regrAllEqns: All Equations of a Linear Model
In regr: Building Regression Models

Description Usage Arguments Details Value Note Author(s) See Also Examples

Selection of best linear models by exhaustive search

regrAllEqns(formula, data, weights = NULL, nbest = 50, nvmax = 20,
  force.in = NULL, force.out = NULL, codes = NULL, really.big = FALSE,
  ...)
regrAllEqnsXtr(object, nbest=1, criterion="cp")
## S3 method for class 'regrAllEqns'
print(x, nbest = 20, criterion = "cp",
  printcriteria = FALSE, printcodes = TRUE, ...) 
## S3 method for class 'regrAllEqns'
plot(x, criterion = "cp", critrange = 10,
  minnumber = 10, nbest=10, codes = x$codes, ncharhorizontal = 6,
  col="blue", legend = TRUE, mar = 6, main = "", cex = 0.7*par("cex"),
  cex.lab = par("cex.lab"), ...)

`formula`	formula for the full model
`data`	data.frame where the variables are found
`weights`	weights for weighted LS estimation
`nbest`	number of equations to record for each size. For `print.regrAllEqns`, number of equations printed in total.
`nvmax`	maximum size of subsets to examine
`force.in`	formula for terms to be kept in all models
`force.out`	formula for terms to be excluded from all models
`codes`	codes for the terms to be used for identifying the equations in the result 'outmat' and in the plot. Preferably one letter each.
`really.big`	Must be TRUE to perform exhaustive search on more than 50 variables.
`criterion`	Criterion to be used to rank the equations and for the vertical axis in plotting.
`object, x`	result of `regrAllEqns` to be used by `regrAllEqnsXrt` or `print` or `plot`
`printcriteria`	logicel: should the table of criteria be printed?
`printcodes`	logical: should the short codes for the terms (columns in the primary output 'outmat') be listed?
`critrange`	range of criterion that determines how many equations will be used in plotting: those that are less than `critrange` worse than the best.
`minnumber`	minimum number of equations shown; overrides `critrange` if needed.
`ncharhorizontal`	maximum length of string on margin 3 to be shown horizontally. Use `ncharhorizontal=0` to get all codes shown vertically. This avoids overwriting.
`col`	color used for the equation identifiers in the plot
`legend`	logical: Should a list of short and long codes for terms be shown as a legend? Alternatively, a position for the legend: keyword like `"bottomleft"` or 2 coordinates.
`mar`	margins: either all 4 margins of the plot or 1 number for the top margin.
`main`	main title
`cex, cex.lab`	character expansion for codes in the plot and on the upper margin
`...`	Other arguments, to be passed to `regsubsets`, `print` or `plot`, respectively.

This is a wrapper function for function regsubsets od package leaps. When factors (and other multicolumn terms) occur in the full model, it returns models that either include the factor or exclude it, and avoids those that contain some, but not all dummy variables related to it.

regrAllEqns also provides convenient information for the the plotting method. The results of regsubsets and its summary, which include the "avoided" eaquations, are also contained in the value.

regrAllEqns returns a list, containing

`which`
`criteria`
`codes`
`force.in`
`force.out`
`outmat`
`allsubsets`
`obj`	see `?regsubsets`

regrAllEqnsXtr returns a formula or a list of formulas (if nbest>1). plot.regrAllEqns invisibly returns the coordinates of the models used in the plot.

For further details and notes, see ?regsubsets

Werner A. Stahel

regsubsets, leaps, step.regr,

data(d.birthrates, package="plgraphics")
t.formula <- fertility ~ catholic + single24 + single49 + eAgric + eIndustry +
eCommerce + eTransport + eAdmin + gradeHigh + gradeLow + educHigh +
bornLocal + bornForeign + altitude + language
## altitude and language are factors

r.ae <- regrAllEqns(t.formula, data=d.birthrates, nbest=100,
really.big=TRUE)
names(r.ae)
showd(r.ae$which)
plot(r.ae, mar=9, ncharhorizontal=0, main="birthrates example",
     legend="bottomright", xlim=c(4,20))

## extract the formula of the best model
( t.formula <- regrAllEqnsXtr(r.ae) )
## ... and fit it
regr(t.formula, data=d.birthrates)