Build a deviation contrast matrix, a type of effects contrast matrix.
Arguments
- n
a vector of levels for a factor, or the number of levels.
- base
an integer specifying which group is considered the baseline group. Ignored if
contrastsisFALSE.- contrasts
a logical indicating whether contrasts should be computed.
- sparse
logical indicating if the result should be sparse (of class
dgCMatrix), using package Matrix.
Details
In effects coding, unlike treatment/dummy coding
(stats::contr.treatment()), each contrast sums to 0. In regressions models,
this results in an intercept that represents the (unweighted) average of the
group means. In ANOVA settings, this also guarantees that lower order effects
represent main effects (and not simple or conditional effects, as is
the case when using R's default stats::contr.treatment()).
Deviation coding (contr.deviation) is a type of effects coding. With
deviation coding, the coefficients for factor variables are interpreted as
the difference of each factor level from the base level (this is the same
interpretation as with treatment/dummy coding). For example, for a factor
group with levels "A", "B", and "C", with contr.devation, the intercept
represents the overall mean (average of the group means for the 3 groups),
and the coefficients groupB and groupC represent the differences between
the A group mean and the B and C group means, respectively.
Sum coding (stats::contr.sum()) is another type of effects coding. With sum
coding, the coefficients for factor variables are interpreted as the
difference of each factor level from the grand (across-groups) mean. For
example, for a factor group with levels "A", "B", and "C", with
contr.sum, the intercept represents the overall mean (average of the group
means for the 3 groups), and the coefficients group1 and group2 represent
the differences the
A and B group means from the overall mean, respectively.
Examples
if (FALSE) { # !identical(Sys.getenv("IN_PKGDOWN"), "true")
# \donttest{
data("mtcars")
mtcars <- data_modify(mtcars, cyl = factor(cyl))
c.treatment <- cbind(Intercept = 1, contrasts(mtcars$cyl))
solve(c.treatment)
#> 4 6 8
#> Intercept 1 0 0 # mean of the 1st level
#> 6 -1 1 0 # 2nd level - 1st level
#> 8 -1 0 1 # 3rd level - 1st level
contrasts(mtcars$cyl) <- contr.sum
c.sum <- cbind(Intercept = 1, contrasts(mtcars$cyl))
solve(c.sum)
#> 4 6 8
#> Intercept 0.333 0.333 0.333 # overall mean
#> 0.667 -0.333 -0.333 # deviation of 1st from overall mean
#> -0.333 0.667 -0.333 # deviation of 2nd from overall mean
contrasts(mtcars$cyl) <- contr.deviation
c.deviation <- cbind(Intercept = 1, contrasts(mtcars$cyl))
solve(c.deviation)
#> 4 6 8
#> Intercept 0.333 0.333 0.333 # overall mean
#> 6 -1.000 1.000 0.000 # 2nd level - 1st level
#> 8 -1.000 0.000 1.000 # 3rd level - 1st level
## With Interactions -----------------------------------------
mtcars <- data_modify(mtcars, am = C(am, contr = contr.deviation))
mtcars <- data_arrange(mtcars, select = c("cyl", "am"))
mm <- unique(model.matrix(~ cyl * am, data = mtcars))
rownames(mm) <- c(
"cyl4.am0", "cyl4.am1", "cyl6.am0",
"cyl6.am1", "cyl8.am0", "cyl8.am1"
)
solve(mm)
#> cyl4.am0 cyl4.am1 cyl6.am0 cyl6.am1 cyl8.am0 cyl8.am1
#> (Intercept) 0.167 0.167 0.167 0.167 0.167 0.167 # overall mean
#> cyl6 -0.500 -0.500 0.500 0.500 0.000 0.000 # cyl MAIN eff: 2nd - 1st
#> cyl8 -0.500 -0.500 0.000 0.000 0.500 0.500 # cyl MAIN eff: 2nd - 1st
#> am1 -0.333 0.333 -0.333 0.333 -0.333 0.333 # am MAIN eff
#> cyl6:am1 1.000 -1.000 -1.000 1.000 0.000 0.000
#> cyl8:am1 1.000 -1.000 0.000 0.000 -1.000 1.000
# }
}