Causal Structure & Model Diagnostics

Inspecting Directed Acyclic Graph (DAG) backdoor paths, SHAP feature attributions, refutation test results, and CATE income-tier heterogeneity.

Global ATE Impact
+0.5960
Deaths avoided per 1,000 live births
Monte Carlo Refutation (50-Seed)
[0.5716, 0.6703]
95% CI stability range
Placebo Test Status
PASSED (p < 0.001)
Treatment effect drops to zero
Confounder Conditioning
Blocked (W, X)
Two-way fixed effects applied

Causal Structure (DAG) Identification

DoWhy Identification Verified
Nutrition (T)
Primary Treatment
────── Backdoor Adjustment (W, X) ──────►
Direct CATE: +0.5960
Mortality (Y)
Under-5 Mortality
Blocked Confounders: Basic Sanitation, Literacy Rate, Log GDP Per Capita, Mortality Shock Index, Country & Year Fixed Effects.

EDGE WEIGHTS & REFUTATIONS

GDP ──► Mortality -0.41
GDP ──► Nutrition +0.68
Nutrition ──► Mortality (ATE) +0.5960
Refutation Proof: Placebo treatment test (p -> 0), Random Common Cause test passed, 50-seed Monte Carlo stability range [0.5716, 0.6703].

SHAP Feature Impact

Intervention Response by Income Tier (CATE)