Required packages for install (and their dependencies)
MASS
relaimpo
Rcmdr
quantreg
BMA


RegModel.1 = lm(depression~fatalism+simplicity, data=Ginzberg)
summary(RegModel.1)

#Variable Importance
library(relaimpo)
calc.relimp(RegModel.1, rela=TRUE) #Rela=T tells it to return the percentage of Rsq attributable to the predictor. I prefer this since I would be reporting a bias-adjusted Rsq not the original
boot.import=boot.relimp(depression ~ fatalism + simplicity, data = Ginzberg, b=500)
infer.import=booteval.relimp(boot.import, bty="bca")
infer.import

#Robust regression based on M-estimators
library(MASS)
RegModel.robust = rlm(depression~fatalism+simplicity, data=Ginzberg)
summary(RegModel.robust)
RegModel.robust$w

#Polynomial regression.  Tests the model dep predicted by fatalism + fatalism^2
RegModel.curve = lm(depression~poly(fatalism,degree=2), data=Ginzberg)
summary(RegModel.curve)

#Interaction (moderation)
RegModel.moderate = lm(depression~fatalism+simplicity+fatalism:simplicity, data=Ginzberg)
summary(RegModel.moderate)
library(effects)
plot(all.effects(RegModel.moderate), ask=FALSE) #this is Rcmdr code.  Simply do it by menu though.

#Quantile regression
library(quantreg)
RegModel.quant = rq(depression~fatalism+simplicity, tau=c(.1,.25,.5,.75,.9),data=Ginzberg)
summary(RegModel.quant)
plot(summary(RegModel.quant))
plot(summary(RegModel.quant), col=c("black","papayawhip")) #I don't like that default gray band

#Bayesian Model Averaging.  This will automatically dummy code factors also.
library(BMA)
world=read.table("http://www.unt.edu:8080/rss/class/mike/data/world.txt", header=T)

#with world data, took out categorical predictors due to low N and redundant predictors (e.g. took out literacy for male and female, kept overall literacy), though some are highly correlated still.
worldpred=world[,c(2:4,6,8:9,11,13,20:21)]
aidsrate=world[,17]
RegModel.bayes=bicreg(worldpred, aidsrate)
summary(RegModel.bayes)
imageplot.bma(RegModel.bayes)


#Ridge regression
library(MASS)
RegModel.ridge = lm.ridge(depression~fatalism+simplicity, data=Ginzberg)
RegModel.ridge$coef  #Compare to original model

#Lasso, a variation on the Ridge theme.  Don't ask, I just know about it but few details.  Requires data to be in matrix/vector rather than data frame form.
library(lars)
x=matrix(c(Ginzberg$simplicity,Ginzberg$fatalism), ncol=2)
RegModel.lasso=lars(x,Ginzberg$depression, type = "lasso")
RegModel.lasso
summary(RegModel.lasso)
coef(RegModel.lasso)