Model List Object

DESCRIPTION:

Objects of class model.list are typically produced as a side-effect of the model-fitting process in , and . They exist somewhere between the data contained in a object and the actual fitted model object.

GENERATION:

This class of objects is returned from fitting routines by specifying method=model.list in the calling sequence.

METHODS:

The following generic functions have methods for model.list:
, , .

INHERITANCE:

The class model.list inherits from .

VALUE:

All model lists have the same basic structure: a "data" component containing the data to be fit; a "formula" component containing the model formula; a "call" component containing the call generating the model list; and a "fit" component containing other data needed to perform the fit.

The components differ from one modeling function to another, but an important common sub-component of "fit" is "fitexpr", which contains the expression to actually compute the fit of the data contained in the model list. See the default method for a given modeling function to see the exact structure of the "fit" component.

The "data" component of a model list typically contains sub-components such as

X
the explanatory variables and their values
Y
the response variable values
weights
weights used in the fit, if any

The "data" component is further described below.

DETAILS:

The model.list object was created to support resampling functions such as , which may call model fitting routines such as thousands of times. The construction of a fitted model object from scratch follows three basic steps: constructing the model frame from the data; extracting the relevant pieces from the model frame; and performing the fit. The first two steps are redundant from the point of view of resampling. Therefore the data from the second step, along with other parameters and expressions needed to perform the fit, has been organized into the model.list object. During resampling, the model.list is created once, then repeatedly resampled and fit using model.list methods for fitting routines ( , for example).

A model list can be returned by fitting routines (without fitting) via the method argument.

The "data" component of a model.list object contains those data that are expected to be sampled with replacement by the resampling functions. All other non-sample-specific data should be in component "fit". This principle, which one should keep in mind when designing model lists for other modeling functions, is born out in the subscripting method for model.lists: the result of, for example ml[rind,], where ml is a model.list and rind a set of indices, is a model.list whose data component contains the rind selected rows of the original data components, and whose other components are unchanged.

SEE ALSO:

, , , , , , , ,