Data sources

Where the maps come from

Every map in the game is drawn from public data. Parties and candidates are made up; the places are real.

What's real, what's estimated, what's made up

The game mixes real data with simplifications, estimates and parts that are made up. Here is which is which.

Real

  • Every state and county, more than 14,000 cities, towns and communities, and their borders.
  • The local offices and their titles.
  • State House and State Senate districts, the congressional districts Maine and Nebraska award electoral votes by, and (in 47 cities) city council districts.
  • Each state's rules for governor, Senate and president (runoffs, ranked-choice counts, split electoral votes) and the years those races are held.
  • How the 50 largest cities elect their mayor and council, where the research is confirmed (98 of 100 offices so far).
  • The 538 electoral votes.
  • How each county voted for president from 2016 to 2024 (Alaska's boroughs are estimated, because Alaska reports its votes by legislative district).
  • How each neighborhood voted for president in 2020, from official precinct results moved onto census block groups, in every state but Alaska and Hawaii.

Simplified

In the 50 largest cities, a confirmed council or mayor race follows the city's own rule (runoff, ranked choice, a top-two primary, or most votes wins), calendar and terms, listed with their sources on Local government. Even there:

  • the rest of the field is played as one independent candidate;
  • a race decided by a top-two primary starts after the primary;
  • a second round more than six weeks away is played as six weeks;
  • a district that elects several members is played for one seat;
  • your party's name stays on screen where the real ballot shows none.

Every other city and county race follows one set of rules: most votes wins, races in even-numbered years, and terms of two years for the first seat and four for the second (a seat in the legislature keeps that chamber's real term), though many real local elections are held in odd-numbered years. Candidates run with a party, except in city council and mayor races in the 29 states whose cities usually print no party label on the ballot, by state law or common practice (most US cities hold nonpartisan elections, says the National League of Cities).

Estimated

  • How each town and neighborhood leaned in 2016 and 2024: its real 2020 result, moved by its county's real swing (precinct results for those years aren't available on open terms).
  • In Alaska and Hawaii, how each town and neighborhood leans at all: each county's or borough's real result spread over its neighborhoods using census shares of education, age, income, density, and race and ethnicity. Race and ethnicity are used only for this estimate, never shown in the game and never part of the voter groups.
  • The mix of voter groups in each place, from education, density, age and income.
  • Where each voter group and coalition stands on the platform's six issues, from Pew Research Center surveys (2018 to 2026) by age, education and community type, centered so the two coalitions mirror each other. The platform's names (Liberal, Conservative, Libertarian, Populist) follow the four camps in Drutman, Voter Study Group, 2017.
  • How far ahead your opponent starts is set by the difficulty you pick, not by a real poll.

Made up

The parties, candidates, opponents and staff; the events, headlines and decisions; and the campaign itself. How name recognition spreads, Political Capital and upgrades are game rules, not a model of real campaigns.

Where the game follows research

A few effects follow what political scientists have measured. Each topic below says what the studies found and what the game does with it; where the game picks its own number, it says so.

Each topic has a page of its own with the studies in full, every claim checked by a second reader: do candidate visits change votes, do party conventions and running mates move the polls, do debates change votes, does knocking on doors win votes, or just bring out voters, do endorsements change votes, how much does being the incumbent help, does a landslide give the winner a mandate, why does the president's party lose seats in midterm elections, do attack ads work, and should a candidate answer them, how accurate are election polls and do yard signs win votes.

Conventions, running mates, runoffs and midterms

A national convention's bounce has shrunk: the conventions of 2012 and 2016 moved the nominee's polling average between a point down and three points up, against five points on average since 1968 (Skelley, FiveThirtyEight, 2020). For 1964 to 1988, the direct effect wore off over about two months (Campbell, Cherry and Wink, 1992), and one estimate for 1948 to 2000 is that about a third of a bounce is temporary (Campbell, 2001). The game's bounce is about 2 points, the 2004 to 2016 average, and about half of it is gone in four weeks, as in the 1964 to 1988 study.

A running mate rarely changes the result: one team of researchers finds no general home-state advantage for running mates, except for long-serving officials from less populous states (Devine and Kopko, 2013), while another study finds a larger home-state boost (Heersink and Peterson, 2016); the game gives a popular governor 2 points at home, between the two.

Fewer people vote in a Georgia runoff, and the president's party usually loses ground in a midterm.

Campaign stops

A campaign stop can lift support for the candidate and for the opponent too: in a randomized test of a Texas governor's schedule both rose, and the effects lasted at least a week (Shaw and Gimpel, 2012); a survey late in the 2012 presidential race found visits had only a small, fleeting effect on voting plans (Wood, 2016); and local news coverage rose around presidential rallies from 2008 to 2016 (Snyder and Yousaf, 2020, a working paper).

In the game, a stop brings a little support for a week and a little for the opponent.

The final week

Turning out your supporters is the push that works best: a door-knock that reaches someone raises the chance that person votes by about two and a half percentage points, across US experiments (Bhatti, Dahlgaard, Hansen and Hansen, 2019), while late persuasion moves almost no one in a general election, for offices at every level of government (Kalla and Broockman, 2018).

Persuasion shows up more in ballot measure votes and in some primaries, which those authors put down to the missing party label; the game applies that reasoning to city races with no party on the ballot, where persuading moves more voters. That extension is the game's: the study did not test city races.

Going negative, and answering attacks

A review of the research finds no good evidence that going negative is an effective way to win votes, though attacks tend to be more memorable (Lau, Sigelman and Rovner, 2007), and in US Senate races from 1988 to 1998 going negative tended to work better for challengers than for incumbents, depending on what the opponent did (Lau and Pomper, 2002).

So in the game it works better for a challenger (a Long Shot race is against a strong incumbent) and less well for an incumbent. How much it helps is the game's own choice: enough that, for a challenger, going negative and turning out the base are about an even choice that depends on the race, which the studies don't show.

In two US experiments, several kinds of answers to an attack, both positive and negative, reduced its effect on voters, though some worked better than others (Craig, Rippere and Grayson, 2014).

Debates

Debates rarely change many votes (Le Pennec and Pons, 2023), but studies of US presidential debates find that viewers learn about the issues (Benoit, Hansen and Verser, 2003). In a study of 56 TV debates, no effect on how people planned to vote could be found one to three days after a debate (Le Pennec and Pons, 2023), so in the game a debate winner's small lift is gone within days; that a debate also makes both of you better known is the game's own rule.

How a debate answer plays is a game model built on research. In ratings given during a 2002 German TV debate, widely shared statements and candidates' praise of themselves won approval from every camp, while attacks, facts and detailed plans split the two candidates' supporters (Reinemann and Maurer, 2005); in lab experiments with a staged talk show, rude exchanges lowered viewers' trust (Mutz and Reeves, 2005). The game follows those directions, and an attack goes better for a challenger than for an incumbent, as going negative did in the Senate races above; the points it gives each answer, and the coin flip of how a night goes, are its own.

Real races hold fewer debates than they used to: as of late September 2026, most governor and Senate races had no debate scheduled between the major-party candidates (ABC News). The game holds one in every race for mayor, governor and senator, and two for president, as recent presidential elections have.

Landslides, part research and part game rule

In 1980 US House races, the incumbent's previous vote margin helped predict whether an experienced, well-funded challenger ran (Bond, Covington and Fleisher, 1985), though after a close race, winning scares off few experienced challengers (Hall and Snyder, 2015).

In US House races from 1946 to 1990, having an experienced candidate when the other side did not was worth from under 1 to almost 8 points of the vote, depending on the year (Cox and Katz, 1996; the game uses about 3, near the average of those yearly estimates), and strong candidates win more votes than others in similar races (Jacobson, 1989).

The weaker challenger in a re-election after a landslide, worth a few tenths of a point, is the game's own rule built on the first of those findings. The extra in-office card after a landslide that beat the final Election Day poll by 5 points or more is the game's own reward too.

In US elections from 1960 to 1998, a bigger presidential margin on its own did not move more members of Congress toward the winner, while signs that voters had changed, like a party gaining many seats, taking over a chamber or winning the White House from the other party, moved more of them (Grossback, Peterson and Stimson, 2005), and one study argues the shift fades as the election recedes (Peterson, Grossback, Stimson and Gangl, 2003). Tying the extra card to beating the final poll is the game's own reading of an unexpected result. One classic essay calls the presidential mandate a myth (Dahl, 1990).

Re-elections, part research and part game rule

An officeholder running again starts better known across the district (a game rule) and with an edge in the vote, which has shrunk: in US House races it fell from 8.6 points in 2000 to 1.2 in 2024 (Jacobson, 2025; Jacobson, 2015), with Senate and governor races nationalizing too (Sievert and McKee, 2019).

The game's own values are about 1 point for president, 2 for US Senate and 1 for governor, set well below the older estimates for statewide offices of about 10 points in the 1980s and 1990s (Ansolabehere and Snyder, 2012) because the House edge has since fallen sharply; no study we read gives a recent figure for these offices.

For mayors and city councils the game uses 2.4 points, a choice at the very low end of the evidence. Vote-share estimates for city offices are few: about 10 points for council members in four Southwestern cities from 1915 to 1985 (Trounstine, 2011); about 8 points for the party holding the mayor's office in partisan races from 1950 to 2014 (de Benedictis-Kessner, 2018, appendix); and 2.4 points, not statistically significant, for mayors in five cities from 1900 to 1985 in an unpublished paper (Trounstine, Modern Machines working paper).

Term limits

Every state's limit on its governor, from its constitution (Wyoming's from a statute its courts have not tested), as of November 2026, including Louisiana's November 2026 amendment (each state's limit and its source); the limits on state legislators in the 16 states that have them, where a career's local seat is a legislative one (NCSL); and the 22nd Amendment's two terms for president. Members of Congress have none (U.S. Term Limits, Inc. v. Thornton, 1995). Where no real limit applies, the game's own rule allows three terms in a row in one seat, and a re-election pays half a race's points.

World events, part research and part game rule

Some events move voters toward one end of an issue, sized to what studies found, and the candidate standing nearer that end gains, whichever side they are on. Peaceful civil rights protests in the 1960s moved nearby counties' presidential vote about 1 to 2 points toward the party seen as backing the protesters (Wasow, 2020); the game reads a peaceful protest after a police shooting as a move toward police reform. The same study found that violent unrest moved nearby white voters the other way, 1.5 to 7.9 points in 1968, so the game's riot moves voters toward law and order by the same amount as a peaceful protest moves them toward reform. A school shooting's effect on candidate votes is small once local trends are counted, under a point (Hassell and Holbein, 2024), so the game's is small too. Wildfires moved environmental votes only near the fire (Hazlett and Mildenberger, 2020). After the 2022 ruling that ended a national right to abortion, the issue moved a small share of voters (Mutz and Mansfield, 2024). Local crime news makes viewers more supportive of tough policies (Gilliam and Iyengar, 2000), and high gas prices raise support for more drilling in polls (Pew Research Center, 2008); for both, the size of the vote effect is the game's estimate. Rain on Election Day cuts turnout by about a point per inch (Gomez, Hansford and Krause, 2007), less in close races (Fraga and Hersh, 2010).

Storms, outages, layoffs, a hospital closing, a strike and an immigration raid have no measured, one-way effect on votes, so in the game they move nothing by themselves: you answer them with a card, and what each answer does is a game rule. A measles outbreak and a local team's championship are news only; the claim that sports wins help incumbents has not held up well (Fowler and Montagnes, 2015). An opponent's gaffe costing them a little where it happened is a game rule.

New candidates, opponents and years

Veterans gain little vote share on average (Schroeder, Best and Teigen, 2023) but are trusted more on security (Teigen, 2013), so the Veteran wins a little favor with the voters who weigh security most, a game rule sized small. Self-funded candidates usually lose (Steen, 2006), and in one study of House races their own money helped them less than money raised from donors: the Big spender opponent has plenty of money, spent at a premium. Door-to-door canvassing turns out voters far better than it persuades them (Kalla and Broockman, 2018): the Grassroots insurgent has a big turnout drive and little else. A scandal costs a candidate about 5 points when it breaks (Basinger, 2013), and about half of that lingers a cycle later (Praino, Stockemer and Moscardelli, 2013): the Damaged favorite starts about 1.5 points behind where their money would put them. New district lines cut an incumbent's edge about in half among new voters (Ansolabehere, Snyder and Stewart, 2000); since only part of a redrawn district is new, the game cuts a re-election's edge by a fifth. Low trust in government helps third candidates (Hetherington, 1999), which the game raises; how much, and the donors who hold back, are game rules. No study finds a vote edge for policy experts or business owners, so the Policy wonk and Small-business owner are game rules, kept small, and so is a crisis abroad, which the game models only as crowded news.

Sounds and music

Every sound in the game is in the public domain (CC0), so no credit is required. We give it anyway. Sound and music can be switched off from the speaker button in the game, and the choice is saved on your device.

Elections and people

Boundaries

County board districts

City council districts