Data supporting "Prescription drugs at the core of darknet drug markets"

This dataset contains derived data underlying the figures, tables, and reported statistical analyses in the accompanying manuscript, “Prescription drugs at the core of darknet drug markets.” The data are derived from four darknet markets crawled between June 2021 and January 2022. The deposit does not contain raw marketplace crawls, listing text, URLs, seller handles, buyer or reviewer identifiers, or exact seller shipping routes. These data are excluded because they were provided for research use and are not approved for public redistribution. The captured four-market drug corpus comprises 402,779 review records associated with 40,380 reconstructed listing proxies and 3,209 market-specific vendor accounts. Analyses requiring valid vendor and product classifications use a filtered analytical sample of 400,441 reviews, 39,860 listing proxies, and 3,195 vendor accounts. Individual files may therefore have different sample bases depending on their unit of analysis and model eligibility criteria. Contents source_data/ contains the values underlying the main and supplementary figures. analysis_results/ contains reported coefficient tables, null-model results, and aggregate portfolio summaries. networks/ contains product-level GraphML networks. Nodes represent product labels and edges represent statistically disproportionate co-offering by vendors. documentation/ contains the product-category mapping and additional documentation. MANIFEST.csv describes every deposited file and provides SHA-256 checksums. Units of analysis Product-market: one product label observed in one market. Product tie: one product pair in one market. Vendor-product: one anonymized seller account and product combination in one market. Market-category: one broad product category in one market. Null draw: one randomized value for one market. Market-specific vendor accounts are treated as separate observations. Product categories correspond to the platform-derived categories used in the study. In particular, Prescription is distinct from the separately coded Benzos and Opioids categories. Reuse All CSV files are UTF-8 encoded and include a header row. Missing values are represented by empty fields. GraphML files can be opened using common network-analysis software and graph libraries in Python or R.