timereg (version 1.8.6)

cum.residuals: Model validation based on cumulative residuals

Description

Computes cumulative residuals and approximative p-values based on resampling techniques.

Usage

cum.residuals(object,data=sys.parent(),modelmatrix=0,cum.resid=1,
              n.sim=500,weighted.test=0,max.point.func=50,weights=NULL)

Arguments

object

an object of class 'aalen', 'timecox', 'cox.aalen' where the residuals are returned ('residuals=1')

data

data frame based on which residuals are computed.

modelmatrix

specifies a grouping of the data that is used for cumulating residuals. Must have same size as data and be ordered in the same way.

n.sim

number of simulations in resampling.

weighted.test

to compute a variance weighted version of the test-processes used for testing constant effects of covariates.

cum.resid

to compute residuals versus each of the continuous covariates in the model.

max.point.func

limits the amount of computations, only considers a max of 50 points on the covariate scales.

weights

weights for sum of martingale residuals, now for cum.resid=1.

Value

• returns an object of type "cum.residuals" with the following arguments:
• cum
  cumulative residuals versus time for the groups specified by modelmatrix.
• var.cum
  the martingale based pointwise variance estimates.
• robvar.cum
  robust pointwise variances estimates of cumulatives.
• obs.testBeq0
  observed absolute value of supremum of cumulative components scaled with the variance.
• pval.testBeq0
  p-value covariate effects based on supremum test.
• sim.testBeq0
  resampled supremum value.
• conf.band
  resampling based constant to construct robust 95% uniform confidence bands for cumulative residuals.
• obs.test
  absolute value of supremum of observed test-process.
• pval.test
  p-value for supremum test statistic.
• sim.test
  resampled absolute value of supremum cumulative residuals.
• proc.cumz
  observed cumulative residuals versus all continuous covariates of model.
• sim.test.proccumz
  list of 50 random realizations of test-processes under model for all continuous covariates.

References

Martinussen and Scheike, Dynamic Regression Models for Survival Data, Springer (2006).

Examples

data(sTRACE)
# Fits Aalen model and returns residuals
fit<-aalen(Surv(time,status==9)~age+sex+diabetes+chf+vf,
           data=sTRACE,max.time=7,n.sim=0,residuals=1)

# constructs and simulates cumulative residuals versus age groups
fit.mg<-cum.residuals(fit,data=sTRACE,n.sim=100,
modelmatrix=model.matrix(~-1+factor(cut(age,4)),sTRACE))

par(mfrow=c(1,4))
# cumulative residuals with confidence intervals
plot(fit.mg);
# cumulative residuals versus processes under model
plot(fit.mg,score=1); 
summary(fit.mg)

# cumulative residuals vs. covariates Lin, Wei, Ying style 
fit.mg<-cum.residuals(fit,data=sTRACE,cum.resid=1,n.sim=100)

par(mfrow=c(2,4))
plot(fit.mg,score=2)
summary(fit.mg)