Result that twice the log likelihood ratio approaches a chi-squared distribution, which turns fits into significance.Wilks' theorem is a cornerstone of statistical inference, stating that under certain regularity conditions, twice the log of the likelihood ratio between two nested models converges to a chi-squared distribution as the sample size grows. The number of degrees of freedom equals the difference in the number of free parameters between the models. This theorem turns a complex likelihood comparison into a simple significance test, making it indispensable in fields where data are expensive and hypotheses are precise, like particle physics.
The theorem was proved by Samuel S. Wilks in 1938, in a paper titled 'The Large-Sample Distribution of the Likelihood Ratio for Testing Composite Hypotheses'.
The theorem relies on the fact that, near the true parameter values, the log-likelihood surface is approximately quadratic (a parabolic bowl). For nested models, the difference in their maximum log-likelihoods then behaves like the sum of squares of independent standard normal variables, which is exactly a chi-squared distribution. This approximation improves with more data, so in practice it works well for large samples, though it can fail near parameter boundaries or when the true parameters are at the edge of the allowed space.
It is used to compute significance levels for new particle discoveries at the Large Hadron Collider, where the 'local p-value' is derived from a chi-squared distribution via Wilks' theorem.
Despite its ubiquity, Wilks' theorem is an asymptotic result — it assumes infinite data — and in real experiments with limited statistics, physicists often use Monte Carlo simulations instead of the theorem.



