ffAnova {ffmanova}

Type II* Anova

Description

Analysis of variance table for a linear model using type II* sums of squares, which are described in Langsrud et al. (2007). Type II* extends the type II philosophy to continuous variables. The results are invariant to scale changes and pitfalls are avoided.

Usage

ffAnova(formula, data = NULL)

Arguments

formula

A model formula or an R object (preferably an lm object).

data

An optional data frame, list or environment.

Details

This function is a variant of ffmanova for the univariate special case. The two input parameters will be interpreted by model.frame.

Value

An object of class "anova" (see anova).

Author(s)

Øyvind Langsrud and Bjørn-Helge Mevik

References

Langsrud, Ø., Jørgensen, K., Ofstad, R. and Næs, T. (2007): “Analyzing Designed Experiments with Multiple Responses”, Journal of Applied Statistics, 34, 1275-1296.

Examples

# Generate example data
set.seed(123)
a <- c(0, 0, 0, 10, 10, 10, 1, 1, 1)
A <- as.character(a)  # A is categorical
b <- 1:9
y <- rnorm(9)/10 + a  # y depends strongly on a (and A)
a100 <- a + 100  # change of scale (origin)
b100 <- b + 100  # change of scale (origin)

# Four ways of obtaining the same results
ffAnova(y ~ A * b)
ffAnova(y ~ A * b100)
ffAnova(lm(y ~ A * b))
ffAnova(y ~ A * b, data.frame(A = A, y = y, b = 1:9))

# Second order continuous variable
ffAnova(y ~ a + I(a^2))

# Model equivalent to 'y ~ A * b'
ffAnova(y ~ (a + I(a^2)) * b)

# Demonstrating similarities and differences using package car
if (!require(car))        # Package car is loaded if available 
  Anova <- function(...) {  # Replacement function if car not available
    warning("No results since package car is not available")}

lm_Ab <- lm(y ~ A * b)
lm_Ab100 <- lm(y ~ A * b100)

# Type II same as type II* in this case
Anova(lm_Ab)      # Type II
Anova(lm_Ab100)   # Type II
ffAnova(lm_Ab)    # Type II*
ffAnova(lm_Ab100) # Type II*

# Type III depends on scale
Anova(lm_Ab, type = 3)
Anova(lm_Ab100, type = 3)

lm_a <- lm(y ~ a + I(a^2))
lm_a100 <- lm(y ~ a100 + I(a100^2))

# Now Type II depends on scale
Anova(lm_a)      # Type II
Anova(lm_a100)   # Type II
ffAnova(lm_a)    # Type II*
ffAnova(lm_a100) # Type II*