Interactive demo · First-year statistics · Estimation

Estimators on trial: bias, spread and the n − 1

An estimator is a recipe that turns data into a guess. To judge a recipe, repeat the study many times and look at the spread of its guesses: its sampling distribution. Each run draws a fresh sample of n values from a population we chose ourselves, so we know the truth (dashed gold line). Two competing recipes are compared: teal on top, pink below. The solid line in each panel is the average of its guesses: if it sits on the gold line, the recipe is unbiased.

Compare

Population

Sample size (n)

Run

Try these

Unbiased means right on average, not right every time. Among unbiased recipes, prefer the one with the smaller spread (standard error); in general compare the mean squared error, MSE = bias² + variance.

In the lecture notes