Turning early votes
into election projections.
The Election Data Guy model analyses early voting returns against historical state voting patterns to estimate the eventual outcome of an election. Where party identification is unavailable, county-level data provides an additional layer of analysis.
A model built around the vote already cast.
Rather than relying on pre-election polling, EDG uses observed early voting data as the foundation of its projections. That data is evaluated in the context of previous election results and the voting patterns recorded within each state.
Observed early voting
The model starts with the available early vote returns, examining both raw vote totals and the distribution of votes between parties where identification is recorded.
Historical context
Current early voting patterns are compared with historical election results within the relevant state to assess how the present data differs from previous voting behaviour.
State-specific analysis
The interpretation of early vote data depends on the state. Party registration practices, county-level patterns and historical results affect how the available figures are assessed.
Estimated final outcome
The observed data is extrapolated to produce a projection of the eventual result. The projection can change as additional early voting returns become available.
The central principle: early voting figures are not the same as final election results. EDG uses the available figures and historical comparisons to estimate how the final vote may develop.
From raw returns to a projection.
The model follows a common analytical framework, with the treatment of incoming data adapted to whether party identification is available.
Collect
Gather the available early voting figures.
Compare
Assess vote shares and raw totals against historical patterns.
Extrapolate
Apply the relevant state methodology to estimate the final vote.
Project
Produce an estimated outcome that can evolve with new returns.
One framework. Two approaches to the data.
The availability of party identification determines how EDG interprets early voting returns before applying its projection methodology.
Where voter party registration is available.
In states where early voting records identify a voter's registered party, EDG compares the current distribution of early votes with historical voting patterns within the state.
The analysis considers changes in party vote share alongside changes in raw vote totals. These measures are evaluated against the state's historical election results to estimate how the early vote pattern may translate into the eventual outcome.
The resulting projection is an extrapolation from the available data, not a direct count of all votes that will ultimately be cast.
Important distinction: party registration identifies registration, not necessarily how an individual votes. The model estimates election outcomes; it does not observe each voter's ballot choice.
Where party identification is unavailable.
In states that do not provide usable party identification for early voters, EDG uses county-level voting data to estimate how the observed early vote may be distributed between parties.
County-level figures are assessed in the context of historical voting patterns to estimate a party allocation for the available early vote data. That allocation is then used as the input to the same underlying extrapolation framework used for states with party identification.
This approach allows the model to produce projections in states where party registration figures cannot provide a direct breakdown of the early vote.
Important distinction: party allocation in these states is estimated, not directly observed through individual voter registration records. This introduces additional uncertainty into the analysis.
Why the numbers can move.
Early voting is a developing dataset. New returns can change the observed party distribution and the relationship between current voting patterns and historical results.
Additional returns
New batches of early votes can materially change the totals and party shares available to the model.
Different voting patterns
The mix of voters participating early can vary over time. Early leads should not automatically be treated as a prediction of the final result.
Updated extrapolation
As the data changes, the comparison with historical patterns can change, producing a different estimated outcome.
A developing picture
Projections may become more informative as additional data arrives, although the amount of remaining uncertainty varies by state and race.
How a projection might evolve
Demonstration only. Values are hypothetical and do not represent actual EDG outputs, historical performance, a live race or a validated forecast.
What goes into the model — and what doesn't.
EDG's projections are based on a defined set of inputs. Understanding the boundaries of that analysis is essential to interpreting the results responsibly.
Early voting data
Available early vote figures, raw totals, party distributions where recorded, county-level data where needed, and historical election patterns.
Polling and discretionary adjustments
The methodology described here does not incorporate polling data or discretionary adjustments beyond the model's early-vote and historical-data framework.
Early vote limitations
Early voters may not be representative of all voters. Turnout patterns can differ across the early voting period and Election Day.
Data availability
Reporting practices, data completeness and the availability of party identification differ between states.
Estimation uncertainty
County-level party allocation involves estimation. Historical patterns may not fully reflect current voting behaviour.
Understanding the EDG model.
Does EDG use polling data?
Why can a projection change from one day to the next?
How does EDG handle states without party ID?
Does a party's early vote lead mean it will win?
Are EDG projections official election results?
Does this methodology guarantee accuracy?
Follow the data. Understand the projection.
EDG exists to make early voting data easier to understand and to explore what it may reveal about the direction of American elections. Every projection should be interpreted in the context of its available data, assumptions and limitations.