glmglrt: glmglrt: GLRT P-Values in Generalized Linear Models

In glmglrt: GLRT P-Values in Generalized Linear Models

Description

This package has been developed to provide Generalized Likelihood Ratio Tests (GLRT) also known as Likelihood Ratio Tests (LRT) to Generalized Linear Models (GLMs). The stats package do support LRT P-values with anova and derived confidence intervals with confint(), but provides Wald's P-values with the summary function. This is unfortunate for two reasons: Wald's P-values may be inconsistent with profile-likelihood confidence intervals and Wald's P-values, on small samples are more biased than LRT P-values, for non-gaussian models. The anova function is not as simple as summary, since it requires manually fitting two models.

Summary function

This package provides a way to override (see override_summary) the standard summary.glm function by a summarylr function that provides LRT and/or Rao's score P-values.

Functions to estimate contrasts

It also provides functions estimate_contrast, confint_contrast and p_value_contrast to estimate contrasts of coefficients of GLMs with LRT, Rao's and Wald's hypothesis tests and confidence intervals This is an alternative to multcomp::glht without Wald's approximation ! It also provides a less powerful p_value.glm method for the S3 generic parameters::p_value. It also extends this S3 generic for a variety of models as p_value.default. That time, the only method supported for all models, is Wald's method.

glmglrt documentation built on Aug. 7, 2020, 9:10 a.m.