make.contrasts {gmodels}

Construct a User-Specified Contrast Matrix

Description

This function converts human-readable contrasts into the form that R requires for computation.

Usage

make.contrasts(contr, how.many = ncol(contr))

Arguments

contr

vector or matrix specifying contrasts (one per row).

how.many

dimensions of the desired contrast matrix. This must equal the number of levels of the target factor variable.

Details

Specifying a contrast row of the form c(1,0,0,-1) creates a contrast that will compare the mean of the first group with the mean of the fourth group.

Value

make.contrasts returns a matrix with dimensions (how.many, how.many) containing the specified contrasts augmented (if necessary) with orthogonal "filler" contrasts.

This matrix can then be used as the argument to contrasts() or to the contrasts argument of model functions (eg, lm()).

Author(s)

Gregory R. Warnes [e-mail]

See Also

• stats::lm(), stats::contrasts(), stats::contr.treatment(), stats::contr.poly(),

• Computation and testing of General Linear Hypothesis: glh.test(),

• Computation and testing of estimable functions of model coefficients: estimable(),

• Estimate and Test Contrasts for a previously fit linear model: fit.contrast()

Examples

set.seed(4684)
y <- rnorm(100)
x.true <- rnorm(100, mean=y, sd=0.25)
x <-  factor(cut(x.true,c(-4,-1.5,0,1.5,4)))
reg <- lm(y ~ x)
summary(reg)

# Mirror default treatment contrasts
test <- make.contrasts(rbind( c(-1,1,0,0), c(-1,0,1,0), c(-1,0,0,1) ))
lm( y ~ x, contrasts=list(x = test ))

# Specify some more complicated contrasts
#   - mean of 1st group vs mean of 4th group
#   - mean of 1st and 2nd groups vs mean of 3rd and 4th groups
#   - mean of 1st group vs mean of 2nd, 3rd and 4th groups
cmat <- rbind( "1 vs 4"    =c(-1, 0, 0, 1),
               "1+2 vs 3+4"=c(-1/2,-1/2, 1/2, 1/2),
               "1 vs 2+3+4"=c(-3/3, 1/3, 1/3, 1/3))

summary(lm( y ~ x, contrasts=list(x=make.contrasts(cmat) )))
# or
contrasts(x) <- make.contrasts(cmat)
summary(lm( y ~ x ) )

# or use contrasts.lm
reg <- lm(y ~ x)
fit.contrast( reg, "x", cmat )

# compare with values computed directly using group means
gm <- sapply(split(y,x),mean)
gm %*% t(cmat)


#
# Example for Analysis of Variance
#

set.seed(03215)
Genotype <- sample(c("WT","KO"), 1000, replace=TRUE)
Time <- factor(sample(1:3, 1000, replace=TRUE))
data <- data.frame(y, Genotype, Time)
y <- rnorm(1000)

data <- data.frame(y, Genotype, as.factor(Time))

# Compute Contrasts & obtain 95% confidence intervals

model <- aov( y ~ Genotype + Time + Genotype:Time, data=data )

fit.contrast( model, "Genotype", rbind("KO vs WT"=c(-1,1) ), conf=0.95 )

fit.contrast( model, "Time",
            rbind("1 vs 2"=c(-1,1,0),
                  "2 vs 3"=c(0,-1,1)
                  ),
            conf=0.95 )


cm.G <- rbind("KO vs WT"=c(-1,1) )
cm.T <- rbind("1 vs 2"=c(-1,1,0),
              "2 vs 3"=c(0,-1,1) )

# Compute contrasts and show SSQ decompositions

model <- model <- aov( y ~ Genotype + Time + Genotype:Time, data=data,
                      contrasts=list(Genotype=make.contrasts(cm.G),
                                     Time=make.contrasts(cm.T) )
                      )

summary(model, split=list( Genotype=list( "KO vs WT"=1 ),
                           Time = list( "1 vs 2" = 1,
                                        "2 vs 3" = 2 ) ) )