Bias in statistics is a systematic error that causes estimates to deviate from the true population parameter in a consistent direction. An estimator is biased if its expected value differs from the parameter it estimates: Bias(θ̂) = E[θ̂] − θ. A biased estimator consistently overestimates or underestimates, regardless of sample size.
Bias enters at multiple stages. Sampling bias occurs when the sample does not represent the population (self-selection, convenience sampling, survivorship bias). Measurement bias occurs when the measurement process systematically distorts values. Estimation bias occurs when the statistical method itself introduces systematic error — using n rather than n − 1 in the variance formula, for example, produces a biased estimator.
An unbiased estimator has E[θ̂] = θ: on average, it hits the true value. However, unbiasedness is not always the most important property — a slightly biased estimator with lower variance may produce estimates closer to the truth in practice. The bias-variance tradeoff formalizes this: total error = bias² + variance, and the best estimator minimizes their sum, not bias alone.