library(MASS)
	attach(wtloss)	#
#----------------LINEAR MODEL------------------------------------------
#
par(pty = "s", oma = c(0., 0., 4., 0.))
	par(mfrow = c(1, 1))
	plot(Days, Weight, xlab = "Days", ylab = "Weight (kg)")
	par(mfrow = c(2, 2))
	lin.lm <- lm(Weight ~ Days, model = T)
	print(summary(lin.lm, correlation = F))
	plot(lin.lm,which=1:2)
	matplot(Days, cbind(lin.lm$fitted, Weight), type = "lp", pch = "*")	#
	print(anova(lin.lm, update(lin.lm,  ~ . + I(Days^2)), update(lin.lm,  ~ 
		. + I(Days^2) + I(Days^3)), update(lin.lm,  ~ . + I(Days^2) + I(
		Days^3) + I(Days^4))))	
#--------------FORWARD SELECTION-----------------------------------------------
#
	step.lm <- step(lin.lm, scope = list(upper =  ~ . + I(Days^2) + I(Days^
		3) + I(Days^4) + I(Days^5)), direction = "forward", trace = F)
	print(step.lm$anova)
#	
#--------------STEPWISE SELECTION-----------------------------------------------
#
	step.lm <- step(lin.lm, scope = list(upper =  ~ . + I(Days^2) + I(Days^3) +
		I(Days^4) + I(Days^5)), direction = "both", trace = F)
	print(step.lm$anova)
#
#----------------CUBIC MODEL------------------------------------------------
#
	par(mfrow = c(2, 2))
	c.lm <- update(lin.lm,  ~ . + I(Days^2) + I(Days^3), model = T)
	print(summary(c.lm, correlation = F))
	plot(c.lm,which=1:2)
	matplot(Days, cbind(c.lm$fitted, lin.lm$fitted, Weight), type = "llp", pch = "*")	
      boxcox(c.lm,lambda=seq(0,5,.1))


      par(mfrow=c(2,2))
      c2.lm <- update(c.lm, Weight^2 ~ .,model=T)
      print(summary(c2.lm, correlation = F))
	plot(c2.lm,which=1:2)
	matplot(Days, cbind(sqrt(c2.lm$fitted), c.lm$fitted, Weight), type = "llp", pch = "*")