| Type: | Package |
| Title: | Post-Selection Inference via Simultaneous Confidence Intervals |
| Version: | 0.1.2 |
| Description: | Post-selection inference in linear regression models, constructing simultaneous confidence intervals across a user-specified universe of models. Implements the methodology described in Kuchibhotla, Kolassa, and Kuffner (2022) "Post-Selection Inference" <doi:10.1146/annurev-statistics-100421-044639> to ensure valid inference after model selection, with applications in high-dimensional settings like Lasso selection. |
| License: | MIT + file LICENSE |
| Encoding: | UTF-8 |
| Suggests: | knitr, pbapply, rmarkdown, testthat (≥ 3.0.0), dplyr, glmnet |
| Config/testthat/edition: | 3 |
| RoxygenNote: | 7.3.2 |
| URL: | https://github.com/Chukyhenry/PosiR |
| BugReports: | https://github.com/Chukyhenry/PosiR/issues |
| Imports: | graphics, parallel, stats |
| VignetteBuilder: | knitr |
| Depends: | R (≥ 4.0.0) |
| NeedsCompilation: | no |
| Packaged: | 2025-04-29 08:49:59 UTC; henrychukwuma |
| Author: | Henry Chukwuma [aut, cre] |
| Maintainer: | Henry Chukwuma <chukyhenry55@gmail.com> |
| Repository: | CRAN |
| Date/Publication: | 2025-04-30 15:10:06 UTC |
Fit OLS model using lm.fit (Internal Helper)
Description
Lightweight and robust wrapper around lm.fit() for use in bootstrap procedures.
Designed to handle possible rank-deficiency gracefully by returning NA-coefficients
for linearly dependent terms. Primarily used internally within simultaneous_ci().
Usage
fit_model_q(X_full, y, q_indices)
Arguments
X_full |
Numeric matrix. Full design matrix including intercept if present.
Column names must be unique. Typically derived from |
y |
Numeric vector. Response variable, same length as |
q_indices |
Integer vector. Column indices (1-based) specifying the submodel to fit. |
Value
Named numeric vector of estimated coefficients. If fitting fails or coefficients are dropped due to collinearity, NA values are returned with expected names.
Plot Simultaneous Confidence Intervals
Description
Visualizes confidence intervals returned by simultaneous_ci() using base R graphics.
Estimates are shown as points with corresponding CI segments, grouped and labeled by
model and coefficient name. Supports customization for log scale, character sizes,
label trimming, and reference lines.
Usage
## S3 method for class 'simultaneous_ci_result'
plot(
x,
y = NULL,
subset_pars = NULL,
log.scale = FALSE,
cex = 0.8,
cex.labels = 0.8,
las.labels = 1,
pch = 16,
col.estimate = "blue",
col.ci = "darkgray",
col.ref = "red",
ref.line.pos = 0,
lty.ref = 2,
main = "Simultaneous Confidence Intervals",
xlab = NULL,
label.trim = NULL,
...
)
Arguments
x |
An object of class |
y |
Ignored. |
subset_pars |
Optional character vector. Coefficient names to subset the plot. Default: all. |
log.scale |
Logical. Plot on logarithmic scale. Intervals crossing 0 or with nonpositive bounds are excluded. |
cex |
Point size for estimates. Default = 0.8. |
cex.labels |
Label size for y-axis. Default = 0.8. |
las.labels |
Orientation of y-axis labels (0, 1, 2, or 3). Default = 1. |
pch |
Plot character for point estimates. Default = 16. |
col.estimate |
Color of point estimates. Default = "blue". |
col.ci |
Color of confidence interval lines. Default = "darkgray". |
col.ref |
Color of reference line(s). Default = "red". |
ref.line.pos |
Position(s) for vertical reference line(s). Default = 0. Set to NULL to omit. |
lty.ref |
Line type for reference lines. Default = 2 (dashed). |
main |
Plot title. Default = "Simultaneous Confidence Intervals". |
xlab |
X-axis label. If NULL and |
label.trim |
Integer. Trims long coefficient labels to this width (adds "..."). Optional. |
... |
Additional arguments passed for future use (currently ignored). |
Value
Invisibly returns a list:
-
ycoords: Named vector of y-axis positions for each label -
xlim: Range of x-axis limits used -
ylim: Range of y-axis limits used
If no valid intervals are available for plotting, returns invisible(NULL).
Examples
set.seed(1)
X <- matrix(rnorm(100*2), 100, 2, dimnames = list(NULL, c("X1", "X2")))
y <- 1 + X[,1] - X[,2] + rnorm(100)
res <- simultaneous_ci(X, y, list(mod = 1:3), B = 100, add_intercept = TRUE)
plot(res)
Compute Simultaneous Confidence Intervals via Bootstrap (Post-Selection Inference)
Description
Implements Algorithm 1 from the reference paper using bootstrap-based max-t statistics to construct valid simultaneous confidence intervals for selected regression coefficients across a user-specified universe of linear models.
Usage
simultaneous_ci(
X,
y,
Q_universe,
alpha = 0.05,
B = 1000,
add_intercept = TRUE,
bootstrap_method = "pairs",
cores = 1,
use_pbapply = TRUE,
seed = NULL,
verbose = TRUE,
...
)
Arguments
X |
Numeric matrix (n x p): Design matrix. Must have unique column names.
Do not include an intercept if |
y |
Numeric vector (length n): Response vector. |
Q_universe |
Named list of numeric vectors. Each element specifies a model as a
vector of column indices (accounting for intercept if |
alpha |
Significance level for the confidence intervals. Default is 0.05. |
B |
Integer. Number of bootstrap samples. Default is 1000. |
add_intercept |
Logical. If TRUE, adds an intercept as the first column of the design matrix. Default is TRUE. |
bootstrap_method |
Character. Bootstrap type. Only "pairs" is currently supported. |
cores |
Integer. Number of CPU cores to use for bootstrap parallelization. Default is 1. |
use_pbapply |
Logical. Use |
seed |
Optional numeric. Random seed for reproducibility. Used for parallel-safe RNG. |
verbose |
Logical. Whether to display status messages. Default is TRUE. |
... |
Reserved for future use. |
Details
Supports parallel execution, internal warnings capture, and returns structured results with estimates, intervals, bootstrap diagnostics, and inference statistics.
Value
A list of class simultaneous_ci_result with elements:
-
intervals: Data frame with estimates, confidence intervals, variances, and SEs -
K_alpha: Bootstrap (1 - alpha) quantile of max-t statistics -
T_star_b: Vector of bootstrap max-t statistics -
n_valid_T_star_b: Number of finite bootstrap max-t statistics -
alpha,B,bootstrap_method: Metadata -
warnings_list: Internal warnings collected during bootstrap/model fitting -
valid_bootstrap_counts: Valid bootstrap replicates per parameter -
n_bootstrap_errors: Total bootstrap fitting errors
References
Kuchibhotla, A., Kolassa, J., & Kuffner, T. (2022). Post-selection inference. Annual Review of Statistics and Its Application, 9(1), 505–527.
Examples
set.seed(123)
X <- matrix(rnorm(100 * 2), 100, 2, dimnames = list(NULL, c("X1", "X2")))
y <- X[,1] * 0.5 + rnorm(100)
Q <- list(model = 1:2)
res <- simultaneous_ci(X, y, Q, B = 100, cores = 1)
print(res$intervals)
plot(res)