Likelihood-Ratio Test

Likelihood Ratio test (often termed as LR test) is a test to compare between two models, concentrating on the improvement with respect to likelihood value. If we keep on adding predictor variables to a linear model, R-square will improve. This holds true for model likelihood value as well. But, the objective is to check if the improvement on likelihood is good enough or not!

This test theoretically can be applied to any kind of regression model. In the presence and simplicity of R2 and Adjusted R2 statistics, it is not generally used for linear regression model and used only for non-linear models.

Application & Interpretation:

In ATH you can run LR Test for various kinds of generalised linear regression model. You need to specify the target variable and the family or link function. You have 8 various options, e.g. binomial (for logistic regression), poisson (for poisson regression), gamma (for gamma regression), etc.

The function tests the significance on improvement on likelihood value due to regression in comparison with no model, i.e. no predictor variable.

Hence, if we compare with the linear regression (refer to “Linear Regression” help), it is similar to the F-statistic table, which tests if the target variable can be predicted using the selected set of variables.

Input and Output:

To run Likelihood-Ratio function, you can log in to Analyttica TreasureHunt and should select variables from the dataset and apply.

See Also:

Kolmogorov Smirnov Diagnostics, Bayesian Information Criterion, Hosmer-Lemeshow Goodness-of-Fit Test.