Poll aggregation · uncertainty
The Democratic candidates for governor, read the way a forecaster reads a race — every public poll combined into one picture, with the uncertainty out loud. It never names a winner, because the honest answer depends on assumptions you can move yourself.
A three-part lesson: a lead is not a lock · the witnesses disagree · you be the forecaster.
Loading polls…
Is this legit?
This page isn’t a black box. Here’s what election-forecasting research says about each signal — including where it’s weak. The short version: the polls are the strongest signal, not a certain one, and the skepticism about buzz, ad spend, and news tone isn’t ours — it’s the academic consensus.
Averaging many polls — counting newer and larger ones more — is a recognized technique. The catch: averaging removes random noise but not systematic bias. Polls can all lean the same way by about 2 points, so the polls are the strongest signal here, not a certain one.
Jackman 2005, Pooling the Polls ↗Shirani-Mehr et al. 2018, Bias & Variance in Election Polls ↗
Simulating the race thousands of times within the polling error is standard practice — it's exactly why a lead is not a lock. The error on the gap between two candidates is about double the margin of error you usually hear quoted.
Gelman et al. 2020, Information, Incentives & Goals in Election Forecasts ↗Pew 2016, Understanding the Margin of Error ↗
Mixing polls with non-poll “fundamentals” is a real forecasting tradition, but the accuracy gain is inconsistent. That's why the blend here is a teaching tool — you set the weights — and never prints a single number.
Abramowitz 2008, Time-for-Change ↗Heidemanns, Gelman & Morris 2020, The Economist model ↗
The claim that social-media buzz predicts vote share has been debunked repeatedly. This is why the page says buzz ≠ votes — that's the research's actual finding, not a disclaimer we added.
Gayo-Avello 2012, No, You Cannot Predict Elections with Twitter ↗Jungherr, Jürgens & Schoen 2012 ↗
Campaign spending's persuasive effect is close to zero in rigorous experiments, and winners raise more because they're already winning. So ad spend is a revealed belief — where a campaign thinks the vote is — not a result.
Kalla & Broockman 2018, The Minimal Persuasive Effects of Campaign Contact ↗Feigenbaum & Shelton 2013, The Vicious Cycle ↗
Coverage tone tracks the polls more than it moves them — it's largely a mirror of where a candidate already stands, not an independent predictor. So we show it, link every headline, and don't lean on it.
Primary electorates really are smaller and older than the general-election electorate. (The common claim that they're far more ideologically extreme is overstated.) That's why turnout is a slider, framed as illustrative.
Sides, Tausanovitch, Vavreck & Warshaw 2020, Representativeness of Primary Electorates ↗
The honest bottom line: the “move the weights, watch the leader change” tool is a way to think about a race and its uncertainty — not a calibrated statistical model. It never names a winner on purpose. See the site’s full methodology for how the underlying data is sourced and checked.