model selection table

Build a model selection table.

model.sel(object, ...)

# Default S3 method
model.sel(object, ..., rank = NULL, rank.args = NULL,
  beta = c("none", "sd", "partial.sd"), extra)
# S3 method for class 'model.selection'
model.sel(object, rank = NULL, rank.args = NULL, fit = NA,
  ..., beta = c("none", "sd", "partial.sd"), extra)

model.sel(x) <- value

Arguments

object,value: a fitted model object, a list of such objects, or a "model.selection" object.
...: more fitted model objects.
rank: optional, custom rank function (returning an information criterion) to use instead of the default AICc, e.g. QAIC or BIC, may be omitted if object is a model list returned by get.models.
rank.args: optional list of arguments for the rank function. If one is an expression, an x within it is substituted with a current model.
fit: logical, stating whether the model objects should be re-fitted if they are not stored in the "model.selection" object. Set to NA to re-fit the models only if this is needed. See ‘Details’.
beta: indicates whether and how the component models' coefficients should be standardized. See the argument's description in dredge.
extra: optional additional statistics to include in the result, provided as functions, function names or a list of such (best if named or quoted). See dredge for details.
x: a "model.selection" object.

Value

An object of class c("model.selection", "data.frame"), being a data.frame, where each row represents one model and columns contain useful information about each model: the coefficients, df, log-likelihood, the value of the information criterion used, Δ_IC and ‘Akaike weight’. If any arguments differ between the modelling function calls, the result will include additional columns showing them (except for formulas and some other arguments).

See model.selection.object for its structure.

Details

model.sel used with "model.selection" object will re-fit model objects, unless they are stored in object (in attribute "modelList"), if argument extra is provided, or the requested beta is different than object's "beta" attribute, or the new rank function cannot be applied directly to logLik objects, or new rank.args are given (unless argument fit = FALSE).

The replacement function appends new models to the existing "model.selection" object.

Author

Kamil Bartoń

Examples


Cement$X1 <- cut(Cement$X1, 3)
Cement$X2 <- cut(Cement$X2, 2)

fm1 <- glm(formula = y ~ X1 + X2 * X3, data = Cement)
fm2 <- update(fm1, . ~ . - X1 - X2)
fm3 <- update(fm1, . ~ . - X2 - X3)

## ranked with AICc by default
(msAICc <- model.sel(fm1, fm2, fm3))
#> Model selection table 
#>      (Int) X1 X2       X3 X2:X3 df  logLik  AICc delta weight
#> fm3  83.25  +                 +  6 -39.584 105.2  0.00  0.558
#> fm1  75.79  +  +  0.07826     +  7 -34.924 106.2  1.08  0.325
#> fm2 104.70       -1.39800     +  4 -47.653 108.3  3.14  0.116
#> Models ranked by AICc(x) 

## ranked with BIC
model.sel(fm1, fm2, fm3, rank = AIC, rank.args = alist(k = log(nobs(x))))
#> Model selection table 
#>      (Int) X1 X2       X3 X2:X3 df  logLik   AIC delta weight
#> fm1  75.79  +  +  0.07826     +  7 -34.924  87.8  0.00  0.967
#> fm3  83.25  +                 +  6 -39.584  94.6  6.75  0.033
#> fm2 104.70       -1.39800     +  4 -47.653 105.6 17.76  0.000
#> Models ranked by AIC(x, k = log(nobs(x))) 
# or
# model.sel(msAICc, rank = AIC, rank.args = alist(k = log(nobs(x))))
# or
# update(msAICc, rank = AIC, rank.args = alist(k = log(nobs(x))))

# appending new models:
model.sel(msAICc) <- update(fm1, . ~ 1)

Arguments

Value

Details

Author

See also

Examples