Replication Data for Causal Inference with Generative AI and Text Treatments

Replication package for the paper 'Causal Inference with Generative Artificial Intelligence: Application to Texts as Treatments'. The data includes internal representations and propensity score estimates used to generate a figure in the original research. The package is hosted by Harvard Dataverse and was last updated on June 9, 2026.

Use Cases

• Replicate causal inference analyses based on the provided internal representations.
• Validate propensity score estimation methods using the included `ps_save.csv` file.
• Benchmark generative AI models for text-as-treatment applications.
• Study the relationship between text features and causal effect estimates.

Strengths

• Includes the specific propensity score estimates (`ps_save.csv`) used to generate a published figure.
• Contains internal representations used for replication, which may not be available in the public code repository.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count and overall data scale are unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

Harvard Dataverse, author Nakamura, Kentaro.

Collection Method

Created as a replication package for an academic paper.

Freshness

Last updated 2026-06-09 16:53:55; freshness should be verified.