The Decision Desk HQ (DDHQ) Election Pulse (EP) provides real-time updates on election night, offering insight into the likelihood of outcomes as votes are reported. The EP starts with a pre-election forecast and adjusts continuously based on incoming vote tallies and patterns. It incorporates both reported votes and estimates of remaining uncounted ballots to refine its predictions, providing a dynamic look at each race as election night unfolds. A predecessor model, which focused primarily on primary elections, was published in the Harvard Data Science Review (HDSR) in October 2024. The general election version incorporates additional logic based on the insights from that earlier work. Model predictions are shown on a delay of X minutes from the live version.
Viewers can follow along with the Election Pulse on the Decision Desk HQ live election results website.
Election Pulse Race Coverage
The DDHQ Election Pulse operates in real-time for most statewide races, including Presidential and Senate contests, but it does not cover many safe states where incoming data streams do not provide enough information to generate accurate predictions. For states such as AZ/NV, predictions on election night are not expected to shift significantly, as many votes will not be counted until after election night. Topline estimates for most states will update throughout the night but are expected to stop around 3-4 AM ET, when vote reporting concludes for the evening.
Topline Probabilities and Control of Government
DDHQ Election Pulse offers real-time “topline” probabilities for several critical outcomes:
- The likelihood of either Trump or Harris winning the Electoral College.
- The probability of Democrats or Republicans gaining control of the US Senate.
- The likelihood of Democrats or Republicans gaining control of the US House.
These probabilities are based on race-level models and race calls for the Senate and Presidential elections, while House estimates rely on race calls. Like the race-specific estimates, the toplines are expected to stop updating when reporting concludes for the night around 3-4 AM ET.
Independence from Race Calls
While the EP provides valuable real-time insights, it operates independently of DDHQ’s race call process. A high win probability in the EP does not automatically trigger a race call, and the race call team may declare a winner based on information not considered by the EP. The race call team has access to direct communications with election officials and other proprietary data that allow for more informed decisions, which may occasionally differ from the EP’s predictions.
Interpreting Changes in the Election Pulse
The EP evaluates candidates’ performances relative to pre-set expectations rather than focusing solely on vote counts. For example, even if a candidate gains raw votes in a specific update, they may lose ground in the EP if their performance falls short of forecasted benchmarks. This allows for a nuanced view of election dynamics beyond simple vote totals. The model accounts for geospatial and demographic data, recognizing patterns in areas that share similar voter profiles.
Adjusting for Reporting and Ballot Types
The DDHQ Election Pulse is designed to handle complex vote reporting patterns, particularly those caused by absentee or mail-in ballots. As voter behavior has shifted in response to new election laws, particularly with the increase in absentee voting, vote counts on election night (especially early on election night in partially reported counties) often do not reflect the final outcome. For example, a candidate who initially appears to be losing may still be favored in the EP if late-counted ballots are expected to shift the race. The EP adjusts for these patterns, factoring in vote types and the timing of reporting to provide a clearer picture of the race.
Special Considerations for Arizona and Nevada
In Arizona and Nevada, where a large portion of votes comes in through mail ballots after Election Day, the shifting patterns between early and late returns can be unpredictable. As a result, the models displayed here are primarily driven by pre-election priors and may not update significantly on election night. The fluid nature of ballot counting in these states means that meaningful changes to the predictions are unlikely on Tuesday Night.
To account for this, topline probabilities will be informed by prior polling and historical correlations, ensuring that even without informative vote data on Election Night in AZ and NV, the projections are accurate. In states with EP models, front-end predictions remain static until votes start reporting.