discSurv (version 1.1.2)

adjDevResid: Adjusted Deviance Residuals

Description

Calculates the adjusted deviance residuals. This function only supports generalized, linear models. The adjusted deviance residuals should be approximately normal distributed, in the case of a well fitting model.

Usage

adjDevResid(dataSet, survModelFormula, censColumn, linkFunc = "logit", idColumn = NULL)

Arguments

dataSet

Original data in short format. Must be of class "data.frame".

survModelFormula

Gives the specified relationship of discrete response and covariates. The formula is designed, that the intercepts for the time dependent base line hazards are always included. Therefore only covariates should be given in this formula. This argument is re

censColumn

Gives the column name of the event indicator (1=observed, 0=censored). Must be of type "character".

linkFunc

Specifies the desired link function in use of generalized, linear models.

idColumn

Gives the column name of the identification number of each person. The argument must be of type "character". Default NULL means, that each row equals one person (no repeated measurements).

Value

• • Output:
  {List with objects:}
  • AdjDevResid:
  {Adjusted deviance residuals as numeric vector.}
• GlmFit:
• {Fit object of class (generalized, linear model used in the calculations)}

item

Input:

References

Gerhard Tutz, (2012), Regression for Categorical Data, Cambridge University Press

See Also

devResid, brierScore, glm, predErrDiscShort

Examples

# Example with cross validation and unemployment data 
library(Ecdat)
data(UnempDur)
summary(UnempDur$spell)

# Extract subset of data
set.seed(635)
IDsample <- sample(1:dim(UnempDur)[1], 100)
UnempDurSubset <- UnempDur [IDsample, ]

# Calculate adjusted deviance residuals for the unemployment data subset
adjDevianceResiduals <- adjDevResid (dataSet=UnempDurSubset, 
survModelFormula=spell ~ age + logwage, 
censColumn="censor1", linkFunc="logit", idColumn=NULL)
adjDevianceResiduals

# Exclude outliers
adjDevResidWoOut <- adjDevianceResiduals$Output$AdjDevResid [
adjDevianceResiduals$Output$AdjDevResid < 
quantile(adjDevianceResiduals$Output$AdjDevResid, prob=0.9) & 
adjDevianceResiduals$Output$AdjDevResid > 
quantile(adjDevianceResiduals$Output$AdjDevResid, prob=0.1)]

# Compare nonparametric density estimate of adjusted deviance residuals 
# with adapted normal distribution
plot(density(adjDevResidWoOut), xlim=c(-10,10), 
main="Density comparison: Normal vs nonparametric estimate", lwd=2, las=1, col="red")
dnorm1 <- function (x) {dnorm (x, mean=mean(adjDevResidWoOut), sd=sd(adjDevResidWoOut))}
curve (dnorm1, n=500, add=TRUE, col="black", lwd=2)
legend("topright", legend=c("Normal", "Nonpar"), lty=1, lwd=2, col=c("black", "red"))