drop1

Drop1 is a statistical utility function found in the R programming language, part of the stats package. It is used to assess the effect of removing individual terms from a fitted model, aiding in model simplification and selection.

How it works: Drop1 tests remove each term in turn and compares the fit of the reduced model with the full model. The comparison uses an appropriate statistical test, typically an F-test for linear models and a likelihood-based test (such as a chi-square test) for generalized linear models. The function returns a summary that includes the term being tested, the test statistic, degrees of freedom, and the p-value, indicating whether dropping the term would meaningfully worsen the fit.

Scope and usage: The user can specify which terms to consider dropping via a scope specification. If scope is not restricted, drop1 evaluates all terms in the model except the response. In practice, it is common to run drop1 on a linear model to identify terms that may be unnecessary, contributing to a more parsimonious model. The results help guide decisions about simplifying the model or proceeding to more automated stepwise procedures.

Output and interpretation: The result is typically presented as a table listing each term with its corresponding test statistic, degrees of freedom, and p-value. Small p-values suggest that dropping the term significantly worsens the model, while large p-values suggest the term may be removed without substantial loss of fit.

Relation to other methods: Drop1 is a tool for one-term-at-a-time evaluation and is often used alongside other model selection approaches, such as stepwise selection or information-criterion-based methods.