G-Computation Risk-Difference Inference for KK Designs with Binary Responses

Fits an all-subject logistic working model \(\mathrm{logit}\,P(Y=1\mid w,x) = \beta_0 + \beta_T w + \beta_X^\top x\) for a KK incidence outcome using treatment \(w\) and, optionally, all recorded covariates \(x\), then estimates the marginal (standardized, g-computation) risk difference \(\hat\theta = n^{-1}\sum_i \{\hat p(1, x_i) - \hat p(0, x_i)\}\) by averaging the fitted-model predicted risks under all-treated and all-control assignments over the empirical covariate distribution (Robins 1986). Matched pairs are treated as clusters and reservoir subjects are treated as singletons when computing the sandwich covariance of the standardized estimator (the delta-method variance of the empirical mean of the two counterfactual-risk contrasts, not the naive logistic-regression coefficient variance).

Computes the randomization distribution of the treatment effect estimate under the sharp null.

Whether compute_rand_two_sided_pval() is actually usable on this instance right now – FALSE exactly when it would stop(): an incidence-response instance with no custom randomization statistic and a design not eligible for design-randomization-based incidence inference (see private$should_use_design_randomization_for_incidence()). TRUE for every other case, including every non-incidence response type. Public, self-contained (only reads already-set instance state, no side effects), so InferenceSuite can check this before attempting the sentinel instead of relying on the stop() being silently swallowed into a pval = NA "ok" row – the single source of truth for both this check and compute_rand_two_sided_pval()'s own guard, so the two can never drift apart.

Value

When debug = FALSE (default), a numeric vector of length r. When debug = TRUE, a list with: values, errors (list of character vectors, one per iteration), warnings (list of character vectors, one per iteration), num_errors, num_warnings, prop_iterations_with_errors, prop_iterations_with_warnings, and prop_illegal_values.

A single logical.

Details

This estimator has likelihood_tier = "none": the fitted logistic model is a working model for standardization only, and reported inference is sandwich/bootstrap-based, not likelihood-based. compute_estimate() fits the model and returns \(\hat\theta\) on the risk-difference (probability) scale; compute_estimate_with_bootstrap_weights() refits under Bayesian-bootstrap subject weights for compute_bayesian_bootstrap_confidence_interval(). Those weights affect the working-model fit only: the resulting predictions are still averaged equally over the retained empirical covariate rows, so every g-computation implementation in EDI targets the same sample-conditional standardized estimand. Jackknife deletes one cluster (matched pair or singleton reservoir subject) at a time. If the working model fails to converge or the design has no treatment-arm variation, the estimate is marked non-estimable via is_nonestimable().

References

Robins, J. (1986). A new approach to causal inference in mortality studies with a sustained exposure period. Mathematical Modelling, 7(9-12), 1393-1512. doi:10.1016/0270-0255(86)90088-6

See also

InferenceIncidKKGCompRiskRatio for the risk-ratio analog on the same standardization machinery.

Super class

Inference -> InferenceIncidKKGCompRiskDiff

Methods

Public methods

• InferenceIncidKKGCompRiskDiff$approximate_randomization_distribution_beta_hat_T()

• InferenceIncidKKGCompRiskDiff$supports_rand_pval_for_incidence()

• InferenceIncidKKGCompRiskDiff$clone()

+ inherited public methods from Inference

• Inference$capabilities()
• Inference$compute_asymp_confidence_interval()
• Inference$compute_asymp_two_sided_pval()
• Inference$compute_estimate()
• 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$get_treatment()
• Inference$initialize()
• Inference$is_nonestimable()
• Inference$set_optimization_alg()
• Inference$set_seed()
• Inference$supports()

---

InferenceIncidKKGCompRiskDiff$approximate_randomization_distribution_beta_hat_T()

Usage

InferenceIncidKKGCompRiskDiff$approximate_randomization_distribution_beta_hat_T(
  r = 501,
  delta = 0,
  transform_responses = "none",
  show_progress = TRUE,
  permutations = NULL,
  debug = FALSE,
  zero_one_logit_clamp = .Machine$double.eps
)

Arguments

r

Number of randomization vectors. Default 501.

delta

The null difference. Default 0.

transform_responses

Type of transformation. Default "none".

show_progress

Show progress bar. Default TRUE.

permutations

Pre-computed permutations. Default NULL.

debug

If TRUE, return a list with the distribution values and per-iteration diagnostics including error messages, warning messages, counts of each, and summary proportions for iterations with errors, warnings, and illegal (non-finite) values. Runs serially. Default FALSE.

zero_one_logit_clamp

The clamping amount for exact 0 and 1 values when logging

---

InferenceIncidKKGCompRiskDiff$supports_rand_pval_for_incidence()

Usage

InferenceIncidKKGCompRiskDiff$supports_rand_pval_for_incidence()

---

InferenceIncidKKGCompRiskDiff$clone()

The objects of this class are cloneable with this method.

Usage

InferenceIncidKKGCompRiskDiff$clone(deep = FALSE)

Arguments

deep

Whether to make a deep clone.

Examples

# \donttest{
seq_des = DesignSeqOneByOneKK14$new(n = 10, response_type = 'incidence')
for (i in 1:10) {
  seq_des$add_one_subject_to_experiment_and_assign(data.frame(x1 = rnorm(1), x2 = rnorm(1)))
}
seq_des$add_all_subject_responses(rbinom(10, 1, 0.5))
inf = InferenceIncidKKGCompRiskDiff$new(seq_des)
inf$compute_estimate()
#> [1] -0.1763281
# }