SCoRES (version 0.1.2)

SCB_functional_outcome: Construct Simultaneous Confidence Bands (SCB) For One Dimensional Functional Data

Description

This function builds simultaneous confidence bands through parametric and bootstrap approaches.

Usage

SCB_functional_outcome(
  data_df,
  object = NULL,
  method,
  fitted = TRUE,
  alpha = 0.05,
  outcome,
  domain,
  subset = NULL,
  id,
  nboot = NULL,
  method_SD = "t",
  weights = "rademacher"
)

Value

A list containing:

mu_hat

Estimated mean function for the group of interest.

domain

The domain used.

se_hat

Standard errors of the estimated means.

scb_low

Lower bound of the simultaneous confidence band.

scb_up

Upper bound of the simultaneous confidence band.

type

A character description of the output type.

Arguments

data_df

A functional data frame that contains both name and values for variables including functional outcome, domain (e.g. time) and ID (e.g. subject names) used to fit the model object.

object

A fitted Function-on-Scalar Regression (FoSR) object (e.g., from mgcv::gam()/mgcv::bam()). Default is NULL

method

A character string specifying the approach:

• "cma" - Correlation and Multiplicity Adjusted (CMA) confidence bands via parametric approach (requires a fitted functional regression model)

• "multiplier" - Dense confidence bands via Multiplier-t Bootstrap method For method = "multiplier", the outcome variable in data_df should not have all-zero entries within any specified domain (except for domain index zero, where this is allowed). Otherwise, the function will return an error. If missing values (NA) exist in the outcome variable in data_df, the function will impute them using fpca.face before performing the Multiplier Bootstrap.

fitted

Logical. Whether to estimate the simultaneous confidence bands for the fitted mean function or the fitted parameter function

• TRUE - Estimate the simultaneous confidence bands for the fitted mean outcome function.

• FALSE - estimate the simultaneous confidence bands for the fitted parameter function.

Default is TRUE.

alpha

Significance level for SCB. Default is 0.05.

outcome

A character string specifying the name of the outcome variable used in the model.

domain

A character string specifying the name of the domain variable (e.g. time) used in the model.

subset

An atomic character vector (e.g., c("user = 1", "age = 30")) specifying the target function for constructing the SCB. Each element must be of the form <name> = <value>, where <name> is the name of a scalar grouping variable and <value> is the desired value. Whitespace is ignored. Binary or categorical character variables should be transformed into numeric. Factors are not allowed here because if the input data contains factor variables, they will be automatically expanded into dummy (indicator) variables when constructing the design matrix, and the resulting variable names may differ from the original factor names. Default is NULL, representing the reference group.

id

A character string specifying the name of the ID variable.

nboot

An integer specifying the number of bootstrap samples used to construct the confidence bands. Default is 10,000 for cma, 5000 for Multiplier Bootstrap.

method_SD

Method for SD estimation: "t" or "regular". Default is "t".

weights

Multiplier type: "rademacher", "gaussian", or "mammen". Default is "rademacher".

Examples

# example using pupil data
if (requireNamespace("mgcv", quietly = TRUE)) {
data(pupil)
# \donttest{
pupil_fpca <- prepare_pupil_fpca(pupil)

fosr_mod <- mgcv::bam(percent_change ~ s(seconds, k=30, bs="cr") +
  s(seconds, by = use, k=30, bs = "cr") +
  s(id, by = Phi1, bs="re") +
  s(id, by = Phi2, bs="re"),
  method = "fREML", data = pupil_fpca, discrete = TRUE)

# CMA approach
results <- SCB_functional_outcome(data_df = pupil, object = fosr_mod,
                                  method = "cma", fitted = TRUE,
                                  outcome = "percent_change", domain = "seconds",
                                  subset = c("use = 1"), id = "id")


# multiplier bootstrap
results <- SCB_functional_outcome(data_df = pupil, object = fosr_mod,
                                  method = "multiplier", fitted = TRUE,
                                  outcome = "percent_change", domain = "seconds",
                                  subset = c("use = 1"), id = "id")
# }

mean_mod <- mgcv::gam(percent_change ~ s(seconds, k = 5, bs = "cr") +
s(seconds, by = use, k = 5, bs = "cr"),
data = pupil, method = "REML")

# multiplier bootstrap
pupil_multiplier <- SCB_functional_outcome(data = pupil, object = mean_mod, method = "multiplier",
                                   outcome = "percent_change",
                                   domain = "seconds", subset= c("use = 1"),
                                   id = "id")
}