dat <- read.csv("bodyfat.csv")
n <- dim(dat)[1]
p <- 4
q <- 2
fit <- lm(bodyfat ~ Weight + Height + Chest + Abdomen, data=dat)
RSS <- sum(resid(fit)^2)
Test for \(H_0 : \beta_{weight} = \beta_{height}, \beta_{chest} = -2 \beta_{abdomen}\). The model can then be rewritten as \[\text{bodyfat} = \beta_0 + \beta_{weight} (\text{weight} + \text{height}) + \beta_{abdomen} (\text{abdomen} - 2 \cdot \text{chest})\]
Use formula \[ \frac{(\text{RSS}(m) - \text{RSS}(M)) / q}{\text{RSS}(M) / (n - p - 1)}. \]
fit0 <- lm(bodyfat ~ I(Weight + Height) + I(-2 * Chest + Abdomen), data=dat)
RSS0 <- sum(resid(fit0)^2)
Fstat1 <- ((RSS0 - RSS) / 2) / (RSS / (n - p - 1))
Can write hypothesis as \(H_0 : L \beta = 0\), where \[L = \begin{bmatrix} 0 & 1 & -1 & 0 & 0 \\ 0 & 0 & 0 & 1 & 2 \end{bmatrix}.\]
Use formula \[\frac{(L\hat{\beta} - c)^\top [L (X^\top X)^{-1} L^\top]^{-1} (L\hat{\beta} - c) / q}{\text{RSS}(M) / (n - p - 1)}.\]
X <- as.matrix(cbind(1, dat[,c("Weight", "Height", "Chest", "Abdomen")]))
y <- as.numeric(dat$bodyfat)
beta_hat <- solve(t(X) %*% X, t(X) %*% y)
y_hat <- X %*% beta_hat
L <- matrix(c(0,0,1,0,-1,0,0,1,0,2), 2)
Fstat2 <- (t(L %*% beta_hat) %*% solve(L %*% solve(t(X) %*% X) %*% t(L)) %*% (L %*% beta_hat) / q) / (sum((y - y_hat)^2) / (n - p - 1))
Check that they match.
Fstat1
## [1] 172.7717
Fstat2
## [,1]
## [1,] 172.7717