Definition
An estimator is unbiased for a parameter when its expected value over repeated samples equals that parameter. Unbiasedness is an average property and does not say that any one estimate is close. Bias, variance and mean squared error should be considered together.
How it is calculated
The expected value of estimator theta-hat over repeated samples equals the target theta.
| Symbol | Meaning | Unit |
|---|---|---|
| E | expectation over repeated samples | theta units |
| theta | target parameter | parameter units |
Interpretation
Imagine repeating the same sampling process many times and averaging the estimates. Unbiasedness says that long-run average lands on the target. Individual estimates can still be noisy, and a small controlled bias can sometimes reduce total error.
Assumptions and conventions
-
The expectation exists under the stated data-generating process.
-
The target parameter is fixed while samples vary.
Both estimator expectation and parameter have the same units.
Limitations and common misreadings
-
Unbiased does not mean accurate in a particular sample.
-
An unbiased estimator can have high variance.
References
- Larry Wasserman. (2004). All of Statistics: A Concise Course in Statistical Inference. Springer Texts in Statistics. Chapter 6, Models, Statistical Inference and Learning. doi:10.1007/978-0-387-21736-9. Source

