Simulate data from a causal zero-inflated beta (BIc) model

Draws potential outcomes from a one-boundary inflated beta regression (gamlss.dist::BEZI, inflation at zero) with a confounded treatment assignment, the data-generating process used in the simulation study of Ospina (2026). Useful for testing bic_gcomp(), bic_curvature() and friends against a model where the true average treatment effect and its discrete/continuous components are known in closed form.

Usage

simulate_bic(
  n,
  gamma = c(-1.5, 0.3, 1.5),
  beta = c(-1, 0.5, 1.2),
  phi = 4,
  rho = 0.4,
  seed = NULL
)

Arguments

n

Sample size.

gamma

Numeric length-3 vector (gamma0, gamma1, gammaT), the inflation-probability (logit) linear predictor coefficients for intercept, covariate W1 and treatment.

beta

Numeric length-3 vector (beta0, beta1, betaT), the continuous-mean (logit) linear predictor coefficients for intercept, covariate W2 and treatment.

phi

Precision of the continuous beta component (constant).

rho

Confounding strength: treatment is drawn with probability plogis(rho * (W1 + W2)), so rho = 0 is a randomized experiment and rho > 0 confounds treatment with the covariates that also drive the outcome.

seed

Optional seed for reproducibility.

Value

A data.frame with columns y (observed response), T (treatment), W1, W2 (covariates), and attributes "truth" (a list with the population tau, tau_alpha, tau_mu) and "params" (the generating coefficients).

Examples

dat <- simulate_bic(500, seed = 1)
attr(dat, "truth")
#> $tau
#> [1] 0.04573403
#> 
#> $tau_alpha
#> [1] -0.08662263
#> 
#> $tau_mu
#> [1] 0.1323567
#>