Interactive demo · First-year statistics · Estimation

Maximum likelihood: climbing the likelihood

The likelihood scores every possible parameter value by how probable it makes the data we actually saw. The top panel shows the log-likelihood of the current sample, shifted so that its top is at 0: the top is the maximum likelihood estimate (MLE), the gold dashed line is the truth (we chose it), and the dashed blue parabola uses the curvature at the top, which gives the standard error. Draw new samples to watch the top move; the bottom panel collects the MLEs of all samples, its sampling distribution, against the Cramér–Rao benchmark.

Model

Sample size (n)

Run

Scoreboard

Try these

The likelihood interval (values within 1.92 of the top) is an approximate 95% confidence interval in regular models: 1.92 is half of 3.84, the 95% point of chi-squared with 1 degree of freedom.

In the lecture notes