gof (version 0.9.1)

cumres.glm: Calculates GoF statistics based on cumulative residual processes

Description

Given the generalized linear models model $$g(E(Y_i|X_{i1},...,X_{ik})) = \sum_{i=1}^k \beta_jX_{ij}$$ the cumres-function calculates the the observed cumulative sum of residual process, cumulating the residuals, $e_i$, by the jth covariate: $$W_j(t) = n^{-1/2}\sum_{i=1}^n 1_{{X_{ij}

Usage

## S3 method for class 'glm':
cumres(model,
    variable = c("predicted", colnames(model.matrix(model))),
    data = data.frame(model.matrix(model)), R = 1000,
    b = 0, plots = min(R, 50), breakties = 1e-12,
    seed = round(runif(1, 1, 1e+09)), ...)

Arguments

model

Model object (lm or glm)

variable

List of variable to order the residuals after

data

data.frame used to fit model (complete cases)

R

Number of samples used in simulation

b

Moving average bandwidth (0 corresponds to infinity = standard cumulated residuals)

plots

Number of realizations to save for use in the plot-routine

breakties

Add unif[0,breakties] to observations

seed

Random seed

...

additional arguments

Value

• Returns an object of class 'cumres'.

References

D.Y. Lin and L.J. Wei and Z. Ying (2002) Model-Checking Techniques Based on Cumulative Residuals. Biometrics, Volume 58, pp 1-12.

John Q. Su and L.J. Wei (1991) A lack-of-fit test for the mean function in a generalized linear model. Journal. Amer. Statist. Assoc., Volume 86, Number 414, pp 420-426.

See Also

cox.aalen in the timereg-package for similar GoF-methods for survival-data.

Examples

sim1 <- function(n=100, f=function(x1,x2) {10+x1+x2^2}, sd=1, seed=1) {
  if (!is.null(seed))
    set.seed(seed)
  x1 <- rnorm(n);
  x2 <- rnorm(n)
  X <- cbind(1,x1,x2)
  y <- f(x1,x2) + rnorm(n,sd=sd)
  d <- data.frame(y,x1,x2)
  return(d)
}
d <- sim1(100); l <- lm(y ~ x1 + x2,d)
system.time(g <- cumres(l, R=100, plots=50))
g
plot(g)
g1 <- cumres(l, c("y"), R=100, plots=50)
g1
g2 <- cumres(l, c("y"), R=100, plots=50, b=0.5)
g2