Description
Compare how well data are explained under a null and an alternative by the ratio of their maximised likelihoods, rejecting the null when the ratio is small.
A likelihood ratio test compares the likelihood of the data at parameter values allowed by each hypothesis. For two simple hypotheses the statistic is the likelihood under the null divided by that under the alternative, and the null is rejected below a critical value chosen for the desired Type I error; the Neyman-Pearson lemma shows this is the most powerful test at that level. The critical region is usually rewritten in terms of a simpler statistic with a known distribution; with discrete data exact levels such as 0.05 may be unattainable. For composite hypotheses the generalised test uses the maximum likelihood under the null restriction over the unrestricted maximum, and in large samples minus twice the log ratio is approximately chi-square with degrees of freedom equal to the number of restricted parameters. This is the machinery behind comparing nested models, for instance a logistic regression with and without a design factor.
Attributes
- Discipline
- Statistics & Modelling
- Loss type
- Latent
- Stage
- Evaluation, Post-launch
- Data
- Quantitative
- Effort
- Hours
Steps
- Write the likelihood for the model of the data.
- Maximise it under the null restriction and without restriction (or evaluate it at the two simple hypotheses).
- Form the ratio of the two maximised likelihoods.
- Find the critical value from the exact distribution of an equivalent statistic, or compare minus twice the log ratio with a chi-square distribution whose degrees of freedom equal the number of restricted parameters.
- Reject the null when the ratio is small (minus twice the log ratio large) and report the p-value with the effect estimates.
Sources
- Mathematical Statistics with Resampling and R (Chihara & Hesterberg, 2011)
- Statistical Inference (Casella & Berger, 2001)
- Applied Logistic Regression (Hosmer & Lemeshow, 2000)
- Nonlinear Regression Analysis and Its Applications (Bates & Watts, 1988)
- Nonlinear Regression (Seber & Wild, 1989)
- Statistical Methods for Reliability Data (Meeker & Escobar, 1998)
- Statistical Methods for Survival Data Analysis (Lee & Wang, 2003)
- A Beginner's Guide to Structural Equation Modeling (Schumacker & Lomax, 2004)
- Latent Variable Models (Loehlin & Beaujean, 2017)
- An Introduction to Categorical Data Analysis (Agresti, 1996)
- Linear Statistical Inference and Its Applications (Rao, 1973)
- Modes of Parametric Statistical Inference (Geisser, 2006)
- Loss Models: From Data to Decisions (Klugman et al., 2012)
Also known as
- LRT
- Generalised likelihood ratio test
- GLRT
- Neyman-Pearson test
- Asymptotic likelihood ratio test
Categories
- Analytics & Behaviour
- Statistics & Modelling