MetaCausal JSS replication: ACIC benchmark data and pre-computed ensemble results

Replication data and pre-computed results for the Journal of Statistical Software manuscript: MetaCausal: Ensemble Estimation of Causal Effects in Python. Contents: acic_data.zip — Raw ACIC benchmark input data (subset used by the manuscript): ACIC 2017 (32 DGPs × 3 seeds) and ACIC 2019 low-dimensional track (16 DGPs × 6 seeds). metacausal_results.zip — Pre-computed ensemble outputs. Extracting this archive and pointing the summarise / plot scripts at it reproduces all tables and figures without re-fitting the models. See the manuscript replication guide (Code/replicate_all.py) for usage instructions. Update. In the Section 5.2 CI-coverage experiment, EconML's CausalForestDML is now excluded from the bootstrap pool: its generalized random forest intermittently crashes (segmentation fault) under the repeated model refits the bootstrap performs, a latent and as-yet-unresolved upstream issue (EconML #470). The acic2019lo_nonparametric.csv and acic2019lo_subsample.csv files in metacausal_results.zip are therefore superseded; the corrected versions ship directly in the replication package. For reproduction, only the ACIC 2017 CATE parquet files in this archive are required (the Section 5.1 PEHE table); the acic2019lo_*.csv files here can be disregarded.