Fit dot Models

Description

Fit a statistical model using different estimators (e.g., robust and least-squares) and/or combine fitted models into a single object. Generic methods then produce side-by-side comparisons of the parameter estimates and diagnostic plots.

Usage

  fit.models(model.list, ...)

Arguments

Details

There are two distinct ways the fit.models function can be used.

The first is to fit the same model using different estimators. In this case, model.list should be a character vector or a list where each element is the name of a modeling function and the remaining arguments (in ...) are the common arguments to the functions in model.list. For example, the following command fits robust and least-squares linear models to Brownlee's Stack Loss Plant Data.

    fit.models(c("rlm", "lm"), stack.loss ~ ., data = stackloss)

The resulting fit.models object is then a list with the output of

    rlm(stack.loss ~ ., data = stackloss)

in the first element and

    lm(stack.loss ~ ., data = stackloss)

in the second. The class attribute of the returned list is set (in this case) to "lmfm" which is the fit.models class (fmclass) for comparing linear-model-like fits.

The second use of fit.models is to combine already fitted model objects. In this case fit.models combines its arguments into a fit.models object (a list where element i is occupied by argument i) and sets the class attribute to the most appropriate fit.models class.

Value

The returned object is a list containing the fitted models. The class of the retuned object depends on the classes of the model objects it contains.

See Also

fmclass.add.class for adding a class to an existing fit.models class and fmclass.register to create a new fit.models class.

Examples

  data(stackloss)

  # First, use fit.models to fit robust and least-squares linear
  # regression models to the Brownlee's Stack Loss Plant Data.

  # Step 1: rlm (robust linear model) is in the MASS package.
  require(MASS)

  # Step 2: tell fit.models rlm can be compared to lm
  fmclass.add.class("lmfm", "rlm")

  fm1 <- fit.models(c("rlm", "lm"), stack.loss ~ ., data = stackloss)

  summary(fm1) #rlm does not provide p-values or Multiple R-squared
  plot(fm1)


  # Second, use fit.models to combine already fitted models into a
  # fit.models object.

  lm.complete <- lm(stack.loss ~ ., data = stackloss)
  lm.clean <- lm(stack.loss ~ ., data = stackloss, subset = 5:20)

  fm2 <- fit.models(lm.clean, lm.complete)

  summary(fm2)
  plot(fm2)


  # Name the models in the fit.models object.
  
  fm3 <- fit.models(c(Robust = "rlm", "Least-Squares" = "lm"),
                    stack.loss ~ ., data = stackloss)

  fm4 <- fit.models(Clean = lm.clean, Complete = lm.complete)

Results