boot.modreg {dirttee}

Estimate confidence intervals and standard errors for the mode regression fit

Description

Performs bootstrap on the modreg object.

Usage

boot.modreg(
  reg,
  nboot,
  level = 0.95,
  newdata = NULL,
  bw = c("variable", "fixed"),
  quiet = FALSE,
  terms = NULL,
  seed = NULL
)

Arguments

reg

an object of class modreg (output of the modreg function)

nboot

number of bootstrap replications

level

confidence level

newdata

Should be a data frame containing all the variables needed for predictions. If supplied, confidence intervals are calculated for the corresponding predictions.

bw

Either "variable" or "fix", determining if the bandwidth of the original fit should be used for the bootstrap fits (fix) or if the bandwith should be recalculated (variable).

quiet

if TRUE, printing of the status is suppressed

terms

character scalar. If supplied, uses this term for confidence intervals of the prediction

seed

the seed to use

Details

A nonparametric residual bootstrap is performed to calculate standard errors of parameters and confidence intervals. More details can be found in Seipp et al. (2022). newdata can be supplied to get confidence intervals for specific predictions. terms can be specified to calculate confidence interval for the contribution of one covariate (useful for P-splines). variable bandwidth is the default, which has higher coverage than fix, but is computationally much more demanding. A seed can be supplied to guarantee a reproducible result.

Value

a list with the following elements

confpredict

data frame, the confidence intervals for the predictions.

confparams

data frame, the confidence intervals and standard errors for the parametric regression coefficients.

level

confidence level

na

scalar, stating the number of NA bootstrap repetitions.

seed

scalar, the used seed.

References

Seipp, A., Uslar, V., Weyhe, D., Timmer, A., & Otto-Sobotka, F. (2022). Flexible Semiparametric Mode Regression for Time-to-Event Data. Manuscript submitted for publication.

Examples

data(colcancer)
colcancer80 <- colcancer[1:80, ]

# linear mode regression
regL <- modreg(Surv(logfollowup, death) ~ sex + age, data = colcancer80)

# bootstrap with a fixed bandwidth and 3 iterations, chosen to speed up the function. 
# Should in practice be much more than 3 iterations.
btL <- boot.modreg(regL, 3, bw = "fixed", level = 0.9, seed = 100)

# coefficients, SE and confidence intervals
cbind(coef(regL), btL$confparams)


## confidence inverval for smooth effect / predictions

reg <- modreg(Surv(logfollowup, death) ~ sex + s(age, bs = "ps"), data = colcancer80, 
              control = modreg.control(tol_opt = 10^-2, tol_opt2 = 10^-2, tol = 10^-3))
ndat <- data.frame(sex = rep(colcancer80$sex[1], 200), age = seq(50, 90, length = 200))

# iterations should in practice be much more than 2!
bt <- boot.modreg(reg, 2, bw = "fixed", newdata = ndat, terms = "s(age)", seed = 100)

pr <- predict(reg, newdata = ndat, type = "terms", terms = "s(age)")[, 1]

plot(ndat$age, pr, ylim = c(-0.75, 1.5), type = "l", xlab = "age", ylab = "s(age)")
lines(ndat$age, bt$confpredict$lower, lty = 2)
lines(ndat$age, bt$confpredict$upper, lty = 2)