G-Computation Mean-Difference Inference for Ordinal Responses

Fits a proportional-odds working model for an ordinal outcome using treatment and, optionally, all recorded covariates (fast_ordinal_regression_with_var_cpp), then estimates the marginal difference in expected ordinal category score by G-computation — see gcomp_ordinal_proportional_odds_post_fit_cpp for the exact standardization formula (mean1 - mean0). Standard errors are obtained by the delta method: a central finite-difference gradient of the mean-difference functional with respect to the fitted \([\alpha, \beta]\) parameters, propagated through the model's fitted variance-covariance matrix, \(\widehat{\mathrm{Var}}(\widehat{\mathrm{md}}) = \nabla^\top \widehat{\mathrm{Var}}(\hat\theta) \nabla\). If that delta-method standard error is unavailable or non-finite, the Wald-style methods ($compute_asymp_confidence_interval(), $compute_asymp_two_sided_pval(), $compute_wald_confidence_interval(), $compute_wald_two_sided_pval()) all silently fall back to a nonparametric bootstrap interval/p-value instead (with a warning), rather than returning NA.

Super class

Inference -> InferenceOrdinalGCompMeanDiff

Methods

Public methods

• InferenceOrdinalGCompMeanDiff$new()

• InferenceOrdinalGCompMeanDiff$compute_estimate()

• InferenceOrdinalGCompMeanDiff$compute_estimate_with_bootstrap_weights()

• InferenceOrdinalGCompMeanDiff$compute_asymp_confidence_interval()

• InferenceOrdinalGCompMeanDiff$compute_asymp_two_sided_pval()

• InferenceOrdinalGCompMeanDiff$compute_wald_confidence_interval()

• InferenceOrdinalGCompMeanDiff$compute_wald_two_sided_pval()

• InferenceOrdinalGCompMeanDiff$clone()

+ inherited public methods from Inference

• Inference$capabilities()
• Inference$compute_exact_confidence_interval()
• Inference$compute_exact_two_sided_pval_for_treatment_effect()
• Inference$duplicate()
• Inference$get_analysis_data()
• Inference$get_covariates()
• Inference$get_design_object()
• Inference$get_model_formula()
• Inference$get_nonestimable_reason()
• Inference$get_nonestimable_stage()
• Inference$get_optimization_alg()
• Inference$get_response()
• Inference$get_response_type()
• Inference$get_treatment()
• Inference$is_nonestimable()
• Inference$set_optimization_alg()
• Inference$set_seed()
• Inference$supports()

---

InferenceOrdinalGCompMeanDiff$new()

Uses the shared randomization two-sided p-value contract; see InferenceRand.

Initialize the ordinal g-computation (G-Comp) inference object for a completed design with an ordinal, uncensored response.

Usage

InferenceOrdinalGCompMeanDiff$new(
  des_obj,
  model_formula = NULL,
  verbose = FALSE,
  smart_cold_start_default = NULL
)

Arguments

des_obj

A completed DesignSeqOneByOne object with an ordinal response.

model_formula

Optional formula for covariate adjustment. If NULL (default), the formula from the design object is used and its pre-computed design matrix is reused. If a formula is provided, a new design matrix is constructed from the design's imputed covariates.

verbose

Whether to print progress messages.

smart_cold_start_default

Whether to use smart cold start values by default.

---

InferenceOrdinalGCompMeanDiff$compute_estimate()

Computes the G-computation standardized mean-difference treatment-effect estimate (see class documentation for the full proportional-odds-based standardization).

Usage

InferenceOrdinalGCompMeanDiff$compute_estimate(estimate_only = FALSE)

Arguments

estimate_only

If TRUE, skip variance component calculations.

---

InferenceOrdinalGCompMeanDiff$compute_estimate_with_bootstrap_weights()

Recomputes the G-computation mean-difference estimate under subject/block bootstrap weights (via fast_ordinal_regression_weighted_cpp plus gcomp_ordinal_proportional_odds_post_fit_cpp), used by the Bayesian bootstrap and related weighted-resampling machinery; see InferenceNonParamBootstrap. Runs side-effect free: the ordinary (unweighted) cached fit, warm-start state, and rank-reduced column selection are saved before the weighted refit and restored afterward (on.exit), so a weighted bootstrap replicate cannot corrupt the class's own point estimate or subsequent fits.

Usage

InferenceOrdinalGCompMeanDiff$compute_estimate_with_bootstrap_weights(
  subject_or_block_weights,
  estimate_only = FALSE
)

Arguments

subject_or_block_weights

Row weights for the bootstrap sample.

estimate_only

If TRUE, skip variance calculations.

---

InferenceOrdinalGCompMeanDiff$compute_asymp_confidence_interval()

Computes a \(1-\alpha\) confidence interval for the G-Comp mean difference using the delta-method standard error (see class documentation), or falls back (with a warning) to a nonparametric bootstrap interval if that standard error is unavailable. Identical to $compute_wald_confidence_interval().

Usage

InferenceOrdinalGCompMeanDiff$compute_asymp_confidence_interval(alpha = 0.05)

Arguments

alpha

The significance level (default 0.05).

---

InferenceOrdinalGCompMeanDiff$compute_asymp_two_sided_pval()

Computes a two-sided Wald p-value testing \(H_0: \mathrm{md} = \code{delta}\) using the delta-method standard error (see class documentation), or falls back (with a warning) to a nonparametric bootstrap p-value if that standard error is unavailable. Identical to $compute_wald_two_sided_pval().

Usage

InferenceOrdinalGCompMeanDiff$compute_asymp_two_sided_pval(delta = 0)

Arguments

delta

The null treatment effect (default 0).

---

InferenceOrdinalGCompMeanDiff$compute_wald_confidence_interval()

Identical to $compute_asymp_confidence_interval() (both compute the same delta-method-based Wald interval, with the same bootstrap fallback); provided as an explicit alias for callers that want to name the Wald method directly rather than via the generic "asymptotic" dispatch.

Usage

InferenceOrdinalGCompMeanDiff$compute_wald_confidence_interval(alpha = 0.05)

Arguments

alpha

The significance level (default 0.05).

---

InferenceOrdinalGCompMeanDiff$compute_wald_two_sided_pval()

Identical to $compute_asymp_two_sided_pval() (both compute the same delta-method-based Wald p-value, with the same bootstrap fallback); provided as an explicit alias for callers that want to name the Wald method directly rather than via the generic "asymptotic" dispatch.

Usage

InferenceOrdinalGCompMeanDiff$compute_wald_two_sided_pval(delta = 0)

Arguments

delta

The null treatment effect (default 0).

---

InferenceOrdinalGCompMeanDiff$clone()

The objects of this class are cloneable with this method.

Usage

InferenceOrdinalGCompMeanDiff$clone(deep = FALSE)

Arguments

deep

Whether to make a deep clone.

Examples

# \donttest{
seq_des = DesignSeqOneByOneBernoulli$new(n = 10, response_type = 'ordinal')
for (i in 1:10) {
  seq_des$add_one_subject_to_experiment_and_assign(data.frame(x1 = rnorm(1)))
}
seq_des$add_all_subject_responses(sample(1:4, 10, replace = TRUE))
inf = InferenceOrdinalGCompMeanDiff$new(seq_des)
inf$compute_estimate()
#> [1] 0.5739362
# }