Founding spots close at launch in Dec 2026. Join free →

Institutional Library

Resources

The research our method stands on, including the papers that argue against it, and the frameworks that govern how Precision Polling and Research works.

Research Library

Do Prediction Markets Beat Polls?

Thirty years of studies on election-market accuracy, with infographics and links to every paper. Read the research →

Annotated Reading List

The research behind the method

The literature that matters here is small, and it does four jobs: comparing markets with polls, combining them, paying people for honest answers, and scoring the result. We list the papers that cut against us in their own section, because a reading list that only quotes the wins would look curated. Every figure below was checked against the source.

1 · Markets Against Polls

The record that made election markets worth taking seriously, and the papers that explain what a market price does and does not mean.

  • 2008Abstract free · full text via journal

    Prediction Market Accuracy in the Long Run

    Joyce Berg, Forrest Nelson & Thomas Rietz (University of Iowa) · International Journal of Forecasting 24(2): 283–298

    The Iowa Electronic Markets against 964 polls across five presidential elections, 1988–2004. The market was closer to the result 74% of the time, and significantly better in every election when forecasting more than 100 days out.

  • 2008Free to read

    Results from a Dozen Years of Election Futures Markets Research

    Joyce Berg, Robert Forsythe, Forrest Nelson & Thomas Rietz · Handbook of Experimental Economics Results, Vol. 1, ch. 80 (2003 working draft linked)

    The earlier vote-share comparison: across 49 Iowa markets, election-eve prices missed by 1.49 points on average against 1.91 for final polls.

  • 2004Free to read

    Prediction Markets

    Justin Wolfers & Eric Zitzewitz · Journal of Economic Perspectives 18(2): 107–126

    The survey economists still assign. Over four presidential elections (1988–2000), Iowa prices in the final week missed the vote share by about 1.5 points on average, against 2.1 for the final Gallup poll.

  • 2006Free to read

    Interpreting Prediction Market Prices as Probabilities

    Justin Wolfers & Eric Zitzewitz · NBER Working Paper 12200 / FRBSF Working Paper 2006-11

    The reply to Manski, below. Under a broad class of models a market price is usually close to traders' average belief: a useful, sometimes biased, estimate of a probability.

  • 2006Free to read

    Interpreting the Predictions of Prediction Markets

    Charles Manski (Northwestern) · Economics Letters 91(3): 425–429 (free as NBER Working Paper 10359)

    The objection: a price reveals nothing about how spread out traders' beliefs are, and pins the average belief down only to a range around the price.

    Our answer: We never present a single market price as the probability. We average three markets with the polls, and show each source next to the blend.

  • 2009Free to read

    Forecasting Elections: Comparing Prediction Markets, Polls, and Their Biases

    David Rothschild · Public Opinion Quarterly 73(5): 895–916

    In 2008 state races, raw Intrade prices were slightly less accurate on average than FiveThirtyEight's debiased poll model. Once both were debiased, markets gave more accurate win probabilities early in the cycle and in races that were not yet certain. The case for correcting each source before blending, not for trusting raw prices.

  • 2016Free to read

    Forecasting Elections

    Leighton Vaughan Williams & J. James Reade · Journal of Forecasting 35(4): 308–328

    The British counterpart: polls, betting markets, expert opinion and a statistical model compared. Markets gave the most precise forecasts and improved as the election neared, while polls did worse; markets and bias-corrected polls were similar on bias.

  • 2008Free to read

    The Promise of Prediction Markets

    Kenneth Arrow, Robert Forsythe, Robin Hanson, Robert Shiller, Vernon Smith, Justin Wolfers et al. · Science 320(5878): 877–878

    A two-page statement by 22 economists, not a study. Useful as evidence the profession took the tool seriously, not as evidence that it works.

2 · The Papers That Cut Against Us

A reading list that quoted only the wins would not survive a serious reader. These are the strongest objections to market forecasts, and how our method responds to each.

  • 2008Free to read

    Are Political Markets Really Superior to Polls as Election Predictors?

    Robert Erikson (Columbia) & Christopher Wlezien (Temple) · Public Opinion Quarterly 72(2): 190–215

    The strongest academic objection to the Iowa record. Once poll leads are discounted for the time left in the campaign, poll-based forecasts beat Iowa vote-share prices, and poll-based win projections beat winner-take-all prices. Traders acted as if campaigns could still move the race more than they do.

    Our answer: We do not use raw poll leads. Our poll input turns a lead into a win chance with a 6-point allowance for a typical polling miss, and in races with markets polls typically carry about a third of the blend or more, so a market that overrates how much the race can still move is pulled back toward the polls. The allowance does not yet shrink as Election Day nears; that is the refinement this paper argues for.

  • 2012Free to read

    Markets vs. Polls as Election Predictors: An Historical Assessment

    Robert Erikson & Christopher Wlezien · Electoral Studies 31(3): 532–539

    Wall Street betting prices from 1880 to 1960, and online markets since 1992. Before scientific polls, prices were good predictors, almost on par with polls since. Once polls existed, last-week prices added nothing beyond the final trial-heat polls, and in 1948 the markets followed the polls in making Dewey the favorite.

    Our answer: If late prices mostly shadow the polls, a blend loses little by including them, and markets still carry the early-campaign signal Berg and Rothschild found. We score the blend and every input separately on our public scorecard, so this can be tested on our own record.

  • 2017Abstract free · full text via journal

    Distilling the Wisdom of Crowds: Prediction Markets vs. Prediction Polls

    Pavel Atanasov, Philip Tetlock, Barbara Mellers, Lyle Ungar et al. (Penn) · Management Science 63(3): 691–706

    More than 2,400 forecasters on 261 geopolitical questions. Market prices beat a simple average of individual forecasts. Team forecasts, aggregated with recency weighting, extra weight for proven forecasters and recalibration, beat the markets.

    Our answer: This is the case for our member panel. A scored panel, weighted by track record, can beat a market. We add panel results to the blend only once they have a published scored history.

  • 2016Free to read

    Trading Strategies and Market Microstructure: Evidence from a Prediction Market

    David Rothschild & Rajiv Sethi · Journal of Prediction Markets 10(1): 1–29

    The 2012 "Romney whale": one Intrade account placed more than a third of all bets on Romney to win, and lost close to $7 million.

    Our answer: A market with under $10,000 traded, or one that jumps 15 points in a day on thin volume, is left out of our blend, the rest count in proportion to the money behind them, and each market's own price is shown beside the combined number rather than replaced by it.

3 · Combining and Scoring Forecasts

Our actual claim is not that markets win. It is that a weighted blend of independent sources beats any one of them, and that every forecast should be scored in public.

4 · Paying for Honest Answers

The research behind the Precision Candor Score™, our implementation of Bayesian Truth Serum, and behind scoring a panel by its track record.

  • 2004Free to read

    A Bayesian Truth Serum for Subjective Data

    Drazen Prelec (MIT Sloan) · Science 306(5695): 462–466

    The scoring rule. Each respondent gives an answer and a prediction of how others will answer. Truthful answers maximize the expected score, even for respondents who believe their answer is a minority view.

  • 2013Free to read

    Creating Truth-Telling Incentives with the Bayesian Truth Serum

    Ray Weaver & Drazen Prelec · Journal of Marketing Research 50(3): 289–302

    Recognition questions salted with made-up items. In one experiment, people paid by the rule claimed to recognize 13% of the fakes, against 24% for controls. When also paid to exaggerate, they claimed 16%, against 55% for people paid to exaggerate without the rule. A lab result on quiz items, not an election.

  • 2017Abstract free · full text via journal

    A Solution to the Single-Question Crowd Wisdom Problem

    Drazen Prelec, H. Sebastian Seung & John McCoy (MIT) · Nature 541(7638): 532–535

    The "surprisingly popular" rule: pick the answer that is more common than people predicted, not the majority answer. Across 490 items it cut errors by 21.3% against majority vote. The Candor Score uses the same two questions to pay the member; this paper uses them to pick the answer.

  • 2017Free to read

    Validating Bayesian Truth Serum in Large-Scale Online Human Experiments

    Morgan Frank, Manuel Cebrian, Galen Pickard & Iyad Rahwan · PLOS ONE 12(5): e0177385

    The test outside the classroom: three online experiments with thousands of paid participants. Where the honest answer was known, the rule moved reports toward it; in a coin-flip task, reported tails rose from 43% to 47%, against 50% expected.

  • 2005Book

    Expert Political Judgment: How Good Is It? How Can We Know?

    Philip Tetlock (Penn) · Princeton University Press

    The finding that matters here is not that experts are bad. It is that forecast accuracy can be measured, and that some forecasters are reliably better than others.

  • 2015Book

    Superforecasting: The Art and Science of Prediction

    Philip Tetlock & Dan Gardner · Crown

    The popular account of the government-funded forecasting tournament: training, teams and track records beat a crowd average.

  • 2014Free to read

    Forecasting Tournaments: Tools for Increasing Transparency and Improving the Quality of Debate

    Philip Tetlock, Barbara Mellers, Nick Rohrbaugh & Eva Chen · Current Directions in Psychological Science 23(4): 290–295

    The short academic write-up of the tournament and why keeping score improves forecasts.

  • 2021Free to read

    Superforecasters: A Decade of Stochastic Dominance

    Christopher Karvetski (Good Judgment Inc) · Good Judgment Inc white paper

    On 108 questions forecast by both groups, superforecasters' daily scores were 35.9% more accurate than the open crowd's. A company's own paper, so read it as a claim.

  • 1995Free to read

    Idea Futures

    Robin Hanson (George Mason) · Wired 3.09

    The readable origin of the research program Hanson started in 1988: settle disputed questions with betting markets. Linked to his own page on the idea.

5 · Why Polls Miss, and How a Panel Is Weighted

The evidence on what is going wrong with polling, and the method we will be asked about for an opt-in, verified panel.

Talks Worth an Hour

  • Finding Truth When Most People Are Wrong

    Drazen Prelec · IMéRA, Aix-Marseille · October 2017

    Prelec walks through the surprisingly popular rule in plain language.

  • Superforecasting

    Philip Tetlock · The Long Now Foundation · November 2015

    The best public account of the forecasting tournament and why a scored crowd beats pundits.

  • Take the Gold!

    Robin Hanson · Manifest 2023, Berkeley

    What prediction markets have and have not become. The market side, not the panel side.

What the Literature Does Not Cover
  • No academic literature exists on survey-reward tokens. The nearest evidence is the Bayesian Truth Serum payment studies above and the scored forecasting tournaments at Penn. That is a gap, and we say so rather than fill it.
  • The Bayesian Truth Serum results come from quiz and recognition tasks, not election questions. We will publish our own panel results, with the scoring rule held fixed, once the panel is in the field.
  • Most market-versus-poll studies use the Iowa Electronic Markets and Intrade. Evidence on today's larger markets (Kalshi, Polymarket) is still thin, which is one reason we score every source on our own scorecard.

Precision White Papers

Our own frameworks

Precision White Paper No. 1 · 2026

Signal Architecture

A Framework for Independent Political Research in the Modern Information Environment

Traditional polling treats public opinion as a single signal captured at a single moment. This paper describes the framework Precision Polling and Research uses instead: a layered signal architecture that reads survey research, media momentum, and organic engagement as distinct instruments — each with known strengths, known failure modes, and a defined role in the final estimate.

Read the Paper
Precision White Paper No. 2 · 2026

Engagement Protocol

Client Onboarding, Methodological Integrity & Research Partnership Standards

This document sets out how Precision Polling and Research engages with clients: how engagements begin, what clients can and cannot influence, how conflicts are handled, and what both parties owe each other. It is published so that prospective clients, respondents, and the public can see the rules before any engagement begins.

Read the Paper

White papers are published for transparency and professional reference. They may be cited with attribution to Precision Polling and Research. Reading-list summaries are our own paraphrase of each source; follow the links for the full text.

Founding spots close at launch · December 2026

Join free. Earn Precision Tokens from day one.

The founding-member window closes when the panel launches in December 2026. Join before then and your founding-member Precision Tokens are credited at launch.

Reserve my founding spot
  1. 1

    Take polls

    Earn Precision Tokens for every member poll you complete, on top of the cash fee after launch.

  2. 2

    Answer honestly

    The truth bonus scores each answer and can double that poll's tokens.

  3. 3

    Bring friends

    Earn tokens for each friend who joins with your link and verifies. They earn tokens too.

Precision Tokens are locked during pre-registration: they cannot be sold, transferred or redeemed and have no cash value. Amounts and rules are subject to final terms. Terms