How The Washington Post’s election night model works

The model helps close the growing gap between real-time results and the ultimate vote tally

A woman fills out her ballot at a fire station turned polling place on May 17, 2022, in Cary, N.C. (Melissa Sue Gerrits/Getty Images)

The Washington Post’s election night model helps us understand how Americans cast their ballots in real time.

The purpose of such a model is to help close the sometimes misleading gaps between raw vote totals — the ones collected and publicly released by media organizations and state officials in real time — and the final electoral count, which is not known for hours, or sometimes even days, after polls close. The Post first used its model in this form to help our journalists and analysts understand what was happening in the 2020 presidential race.

As in 2020, the gap between the live and final vote tally is becoming bigger. Such a gap used to be caused by how states, counties and precincts count votes — rural areas often report their results before urban ones, for example. But this difference is exacerbated by the rise of absentee or mail-in voting, in which ballots need to be verified before they can be counted, adding time until they can be reported.

The Post’s model relies on systematically comparing live results with previous vote tallies. It estimates how much of the overall vote total remains to be counted, and the partisan or candidate split of those votes. This helps our reporters and analysts understand what to expect for the rest of the night. It also helps us discern key voting patterns by comparing final and expected results in counties and precincts with those in other, similar geographic areas in a certain state or across the country.

The model is built using a variety of data sources. The Post uses data from the American Community Survey, run by the U.S. Census Bureau for demographic data on counties. When possible, the model relies on precinct-level data; demographic information for those areas is provided by L2 Political. Demographic factors are included in how we examine electoral contests — including age, race, gender, education level and median household income of the voting-age population for counties and of registered voters for precincts.

The model first turns “on” during a given race when at least 20 counties or seven precincts have counted all of their votes. It adjusts in real time as live results are delivered by the Associated Press or released directly by the states. This process allows The Post to estimate — within a range — where the outstanding vote might eventually land. The resulting estimate is updated in a mathematically rigorous way as more results come in. (Here’s a more technical explanation of the model.)

The Post does not call winners and losers based solely on our elections model. A team of editors and data experts assesses results from the AP, state boards of election and our expected votes model to determine when The Post will report projected race outcomes.