DGP Dossier: synth_placebo_tilted

1. Identity & Status

DGP ID	synth_placebo_tilted
Version	1.4.0
Status	experimental
Difficulty	★★★★☆ (4/5)
Stress Profile	overlap: moderatenoise: gaussianlinearity: lineareffect: constanttarget: both

2. What This DGP Stresses (Intent)

Intent: The “Spurious Correlation Trap.” This DGP induces strong (but manageable) linear confounding where treated units have naturally higher outcomes than control units (\(p(X)\) aligns with \(\mu_0(X)\)).

The Scientific Question: “Can the estimator distinguish Selection Bias from Treatment Effect?” A naive comparison suggests a strong positive effect. The estimator must vigorously adjust for covariates to recover the true zero. This tests resistance to “Bias Leakage.”

3. Identification Assumptions (Explicit)

• Selection on observables holds.
• Overlap is moderately tilted (propensity shifted, but not extreme).
• SUTVA holds.
• Sharp Null: \(Y_1 \equiv Y_0\).

4. Mathematical Specification

Covariates: \(X_1, X_2, X_3, X_4, X_5 \sim \mathcal{N}(0, 1^2)\) independently.

Outcome (control): \[\mu_0(X) = 1 + X_1 + 0.5 X_2\]

Propensity (tilted): \[p(X) = \text{expit}(0.6 X_1 + 0.8 X_2)\]

Treatment effect (sharp null): \[\tau(X) \equiv 0\]

Noise: \[\varepsilon \sim \mathcal{N}(0, 0.5^2)\] (Shared noise \(Y_1 \equiv Y_0\))

5. Oracle Truth Definition

• True ATT is zero by construction.
• True QST is identically zero on the canonical grid.

6. Visual Diagnostics (n = 5000)

7. Empirical Validation

We validate only the oracle estimator to keep builds fast.

Oracle validation (n=1000, seeds=1:20)

mean_bias	rmse
0	0

8. Failure Mode Summary

• Bias Leakage: The primary failure mode is reporting a positive effect (Type I error) because the estimator failed to fully scrub the confounding.
• Under-adjustment: Regularized estimators (like Lasso) might over-shrink coefficients, leaving residual confounding that masquerades as a treatment effect.

9. Implementation Reference

• Code: R/dgp-synth-placebo-tilted.R (v1.4.0)
• Registry: cs_dgp_registry() entry for synth_placebo_tilted
• Oracle: true_att = 0, true_qst = 0
• Oracle QST: cs_get_oracle_qst("synth_placebo_tilted", version = "1.4.0")

10. Validation Checklist

• Sharp null enforced (\(Y_1 \equiv Y_0\))
• Structural ATT = 0
• QST truth is identically zero
• Oracle validation passes

11. Changelog

• v1.0 dossier: Initial placebo dossier
• v1.4.0: Reduced tilt coefficients to 0.6/0.8. v1.3.0 deprecated to avoid confounding variance explosion with bias.

Appendix: Implementation

Source code for: dgp_synth_placebo_tilted_v140

function (n, seed = NULL, include_truth = TRUE, oracle_only = FALSE) 
{
    if (!is.null(seed)) {
        cs_set_rng(seed)
    }
    X1 <- stats::rnorm(n, mean = 0, sd = 1)
    X2 <- stats::rnorm(n, mean = 0, sd = 1)
    X3 <- stats::rnorm(n, mean = 0, sd = 1)
    X4 <- stats::rnorm(n, mean = 0, sd = 1)
    X5 <- stats::rnorm(n, mean = 0, sd = 1)
    mu0 <- 1 + X1 + 0.5 * X2
    tau <- rep(0, n)
    p <- stats::plogis(0.6 * X1 + 0.8 * X2)
    w <- stats::rbinom(n, size = 1L, prob = p)
    eps <- stats::rnorm(n, mean = 0, sd = 0.5)
    y0 <- mu0 + eps
    y1 <- y0
    if (isTRUE(oracle_only)) {
        return(list(df = tibble::tibble(w = w, y0 = y0, y1 = y1)))
    }
    y <- ifelse(w == 1L, y1, y0)
    true_att <- cs_true_att(structural_te = tau, w = w)
    true_qst <- if (isTRUE(include_truth)) 
        cs_get_oracle_qst("synth_placebo_tilted", version = "1.4.0")
    else NULL
    out <- list(df = tibble::tibble(y = y, w = w, y0 = y0, y1 = y1, 
        p = p, structural_te = tau, X1 = X1, X2 = X2, X3 = X3, 
        X4 = X4, X5 = X5), true_att = true_att, true_qst = true_qst, 
        meta = list(dgp_id = "synth_placebo_tilted", version = "1.4.0", 
            type = "synthetic", params = list(n = n, seed = seed), 
            structural_te = tau))
    cs_check_dgp_synthetic(out)
    out
}
<bytecode: 0x55e18b30d470>
<environment: namespace:CausalStress>