pradyuprasad/igm-consensus

IGM Forum Economist Consensus

Every IGM Forum poll from the Kent A. Clark Center for Global Markets (US, Europe, Finance panels), spanning 2011–2026, normalized to a uniform schema with a direction-blind consensus metric.

• Code & blog post: https://github.com/pradyuprasad/igm-consensus
• Interactive chart: https://pradyuprasad.com/writings/economists-agree/
• Source: https://kentclarkcenter.org/surveys/

Files

File	Rows	What it is
statements.csv	1,146	One row per poll question (text, date, panel, vote shares, source URLs).
votes.csv	50,410	One row per economist-question pair (vote, confidence, comment, affiliation).
statements_consensus.csv	1,146	statements.csv enriched with the HHI consensus index, dominant-direction score, percentile rank, and confidence-weighted variants.
methodology.md	—	Coverage stats, panel breakdown, quality checks.

The consensus metric

The chart in the linked blog post ranks questions by HHI (Herfindahl-Hirschman concentration) on the three vote shares:

HHI = share_agree² + share_uncertain² + share_disagree²

• 1.00 = unanimous; 0.33 = perfectly diffuse three-way split.
• Direction-blind — survives wording flips like "X is good" vs. "X is bad".
• Asymmetry-sensitive: scores 60/39/1 (settled) higher than 60/30/10 (real dissent), unlike top-share alone.

Coverage

• 558 poll pages, 0 failures, 0 duplicates
• 1,146 statements: US 644, Europe 377, Finance 125
• 50,410 votes from 193 distinct economists at top departments (MIT, Harvard, Chicago, Stanford, Yale, Berkeley, Princeton, Columbia, plus European/finance equivalents)
• 553/558 pages use the official "Download Poll Data" CSV as source of truth; 5 special crisis-rating pages use HTML chart shares.

See methodology.md for the full audit.

Quick start

from datasets import load_dataset

statements = load_dataset("pradyuprasad/igm-consensus", "statements_consensus", split="train")
print(statements[0])

Or with pandas:

import pandas as pd

url = "https://huggingface.co/datasets/pradyuprasad/igm-consensus/resolve/main/statements_consensus.csv"
df = pd.read_csv(url)
print(df.sort_values("consensus_score", ascending=False).head(10))

License

MIT for the schema and code that produced this dataset. Underlying poll data is from the IGM Forum — consult the Clark Center site for terms governing redistribution of the original surveys.