TSVC (version 1.7.2)

confint.TSVC: Confidence intervals for Varying Coefficient Trees

Description

Construct parametric bootstrap percentile confidence intervals of effects of covariates that vary with the values of one or several effect modifiers. The basic method is described in Spuck et al. (2025).

Usage

# S3 method for TSVC
confint(
  object,
  bootstrap_n = 500,
  alpha = 0.05,
  post_pruning = NULL,
  splits_max = 5,
  trace = FALSE,
  ...
)

Arguments

object

a fitted object of class TSVC.

bootstrap_n

the number of bootstrap samples to be drawn.

alpha

significance level \(alpha\) of the confidence interval.

post_pruning

method to select the maximal number of splits; can be "AIC" or "BIC". If NULL (default), no post-pruning is performed.

splits_max

maximal number of splits to be considered. If post_pruning is NULL it is ignored.

trace

if TRUE, information about the estimation progress is printed.

...

further arguments passed to or from other methods.

Author

Moritz Berger <[e-mail]>
https://www.imbie.uni-bonn.de/people/dr-moritz-berger/

Nikolai Spuck <[e-mail]>
https://www.imbie.uni-bonn.de/people/nikolai-spuck/

Details

The method is so far mainly put to the test for gaussian (family=gaussian) and binary (family=binary(link="logit")) outcome. It should be taken with care for differently scaled outcomes.

References

Berger, M., G. Tutz and M. Schmid (2019). Tree-Structured Modelling of Varying Coefficients. Statistics and Computing 29, 217-229, https://doi.org/10.1007/s11222-018-9804-8.

Spuck, N., M. Schmid, M. Monin and M. Berger (2025). Confidence intervals for tree-structured varying coefficients. Computational Statistics and Data Analysis.

See Also

TSVC, plot.TSVC, predict.TSVC, summary.TSVC

Examples

# Swiss Labour Market 
library(AER)
data("SwissLabor")

# recode factors 
sl <- SwissLabor
sl$participation <- as.numeric(sl$participation)-1
sl$foreign       <- as.numeric(sl$foreign)-1

if (FALSE) {
fit1 <- TSVC(participation~income+age, data=sl, family=binomial(link="logit"), 
perm_test=FALSE, test_linear=FALSE, splits_max=3)
confint(fit1, bootstrap_n=500, alpha=0.05, trace=TRUE)
}