Compare Models (Likelihood Ratio Test)
In regression models, the data analyst generally has to choose which of several statistical models most appropriately reflects the relationship between the response and the predictor variables. A standard step is choosing between two models of different complexity. The Compare Models dialog provides statistical tests that are appropriate for determining whether to select a more complex model with many predictors or a simpler model with a subset of the predictors. The Compare Models dialog may be used with many types of models, including most types of regression models.
To compare models
Create two or more statistical models, being sure to specify unique names in the Save As field on the Model page of the statistical dialog for each model. Choose Statistics Compare Models. The dialog shown below appears.
The Compare Models dialog has the following options:
Select Model
Model Objects
Select the model objects for comparison. The models that you created and saved appear in the dropdown list. An arbitrary number of models can be selected.
Name String Match
Enter a pattern used to restrict the list shown in the Model Object dropdown list. The symbol "*" matches any character. For example, to view all objects that begin with "last", enter last*. Use "[ ]" to denote a list of character options. For example, "model1", "model2", and "model3" match model[123], but "model4" does not.
Model Class
Select the class of models you want to do comparisons with. All models being compared must be from the same class.
Test Statistic
Select the test statistic to use for the model comparison. The test statistics that are available depend upon the Model Class selected.
Results
Enter the name for the object in which to save the results of the analysis.
Print Results
Select this to print out the results of the analysis in the designated output window.
S-Plus language functions related to Model Comparisons
anova, print.anova
Other related S-Plus language functions
aov, coxph, gam, glm, lm, lme, loess, nls, survreg