data <- matrix(scan(),ncol=3,byrow=TRUE)
 1 72 83  1 77 85  1 81 90  1 84 92  1 88 97  1 94 92  1 71 79  1 75 80  1 82 89  1 84 87
 2 91 93  2 85 88  2 89 91  2 95 93  2 80 84  2 96 98  2 79 82  2 87 90  2 84 83  2 85 86

data
AncovaEx1 <- data.frame(data)
colnames(AncovaEx1) <- c("group","pretest","posttest")

rm(data)

AncovaEx1$dummy     <- (AncovaEx1$group -1)
AncovaEx1$grouppre  <- (AncovaEx1$group -1)*AncovaEx1$pretest    # create the variable to test for parallelism
AncovaEx1$group <- as.factor(AncovaEx1$group)                  # to make sure that group is a factor, not numeric

AncovaEx1

plot(AncovaEx1$pretest,AncovaEx1$posttest,pch=AncovaEx1$dummy,
      xlab="pretest",ylab="posttest",main="Analysis of Covariance Example")

options(contrasts = c("contr.SAS","contr.poly"))     # set dummy coding to match the scheme in SAS


 # the regression model approach to ANCOVA

anova(lm(posttest ~ dummy, data=AncovaEx1))
anova(lm(posttest ~ pretest, data=AncovaEx1))
anova(lm(posttest ~ pretest + dummy, data=AncovaEx1))
anova(lm(posttest ~ pretest + dummy + grouppre, data=AncovaEx1))


 # using a linear model with factors (like Proc GLM)

AncovaEx1.pre   <-  lm(posttest ~ pretest, data=AncovaEx1)        # Fit all three models
AncovaEx1.grouppre   <-  lm(posttest ~ pretest + group, data=AncovaEx1)
AncovaEx1.groupXpre  <-  lm(posttest ~ pretest + group + grouppre, data=AncovaEx1)


anova(AncovaEx1.grouppre, AncovaEx1.groupXpre)          # perform F tests for nested models
anova(AncovaEx1.pre, AncovaEx1.grouppre)


  # check residuals for the final model
par(mfrow=c(2,2))  
plot(AncovaEx1.grouppre)          
par(mfrow=c(1,1)) 





rm(AncovaEx1,AncovaEx1.pre,AncovaEx1.grouppre,AncovaEx1.groupXpre)    # remove objects when finished