README.md

In brentonk/coefbounds: Nonparametric Bounds for Regression with Interval-Censored Outcomes

coefbounds

Nonparametric Bounds for Regression with Interval-Censored Outcomes

Author: Brenton Kenkel, Vanderbilt University

Installation

coefbounds is not yet available on CRAN. To install directly from GitHub:

if (!require("devtools")) {
    install.packages("devtools")
    library("devtools")
}

devtools::install_github("brentonk/coefbounds", build_vignettes = TRUE)

Note

coefbounds is in an early stage of development. Backward-incompatible API changes are possible until version 1.0.0 is released.

Usage

Below are some simple usage examples. For more fleshed-out examples, see the package vignette: vignette("introduction", package = "coefbounds").

library("coefbounds")

I'll begin by generating some simple data with an interval-censored outcome.

set.seed(97)
x1 <- rnorm(100)
x2 <- rnorm(100)
y <- 1 - x1 + x2 + rnorm(100)
lwr <- floor(y)
upr <- ceiling(y)

To estimate coefficient bounds, use coefbounds() with a formula of the form yl + yu ~ x1 + x2 + ..., where yl is the lower bound on each response value and yu is the upper bound.

fit_full <- coefbounds(lwr + upr ~ x1 + x2, boot = 100)
fit_full
#> 
#> Call:
#> coefbounds(formula = lwr + upr ~ x1 + x2, boot = 100)
#> 
#> Estimated coefficient bounds:
#>              lower upper
#> (Intercept)  0.652  1.65
#> x1          -1.247 -0.40
#> x2           0.722  1.48

For inference, make sure coefbounds() is run with boot > 0 and use interval_hypothesis() or confint().

interval_hypothesis(fit = fit_full,
                    term = "x1",
                    interval = c(0, 0),
                    type = "subset")
#> 
#> Null hypothesis: identification region for x1 contains [0, 0]
#> Estimated identification region: [-1.247, -0.405]
#> Test statistic: sqrt(N) * directed Hausdorff distance
#> 
#> stat = 4.05, p-value = <0.01
interval_hypothesis(fit = fit_full,
                    term = "x2",
                    interval = c(0.7, 1.6),
                    type = "equal")
#> 
#> Null hypothesis: identification region for x2 equals [0.7, 1.6]
#> Estimated identification region: [0.722, 1.480]
#> Test statistic: sqrt(N) * Hausdorff distance
#> 
#> stat = 1.2, p-value = 0.31

confint(fit_full, level = 0.99)
#>                0.5 %   99.5 %
#> (Intercept)  0.40717  1.89768
#> x1          -1.47298 -0.17848
#> x2           0.47613  1.72565

To Do

• Allow clustering in critical value bootstrap.
• Enable parallel processing in bootstrap.

brentonk/coefbounds documentation built on May 13, 2019, 5:09 a.m.