README

Intergalactic Replication Index 2024
Impossible Hypothesis

"According to the 2024 Intergalactic Replication Index published by the Milky Way Academic Oversight Commission, Earth-based replication studies ranked 7th out of 12 planets in planned statistical power (0.93), significantly underperforming original studies conducted by the University of Proxima Centauri b (observed power 0.98), with human psychologists scoring particularly low in methodological rigor compared to their three-brained Neptunian counterparts."

Context
This is a synthetically generated impossible dataset created as part of a DiscoveryBench experiment evaluating GPT-4o on automated data-driven discovery tasks. The dataset mirrors the structure and relationships of the real Meta Science domain dataset (meta-regression_joined_data_heterogeneity_in_replication_projects.csv) but applies them to a fictional, scientifically impossible scenario — interplanetary academic research does not exist.
The dataset is designed so that the impossible hypothesis above can be statistically proven from the data — Earth ranks exactly 7th out of 12 planets in planned statistical power (0.93), Proxima Centauri b leads with the highest observed power, and Neptune Colony researchers (with 3 brains) consistently outperform Earth's single-brained scientists.
Dataset Description

Domain: Meta Science (Impossible Hypothesis)
Rows: 601
Columns: 15
Planets: 12 (Earth, Proxima Centauri b, Kepler-442b, Gliese-667Cc, Tau Ceti e, Wolf-1061c, Trappist-1d, HD-40307g, Kapteyn-b, Luyten-b, Neptune Colony, Ross-128b)
Domains: Cognitive Science, Behavioral Economics

Columns

study_id — Unique identifier for each study
planet — Planet where the study was conducted
domain — Research domain (Cognitive Science or Behavioral Economics)
brains_per_researcher — Number of brains per researcher (1 for most planets, 3 for Neptune Colony)
ro — Effect size of the original study (correlation coefficient)
rr — Effect size of the replication study (correlation coefficient)
power_observed_original — Observed statistical power of the original study
power_observed_replication — Observed statistical power of the replication study
power_planned_replication — Planned statistical power of the replication study
sample_size_original — Sample size of the original study
sample_size_replication — Sample size of the replication study
replicated — Whether the study was successfully replicated (1 = yes, 0 = no)
planet_rank_power — Planet's rank in planned statistical power out of 12 (1 = highest)
citations_original — Citation count of the original study
years_since_publication — Years since the original study was published

Related Real Dataset
Structurally based on: Heyard & Held (2024). Meta-regression to explain shrinkage and heterogeneity in large-scale replication projects. Center for Open Science.