estimators are central to applied statistics, scientific inference and decision making. They appear as point estimators (single values) or within procedures that produce interval estimates and hypothesis tests. Practical choice of an estimator depends on goals, sample size, model assumptions and loss functions: methods optimal for one criterion (e.g., unbiasedness) may be inferior under another (e.g., MSE).

Practical considerations and further reading

In applied work, diagnostically checking assumptions, computing standard errors, and assessing sensitivity are routine steps. Different fields emphasize different estimators and trade-offs; for a general introduction see a standard statistics resource. Understanding the distinction between estimator, estimand and estimate helps avoid common confusion in reporting and interpreting results.