vcdExtra (version 0.9.6)

LRstats: Brief Summary of Model Fit for glm and loglm Models

Description

For glm objects, the print and summary methods give too much information if all one wants to see is a brief summary of model goodness of fit, and there is no easy way to display a compact comparison of model goodness of fit for a collection of models fit to the same data. All loglm models have equivalent glm forms, but the print and summary methods give quite different results.

Usage

LRstats(object, ...)
# S3 method for glmlist
LRstats(
  object,
  ...,
  saturated = NULL,
  sortby = NULL,
  label = c("name", "formula"),
  label.args = list()
)
# S3 method for loglmlist
LRstats(
  object,
  ...,
  saturated = NULL,
  sortby = NULL,
  label = c("name", "formula"),
  label.args = list()
)
# S3 method for default
LRstats(
  object,
  ...,
  saturated = NULL,
  sortby = NULL,
  label = c("name", "formula"),
  label.args = list()
)

Value

A data frame (also of class anova) with columns c("AIC", "BIC", "LR Chisq", "Df", "Pr(>Chisq)"). Row names are taken from the names of the model object(s) or their model formulas.

Arguments

object

a fitted model object for which there exists a logLik method to extract the corresponding log-likelihood

...

optionally more fitted model objects

saturated

saturated model log likelihood reference value (use 0 if deviance is not available)

sortby

either a numeric or character string specifying the column in the result by which the rows are sorted (in decreasing order)

label

character string specifying how to label the rows: "name" (default) uses the model object names; "formula" uses model formulas or bracket notation obtained from get_models (for glmlist and loglmlist objects) or get_model (for individual model objects passed to the default method).

label.args

a list of additional arguments passed to get_models when label = "formula". Useful arguments include abbrev (logical or integer) to abbreviate factor names and sep to change the separator in bracket notation.

Author

Achim Zeileis, Michael Friendly

Details

LRstats provides a brief summary for one or more models fit to the same dataset for which logLik and nobs methods exist (e.g., glm and loglm models).

The function relies on residual degrees of freedom for the LR chisq test being available in the model object. This is true for objects inheriting from lm, glm, loglm, polr and negbin.

See Also

logLik, glm, loglm,

logLik.loglm, modFit, get_models, get_model

Other glmlist functions: Kway(), get_model(), glmlist(), mosaic.glmlist()

Examples

data(Mental)
indep <- glm(Freq ~ mental+ses,
                family = poisson, data = Mental)
LRstats(indep)
Cscore <- as.numeric(Mental$ses)
Rscore <- as.numeric(Mental$mental)

coleff <- glm(Freq ~ mental + ses + Rscore:ses,
                family = poisson, data = Mental)
roweff <- glm(Freq ~ mental + ses + mental:Cscore,
                family = poisson, data = Mental)
linlin <- glm(Freq ~ mental + ses + Rscore:Cscore,
                family = poisson, data = Mental)

# compare models using object names (default)
LRstats(indep, coleff, roweff, linlin)

# compare models in a glmlist, using formula labels
mods <- glmlist(indep, coleff, roweff, linlin)
LRstats(mods, label = "formula")

# loglmlist example with bracket notation labels
data(Titanic)
tit.joint <- seq_loglm(Titanic, type = "joint")
LRstats(tit.joint)
LRstats(tit.joint, label = "formula")
LRstats(tit.joint, label = "formula", label.args = list(abbrev = TRUE))