Study Design Architecture
Difference-in-Differences
• Identification Guarantee: Calibrated against cluster randomized & panel quasi-experiment power equations (Murray 1998; Abadie et al. 2010).
NIH Study Design Feasibility Verdict
High Statistical Feasibility & Confounder Robustness
Exceeds NIH/NCI minimum 80% statistical power threshold with solid parallel trends test (p > 0.20)
Statistical Power (1-β)
0.912
α = 0.05, Two-Tailed
MDES (% Change)
-8.42%
Min Detectable Effect
Parallel Trends p-val
0.482
Pre-policy Wald test
E-Value (Point / CI)
1.68 / 1.31
VanderWeele Sensitivity
—•— Treated Cohort —•— Control Cohort - - - Synthetic Counterfactual
Intervention Launch: Period 0 (T=4)
Power curve computed across effect sizes (-2% to -25%) given Jt=18, Jc=22, ICC=0.035. Red dash indicates 80% NIH benchmark.
E-value boundary defines the minimal risk ratio strength an unmeasured confounder must have with both treatment and outcome to explain away the observed effect.
| Period (t) | Treated Observed | Counterfactual | Control Base | Treatment Effect (τ) | 95% Conf. Interval | p-value |
|---|
Quasi-Experimental Rigor & Peer-Review Defense Assessment
Under the selected Difference-in-Differences design with 18 treated and 22 control clusters (total effective sample N = 100,000 across 8 waves), the proposed study achieves 91.2% power to detect a hypothesized -12.5% policy impact. The Minimum Detectable Effect Size (MDES) is -8.42%. Pre-intervention parallel trends show no statistically significant divergence (Wald test p = 0.482), and the point estimate E-value of 1.68 (lower CI bound 1.31) demonstrates robust resistance against unobserved social or clinical confounding.