Urban Mobility Data for Causal Inference (Synthetic Spatiotemporal Dataset)
Description
This data is a synthetic spatiotemporal traffic dataset designed to serve as a benchmark for causal inference methods under observed confounding. It simulates traffic dynamics over a 10x10 spatial grid across 90 days at 15-minute intervals (all adjustable), yielding 864,000 observations. The treatment variable, which is the congestion toll variable, is assigned via a probabilistic mechanism that depends on observable covariates, including peak hours, weekends, zone type, and road quality, thereby embedding a known confounding structure. The outcome variable "final delay" is generated from base delay modified by the treatment effect, rainfall, and event shocks. Because the data are fully synthetic, the ground truth treatment effects (0.10, 0.15, and 0.12 for residential, commercial, and mixed zones, respectively) are known to researchers, making the dataset suitable for evaluating whether causal estimators such as inverse probability weighting, regression adjustment, and causal forests can recover true effects in the presence of confounding. The dataset is provided as a CSV file, and the generation code is publicly available on GitHub.
Files
Institutions
- Kyungsung UniversityBusan, Busan