Statistics Glossary → Score (Lagrange multiplier) test
Score (Lagrange multiplier) testconcept
A test of whether a parameter restriction (e.g. coefficient = 0) holds that fits only the restricted model and uses the slope of the likelihood at that point.
Maximum likelihood estimation has three classical tests: the likelihood ratio test fits both models and compares their likelihoods; the Wald test fits only the larger model and looks at how far the coefficient is from zero relative to its standard error; the score test fits only the restricted (smaller) model. If the restriction is true, the slope of the log-likelihood (the score) at the restricted estimate should be close to zero. The score test looks at the size of this slope scaled by the information matrix: the statistic U'I⁻¹U follows, in large samples, a chi-square distribution with degrees of freedom equal to the number of restrictions.
The three tests are equivalent in large samples but can disagree in small ones. The score test's advantage is that it can assess the contribution of a variable not yet in the model, or a parameter not yet freed, without refitting; that is why it is used in stepwise variable selection (the entry step) and for modification indices in structural equation models. The log-rank test is in fact a score test derived from the Cox model. In econometrics the same test is known as the Lagrange multiplier (LM) test.
An everyday example
A cyclist does not need to ride the whole road to know whether she is at the top of a hill: she looks at the slope where she is standing. If it is almost flat she is near the top; if it is steep there is still climbing left. The score test likewise looks at the 'slope' of the likelihood at the restricted model to decide whether lifting the restriction would raise the likelihood meaningfully.
In MerQur
In structural equation modeling, 'Extra output: Modification indices' gives the score (LM) test and EPC for fixed-zero candidates; the measurement invariance output of structural equation modeling and confirmatory factor analysis includes score tests. In logistic regression the entry criterion of stepwise selection is the Rao score test; in Cox regression variable entry uses the score test, the proportional hazards test of Cox and time-dependent Cox regression is the Grambsch-Therneau score test, and the score χ² is also reported in the model test of time-dependent Cox regression.
Analyses where this concept appears
Logistic RegressionConfirmatory Factor Analysis (CFA)Structural Equation Modeling (SEM)Cox Regression (Proportional Hazards Model)Time-dependent Cox Regression
Related concepts
Likelihood ratio test (LRT)Wald test and Wald confidence intervalModification indices