regr: Fitting Regression Models
In regr: Building Regression Models

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

'regr' fits regression models of various types: ordinary, robust, quantile, Tobit, logistic, generalized linear, multinomial, ordered response, multivariate, ...

It is a wrapper function which calls the respective S fitting functions and yields an extensive result, to be used by the respective print and plot methods.

regr(formula, data=NULL, family=NULL, robust = FALSE, method=NULL, 
     nonlinear = FALSE, start=NULL, subset=NULL, weights=NULL, offset=NULL, ...)

## S3 method for class 'regr'
summary(object, ...)

`formula`	a symbolic description of the model to be fit.
`data`	data frame containing the variables in the model.
`family`	character string describing the type of model and the fitting procedure. By default, `regr` will fit a model that is suitable for the response variable, if possible. Can be either "gaussian" for ordinary regression, including multivariate response, "binomial" for logistic regression, "poisson" for Poisson regression, "multinomial" for multinomial regression, "polr" for ordered response (cumulative logits) regression, "weibull", "lognormal", "loglogistic", "extreme" for survival regression with the respective distribution, "ph" for Cox regression, or a suitable argument for `glm`, i.e.: A description of the error distribution and link function to be used in the model. This can be a character string naming a family function, a family function or the result of a call to a family function. (See 'family' for details of family functions.)
`nonlinear`	if TRUE, a non-linear model is expected, to be fitted by `nlm`
`start`	if `nonlin` is `TRUE`, starting values for `nlm`
`robust`	if TRUE, robust fitting is used if a function is available, see Details.
`method`	"rq" or "quantreg" requires fitting by quantile regression. Then, the argument `tau` is used for setting the quantile: `regr(..., method="quantreg", tau=0.9)`. Otherwise, if suitable, a method argument that is passed to the fitting procedure, see also Details.

`subset`	an optional vector specifying a subset of observations to be used in the fitting process. Names of variables in `data` can be used to generate the vector.
`weights`	an optional vector of weights to be used for fitting.
`offset`	sometimes used to help specify the model.
`...`	additional arguments, which are passed to `regr.control`, see `?regr.control`
`object`	'summary.regr' simply returns its only argument `object`.

Contrasts: If (ordered) factors are generated in the formula (like y~ordered(f1)), the contrast functions contr.wsum and contr.wpoly will be equivalent to their unweighted versions.

Robust fitting is available for ordinary linear models, by setting robust = TRUE. By default, the function lmrob from package robustbase is called with the argument setting="KS2011". This default will be changed in future versions. Alternative methods can be obtained by setting the method argument. Choosing method="rlm" calls function rlm from package MASS with default method="MM". Any further elements of method will be passed on to the methods argument of the called function.

The 'summary' generic function does nothing to 'regr' objects, since all useful information is already collected in the object.

Logical explanatory variables are converted into numeric, in contrast to the official R fitting functions, which treat them as factors.

regr returns a list object of class regr and secondary class as produced by the fitting function. The components include generally those of the results of the fitting function and its summary. The important additional components are:

`termtable`	data.frame, table for testing terms both for single and mulitple degree terms (continuous or binary explanatory variables and factors). The columns are The `print` method for the whole `regr` object will by default not show all of them.
`termeffects`	All coefficients. This is a list with a component for each term. Each component characterizes the coefficients similar to the `termtable` just explained, but without the columns `df, stcoef, R2.x`. Note that the interpretation of the coefficients of dummy variables depends on the contrasts used for the factor. Tests need to be interpreted correspondingly, and with caution.
`stdresiduals`	Standardized residuals
`sigma`	estimated standard deviation for normal errors, `sqrt(dispersion)` for `glm`
`leverage`	leverage values
`allvars`	data.frame: variables used for fitting the model
`binfac`	levels of binary factors
`fitfun, funcall`	R function and call that has been used for fitting the model

Werner A. Stahel, Seminar for Statistics, ETH Zurich

lm, glm, rlm, multinom, polr,...

data(d.blast, package="plgraphics")
( r.blast <-
  regr(log10(tremor)~location+log10(distance)+log10(charge), data=d.blast) )

## Anorexia
data(anorexia, package="MASS")
r.anorexia <- regr(Postwt ~ Prewt + Treat + offset(Prewt),
                   data = anorexia)

## Annette Dobson (1990) "An Introduction to Generalized Linear Models".
     ## Page 9: Plant Weight Data.
ctl <- c(4.17,5.58,5.18,6.11,4.50,4.61,5.17,4.53,5.33,5.14)
trt <- c(4.81,4.17,4.41,3.59,5.87,3.83,6.03,4.89,4.32,4.69)
d.dob <- data.frame(group = gl(2,10,20, labels=c("Ctl","Trt")),
                    weight = c(ctl, trt))
(r.dob <- regr(weight ~ group, data=d.dob))

## multinomial regression
data(d.surveyenvir)
d.surveyenvir$dist.unordered <- factor(as.character(d.surveyenvir$disturbance))
t.r <- regr(dist.unordered~age+education+location, data=d.surveyenvir)

## ordered regression
data(housing, package="MASS")
t.r <- regr(Sat ~ Infl + Type + Cont, weights = Freq, data = housing)
plot(t.r)

## multivariate regression
data(d.fossileSamples, package="plgraphics")
r.mregr <-
  regr(cbind(sAngle,lLength,rWidth)~SST+Salinity+lChlorophyll+Region+N,
                data=d.fossileSamples)