How To Tell Who's On Track To Win Pennsylvania On Election Night

How this works: 538’s county-level benchmarks for the 2024 elections estimate what percentage of the vote Democrats or Republicans would need in each county for each statewide (or district-wide, in the cases of Maine and Nebraska) contest to be roughly tied. That means if a candidate is performing consistently better than their party’s benchmark as we get a meaningfully large number of votes tallied, it’s a decent sign that they are on track to come out ahead in the final results (though we also have to be mindful of returns coming in nonuniformly). The benchmarks for each individual county bring together two main pieces of information: (1) the estimated share of the vote for each party in that county and (2) the proportion of a state’s (or district’s) total vote contributed by voters from that county.

We calculated the first of these two metrics based on how much redder or bluer the county was relative to the statewide result in a comparable recent election. For example, in the 2020 presidential race, Wayne County, Michigan (home to Detroit), voted for Biden by a margin of 38.6 percentage points, 69.3 percent to 30.7 percent, while the state as a whole voted for Biden by a margin of only 2.8 points (in both cases, this includes only the two-party vote share — votes for either Biden or Trump). This put the county 35.8 points to the left of the state as a whole.

We then used a county’s lean relative to the statewide margin as the starting point for the target margin in that county, meaning we adjusted each party’s vote share to match that margin in a two-party race — so in the case of Wayne County, a Democratic victory of 35.8 points gives us an initial target vote share of 67.9 percent for Democrats and 32.1 percent for Republicans. (We later adjusted these numbers slightly to account for the estimated share of the statewide vote that county is likely to contribute and the small share of the vote that third-party candidates will roughly attract to get the final county-level benchmarks — more on that below.)

The historical race we used as a reference point to calculate the party vote share that feeds into the final benchmarks differs by state. For 16 states and the District of Columbia, the 2020 presidential vote is the reference race. This mostly includes the most competitive states, such as Michigan, as well as a few dark blue or red states in which no other statewide race differed notably from the presidential result. But for the 34 remaining states that have been less competitive in recent presidential races but held a more competitive statewide election at some point from 2016 through 2023, we used one of those races as a reference race instead.

Because our benchmarks were calculated assuming linear relationships among variables, using these closer races instead of the more lopsided 2020 presidential race involves a relatively smaller swing in the county-level numbers when we calculate target margins, thus mostly avoiding the risk of creating extreme (or even impossible) benchmarks in the bluest counties in red states or the reddest counties in blue states. (For example, using the 2020 presidential vote in quite-red South Dakota would have meant shifting deep-blue Oglala Lakota County so far to the left that its benchmark for Democrats would go beyond 100 percent of the vote, whereas the state’s 2018 gubernatorial race provides a far more competitive baseline.) Using a closer race also acted as a control for some state-specific factors that could potentially lead some counties to shift more than others in a more competitive race, such as the county’s demographics, partisan lean and share of persuadable voters.

Of course, lower-turnout races tend to have different county-level turnout patterns than presidential races, which brings us to our second piece of information: what proportion of a state’s (or district’s) total vote we expect will come from each county in a presidential election. In every case, we used the 2020 presidential contest as our baseline for relative turnout rates, which required a small adjustment in the 34 states where that contest was not the reference race for two-party vote share.

Specifically, we recalculated the statewide vote share in these states (used as the comparison point to calculate each county’s lean) as if each county in the reference race had contributed the same proportion of the vote as it did in the 2020 presidential contest. After this adjustment, for instance, the statewide two-party vote share for the 2017 Alabama Senate contest shifted a tad bit to the right because many Republican-leaning areas had slightly lower turnout in that special election than in the 2020 presidential election.

The last adjustment we made to produce our final county benchmarks accounts for the likely third-party vote share in a state and a given race (which is why a county’s benchmarks may vary somewhat between, say, the presidential and Senate contests in that state). While this is not an exact science, our estimates of how much of the state- or district-wide vote will go to third-party and independent candidates account for results in comparable recent statewide elections, current polling and the historical tendency for levels of third-party support to shrink in the polls as we get closer to Election Day.

Based on these estimates, we adjusted the total combined vote share for the major-party candidates such that the two-party vote share is the same as the benchmark, but comes out of a slightly smaller denominator. For instance, we anticipate roughly 2 percent of Michigan’s presidential vote to go to third parties and independents, so we multiply the initial two-party target vote shares for each county in that state by 98 percent — in our example of Wayne County, that produces a final county benchmark of 66.5 percent for Democrats and 31.5 percent for Republicans.