Completed Project

Independent Research Project

Quasi-Causal Analysis of Dispatch Delays and Late-Delivery Risk in E-Commerce

A quasi-causal analysis of how dispatch-deadline breaches relate to late-delivery risk in e-commerce, using augmented inverse probability weighting with five-fold cross-fitting on real Olist order data.

Dates
December 2025 – January 2026

Research Problem

Late deliveries in e-commerce are often attributed to dispatch delays, but isolating that relationship from confounding operational factors is difficult. This project applies a quasi-causal design — augmented inverse probability weighting with cross-fitting — to estimate the adjusted association between dispatch-deadline breaches and late customer delivery on real marketplace orders, without claiming a definitive causal effect.

Methodology

Causal InferenceAugmented Inverse Probability WeightingFive-Fold Cross-FittingLeakage-Safe Seller HistoriesPre-Treatment Operational CovariatesOverlap AnalysisCovariate-Balance AnalysisPlacebo AnalysisTemporal and Cohort Robustness ChecksAutomated TestsGitHub Actions CI

Tools

Python

Verified Data

  • Analysed 81,941 Olist orders
  • Defined dispatch-deadline breach as the treatment
  • Defined late customer delivery as the outcome

Verified Results

  • Estimated a 14.11 percentage-point higher adjusted late-delivery risk
  • 95% confidence interval: 12.81–15.41 percentage points
  • Maximum weighted absolute standardised mean difference: 0.053

Current Stage

Completed independent research project